Saturday, December 29, 2012

Image Parsing

Thursday's post about the painting machine called Vangobot brought on a lively discussion about the similarities and differences between human painters and programmed machines.

As many of you observed in the comments, Vangobot executes physical paintings, but the results are only as good as the instructions it receives. As a result, it's easy to dismiss Vangobot as a kind of fancy inkjet printer.

Machines like Vangobot may develop the hand skills to manipulate the brushes and paints, but will they ever have artistic judgment? Is it possible for a computer to be programmed to see and interpret the world in the same way that an experienced painter does?

These are qualities of the "eye" or "mind" or even the "soul" more than the "hand." Blog reader M.P. invoked a quote from nineteenth century drawing instructor James Duffield Harding who characterized the great artist as having an instinct for "selection, arrangement, sentiment, and beauty" rather than just replicating reality.

Let's have a look at a photograph of two people in a parklike setting. How would an experienced painter transform this image?

Note the difference in this master painter's interpretation. The details in the faces are accurately drawn, but rougher brushes are used for the foliage. The drapery is painted efficiently with big slashing strokes. The sky is painted loosely with spots of light. The sidewalk tiles are in perspective, but they're just suggested with thin dashing strokes. The colors are warmed up and intensified. 

For a computer to do this, it would have to understand what it was looking at, and have an instinct to do all these subjective interpretations. It would need to be able to handle a lot of brushes in a variety of ways depending on what forms it was painting.

Let's look another example, a photo of a landscape scene composed of sky, trees, water, and rock.

The experienced painter uses a variety of paint handling depending on material. The water uses long horizontal strokes, the rocks are done with flat brushes, and the ducks are painted carefully with small brushes.  

Note how different this is than an off-the-shelf "artistic paint daub filter" from Photoshop, which merely translates everything into blobby strokes indiscriminately across the whole image, without regard to the areas that are psychologically important, such as the faces. 

Now what if I told you that the "master painter" of all of the examples is a computer? 

The creators of the program are Kun Zeng, Mingtian Zhao, Caiming Xiong, Song-Chun Zhu from Lotus Hill Institute and University of California, Los Angeles. The goal for the software was to interpret photographs in painterly terms.
The process begins with "image parsing," where the scene is divided and grouped into various areas of unequal importance and of unequal character, such as foliage, branches, drapery, and faces. Each region of the painting has different meaning to a viewer and therefore requires a different paint handling. This visual meaning is known in the field of artificial intelligence as "image semantics."

The image parsing software works like the facial recognition system in a modern digital camera, but this system does it at a much more sophisticated level, recognizing and classifying various elements in categories such as:

face/skin, hair, cloth, sky/cloud, water surface, spindrift, mountain, road/building
rock, earth, wood/plastic metal, flower/fruit, grass, leaf, trunk/twig, background, and other.

Image parsing is similar to what a human artist does. Painter Armand Cabrera wrote about this recently in his post "Learning to See." 

The authors of the computer program assigned the computer to use a hierarchy of as many as 700 different brushes for each of these forms, with various settings for opacity (depending on whether it's painting a cloud or a rock), stroke direction, dryness and wetness, and, of course color. 

The strokes are applied differently depending on the forms, and they're overlapped spatially, painting from the background to the foreground so that the objects in front "occlude" or paint across the ones behind.

The colors in the painterly images are shifted according to a statistical observation that the typical hue and chroma distribution of colors in photos (left) differ from those of paintings (right). Paintings have less blue and green, and more yellow and red. 

What does this mean for traditional painters? Should we welcome it or be worried? If this software is hooked up to a Vangobot, anyone could buy a really nice wedding portrait painted in oil from a decent wedding photo. A portable Vangobot with this software could start winning plein air competitions, just as computers have won chess matches.

Is there some skill set that is out of reach of programmed machines? As Anonymous mused in the comments of the last post, "maybe it's along the lines of caricature, and the intensification of forms and space and color and emotions and beauty and mystery?" Will computers ever achieve the higher level judgments, what Harding referred to as "selection, arrangement, sentiment, and beauty?" 

I would be inclined to say yes, yes, yes, and yes. Computers will learn to caricature and they'll do a good job at it. They will learn to paint science fiction and fantasy, and to do Van Gogh or Picasso transformations. Anything that can be deconstructed can be programmed. The more these computers advance, the better we understand what we do as painters. As blog reader Todd said so well: "Robots will only be able to represent as much of humanity as we understand about ourselves."

Despite it all, I do believe that there is something elusive, some element of real genius in great artists that will always stay beyond the reach of materialistic or deconstructive analysis. The greatness of Mozart and Rembrandt and Shakespeare can never be matched by a computer. And for more earthbound practitioners like me, I can take comfort in the faith that other humans will always enjoy works that are filtered through the human consciousness and the human hand, just as we prize home cooking, hand knitting, wooden boats, and folk music.

I believe we should congratulate Kun Zeng, Mingtian Zhao, Caiming Xiong, and Song-Chun Zhu and applaud their accomplishment. These painting systems are not faceless robots, but the creations of amazingly bright people. The one thing that is certain is that these new systems will put certain kinds of artists out of business, they will redefine what we hand-skilled artists do, and the tools will bring to the table new creative opportunities that we can't even imagine yet.
I encourage you to read the full paper: "Parse to Paint"
Armand Cabrera's post "Learning to See"
Thanks, Jan Pospíšil for linking me to this paper.
Video showing Photoshops "artistic filters,"
Previously: Vangobot


Anonymous said...

I prefer the photographs in all the above examples. Maybe machines will replace realism, but weren't realistic 'artists' just machines anyway?

António Araújo said...

>Anything that can be deconstructed >can be programmed.

Exactly. And even some things that can't be (explicitely) deconstructed.

Much of the resistance comes from us fleshbots fearing for our fragile egos, and some very fierce resistance comes from professional artists fearing for their income.

I am afraid that professional artists fearing for their income are quite right to fear - see the unfortunate effect that 3D modelling had over the jobs of classical 2D animators (and I do mean unfortunate). Those who did not want to convert to digital puppeteers (no offense - puppeteering is cool too) saw their market shrink appreciably.

But here is one thing that can't be taken away: the pleasure and knowledge that you get from doing the drawing/painting/animation *yourself*, using your own brain and eyes. By definition, that can't be done for you by the computer. And that is the reason why drawing (as an activity rather than a job) isn't going away, ever.

Yes, the computer can interpret and paint the landscape for you. And you can look at it, as a consumer or as a tourist. But you can only really see it if you draw it yourself.

Corollary: people who spend too much time being consumers of pictures rather than makers are really missing out. Save on art books, invest in some art lessons. :)

Anonymous said...

Abstract artists have already been replaced by monkeys, elephants and various other animals.

Anonymous said...

"Abstract artists have already been replaced by monkeys, elephants and various other animals."

And their paintings sold for more than their human contemporaries, no doubt?

Well, at least they gained some respect from their animal friends....

D Palumbo said...

The first thing I'd like to say is that these actually are quite impressive, particularly the landscape.

However, as I mentioned in the previous post, this still requires a photograph of exactly what you want the painting to look like. I think both posts have seriously overlooked the difficulty and expense in obtaining the quality professional photography required for the machine to interpret and produce from. People generally assume that commercial use of photography instead of illustration is often done because it's cheaper, but so far as I've seen it is exactly the opposite. I'm looking at this purely from a "this machine is taking away human jobs" point of view. If photography costs as much or more than illustration, the only way this technology threatens illustrators is through use of stock photos and I don't see how that really changes anything.

As far as portraiture is concerned, I could see it having some success, though I'd imagine many people who commission portraits would be turned off by the idea of it being produced by machine. It loses the unique, one-of-a-kind, hand created quality. I see the issue as similar to how digital painting has had little impact in the fine art market: the perceived "originalness" is not sufficient to satisfy most buyers.

What is more interesting to me is how programs like this might one day become part of the workflow or process of digital artists. Not the program as an end result, but just one in a string of steps in how some artists may create.

D Palumbo said...
This comment has been removed by the author.
Theresa Bayer said...

That would take all the fun out of it for me. I love painting, and a computer app can never come close to the joy I feel with brush in hand..

JC said...

What you are looking at here are not paintings, but photographs; they're simply printed with a different computer technique. And it would make no difference if the computer instructions were fed to a sophisticated robot that could actually use brushes and paints: it would still be a machine following programed instructions to reproduce a photograph, though by somewhat different parameters than ordinary computer printers. The machine could not say to itself, "You know what? I don't care for that, because it sorta bums me out. I think I'll do something a little different." It could be programmed, perhaps, to throw in some random variations, but again, that's just part of the program, and not fundamentally different than what you could do with Photoshop.

Just because something has paint on it, like a door, doesn't mean it's a painting.

This goes back to a peculiar failing of the computer culture, which is a fascination with technique. Technique seems to me to be the concern of a beginner painter, not a mature one. One reason I don't care much about 99% of plein aire painting is that most of it is simply an accumulation of techniques that have nothing to do with nature. You go out an paint a barn in a field and learn how to spray little daubs of paint to represent the flowers that aren't really there, and to make a foreground of foliage that isn't there, etc. No great paintings depend on technique, although the artist's technique may be marvelous -- they essentially depend on something else, involving thought, emotion, personal expression and so on.

The key to fine paintings isn't 700 brushes or techniques, it's the brain. A computer doesn't have one.

Anonymous said...

Thanks for sharing this and keeping your blog so informative. I have learned so much from reading it!
I think this technology is great. Speaking from a fine artist point-of-view, buyers who invest in hand painted paintings are not going to be interested in purchasing paintings that are computer digitized. In fact, it opens up the world of art to those that can't afford it. It also helps accelerate artists that want to learn how to paint, shortcuts on how to improve on their work.
On the digital-point-of-view, I do see what Antonio is saying when it comes to 2d/3d art. It is sad to see 2d art being less prevalent, however, anything that ends up on the digital screen will continue to evolve and change.

KB said...

Dan Wexler et al also offer "Glaze" for iPhone which does some level of parse/pre-processing, looking for faces etc. It's been very well received and Wex gave a talk @ the last Siggraph about it.

The sticking point for me on all of these sorts of programs is their dependency on the input photograph as the "ground truth" -- that is, they all start with a photo as the "correct" image and then layer it with various automated effects. It's true that such images can teach us a few things about perception and so forth but Antonio gets it right (imo) when he says "you can only really see it if you draw it yourself." Software tries to replace the specific character of things with the generic (parsing helps some). Which goes against the grain of Klee's famous observation: "Art does not reproduce the visible; rather, it makes visible."

Jan said...

"The key to fine paintings isn't 700 brushes or techniques, it's the brain. A computer doesn't have one." does. Not a squishy human one, but it does have an equivalent of a brain. And it keeps getting better all the time. But of course, we're still special and our humanity shall never be surpassed by no friggin' toaster! Right? ;P

What you're describing is simply a layer of decision making, one which is not being implemented in this program. Not because it's ultimately impossible, just because these guys decided to do one thing - artistic parsing of photographs.

It'll happen eventually.

It's quite sad to keep reading these dismissive or terrified posts over and over. These scientists are not doing it to put painters out of their jobs. They're doing it, because it's an interesting problem to solve. There's something special about painting indeed, that's why it's worth analyzing and simulating.

And until we manage to add that decision making layer, which will be able to generate content, rather than parse it from pre-existing images, you don't have anything to worry about.

Even then, there will be enough snobbish people out there willing to pay extra for the so called "human touch". ;)

Anonymous said...

an abstract painting robot that uses AI and its aural environment as input:

mdmattin said...

The computer generated paintings are indeed impressive, and a tribute to the ingenuity of the programmers in deconstructing a process, in this case the creation of a traditionally realistic painting, and transforming it into an algorithm and a working program. Similar work has been done in other art forms, e.g. programs that compose music in the style of Mozart.
What these programs don't do is come up with an original concept or vision as a basis for new works of art. One could argue that Mozart's accomplishment was not simply writing a bunch of compositions, but inventing the algorithm for creating Mozart compositions. The same could be said for any visual artist with a recognizable style; art forgers attempt to decode and reverse-engineer their works to find the recipe for making fakes.
Not to say that a computer capable of meaningful artistic innovation couldn't come along someday. Would such a machine be more impressive if it created art that reflected on the human condition, using knowledge supplied by us, or if it communicated to us the essence of its own machine existence?

Dan and Deb said...

I'm trying to analyze why, but for some reason as I paged down and saw the photo of the two people, and then the painting below it, my first perception was that it was a machine generated image. I wasn't yet engaged in reading the post, so this was merely the result of looking at it. You convinced me, as I read the explanation of the various techniques used--although I wondered immediately why the 'master painter' would not be named! I'll have to further analyze why I thought it was mechanically made... Perhaps it was the absolute adherence to the location of every element, without any variation in placement, like a photo that's been traced. I'm not sure. --Deb

Christine Cancelli said...

This is all terrifically interesting to read, but it seems Jan is the only person who realizes that no matter how sophisticated these machines become, they can't see and they can't think. The selection process you (the artist) must go through whenever you sit down to create a work of art should include decisions about what you want to "say" - do you want to capture a light effect, a humorous moment, a mood, etc.? You must then select the image(s) (photos, sketches, memory) and technique(s) (media, paint application, etc.) which will help you successfully execute your painting. Reproducing a photograph via computer, no matter how clever the program, can never make up for the intellectual process (conscious or otherwise)that goes into every successful painting. Even the most technically superb painting cannot have any emotional impact without having a reason for its creation.

Anonymous said...

Nevertheless, it is something that is inevitable like photography. Artists and photographers will have to get used to having it around just like illustrators competing with photographers. This will seperate the men from the boys for sure.

Anonymous said...

every artist have their own style. i cant say the same for computer but perhaps in the future they can mix existing styles to always create a new one.

Brian said...

Hmmmm? I suppose most everything is inevitable and eventually a bi-optical apparatus that can make random and organized and simplified decisions will be set-up on the field or in front of a still-life or model and paint with real paint and the result will look like it came from the hand of an artist. What separates us from any machine? Is it emotion and feeling? Love, hate, sorrow, happiness. Those are the things that create the impulse to create.

Anonymous said...


In transforming the photographs, an artist might do much more than was shown by the computer program, such as alter the composition dramatically to suit himself, varying his interpretation in any one of a bazillion possible combinations. One might suppose a computer using brute force number crunching would generate more possibilities than a human would. However, the chances are slim to none that a computer would generate the same interpretation and even less so, a new creation, at a given time as a given artist.

The Michelangelo/ Mozart/ Shakespearean greatness factor is an interesting conundrum.

Certainly, a computer can recreate and alter a variety of images/sounds/words based upon input. Would a computer create the same image or write the same book any particular person has? Considering the probabilities and multitude variables in specific combination required, it would seem to be just about as unlikely that a computer program would generate any particular artist's interpretation or creation at a given time, as it is that it would create a Michelangelo, Shakespeare, Beethoven, etc.

Can creative thought/judgement/wit/humor/je ne sais quoi be deconstructed and programmed into a computer to be regurgitated to wild acclaim? Some approximation based upon standards and randomness is possible. Perhaps the greatness of a Shakespeare, Michelangelo, Beethoven, Mozart, etc., could theoretically be achieved by computer, though I would suspect it would take a long time to hit upon the result much less also have the program distinguish the great from the not great.

- mp

Anonymous said...

I know a lot has already been said about this, but I think we're missing a special component of art... something machines can't do. I mean, for all of us, it's a lot of fun to just sit alone and paint pictures. I love the smell of the oils and everything, I love being careful with my brushes and concentrating really hard on a specific portion of a canvas... I think everyone who contributes to this blog has those same feelings. Maybe machines will replace a lot of professional painters, but I can't imagine a machine would enjoy the process. A machine won't be proud of its work or anything like that, and a machine won't have favorite color combinations or anything... I think that's the real distinction.

jane gardiner said...

The thing that separates us from the computer presently is experience - the accumulated data of 15 yrs plus. Everything we have been through affects everything we do - and no computer can replicate that yet.

However, a counter argument is that current visual artists have access to imagery that was unachievable previously.How different would Rembrandt have been if he had been able to see Caravaggio's work, for example?

What computers can't do presently is make that big leap to the place nothing has been before. They can copy Mozart, but not become Mozart.

But for most of s working artists they may well end up as competition. Like the market for painted portraits shrank with photography.

(as an aside, the way I prove I'm not a robot may not be that difficult to write a computer programme to replicate)

Don Ketchek said...

It looks like computers can be programmed to make artistic judgements, but it is still the human programmers who are making those judgements, at least for now. So, in a real sense, the programmers are still the artist. What the computer does is make mass production of artistic decisions possible even by those who would have no idea how to make those decisions themselves. And in today's technology-based world, what could be better? You can know little or nothing and still create the same things as those who have spent the time and energy to become learned at their craft.
Just choose a few commands in your art program and you create a work of art. Wow, you will feel so proud of your accomplishment in a world that is becoming increasingly accepting of mediocrity, mass production and finding the easiest way to complete your goals.

Pity those fools who actually spend time learning a skill or craft - that's the way of the future.

Torbjörn Källström said...

Ironically I think those paintings would have been more interesting if a human had actually made them like they are...

Anonymous said...

"Pity those fools who actually spend time learning a skill or craft"

The same is true for music. Most of what you hear these days - from orchestrated movie scores, commercials, the latest pop, hiphop, you-name-it iPod hit, etc, etc, - is nothing but computer generated sound.

Still, there's nothing like hearing Yoyoma playing his cello live.


Brandon Miltgen said...

I'm reminded again of the phrase, "Video killed the radio star." But it really should be changed to, "Video didn't kill the radio star," since it actually didn't. We still love listening to music on the radio and watching videos in our musical culture. To me as an artist, the same rings true anytime something "amazing" and "revolutionary" comes along in the art world. Wacoms and Adobe products haven't killed the traditional execution of art, only enhanced the possibilities of what can be done (or allowed old things to be done in new ways). Image parsing programs may open new avenues and may stem off some of the artwork that might have been commissioned to contractors, but there seems to always be a creative place for anything that emerges on this earth. I agree with the post's ending sentiment, alluding to the new possibilities opened by the new technology.

To me, advances in computer technology don't so much threaten current artists. These advances call forth and shuttle along the next generations of creativity.

António Araújo said...

I have to disagree with Torbjorn...if those pictures, such as they are, had been painted by a human they would be rather commonplace landscapes. Knowing that they were painted by a computer makes them absolutely fascinating to me. Not least because of the human craft that it represents: that of the researchers that made the algorithms.

If one day - as I more or less expect it to happen - one can honestly say that the computer did it itself, i.e., if one day we end up having some type of actual A.I., then I will find whatever clumsy pictures it comes up with even more fascinating than the best of human creations! And if the robot painter comes with wheels or feet I'll be glad to go painting plein air with it :)

As for jobs, I agree with the commenter who said that for a long while these machines will probably be accessories, and later there will always be a market for the handmade (our fleshbot chauvinism will guarantee it :)) just because it is handmade. But I would hope that by the time AI is solved we shall have progressed beyond the worries of wage labour? (wait, now *that* is science fiction! :))

As for the contention that this must start from a photo...well, that is not necessarily so. But one problem at a time.

Finally: The whole idea that a computer is somehow handicapped because it starts from "rules"...well, I am not convinced that we aren't similarly handicapped, except that our brains are orders of magnitude more complex, and built in such devious ways that their rules are hard to understand (for ourselves! meaning, our brains are not strong enough to easily self-deconstruct). The whole question is whether we can build a machine with silica that is just as complex as a human brain. It's a technical matter, not a matter of principle. I may of course be wrong about this, but I've never seen an argument that convinced me I am not a rule-based machine. I am fairly predictable, for one, as are most of my fellow fleshbots (especially when they predictably argue about how unique and unpredictable each single fleshbot is). ;)

stevec said...

Very interesting. I immediately was able to tell the computer generated images were done by computers -- the proportions of everything were sort of "too perfect", the ducks were duplicated too exactly, some photographic artifacts (blasted out overexposed sky) were replicated in the paintings, and so on. That being said, they were very very impressive. Esp. the landscape. If they had not been posted next to the corresponding photos, but by themselves, I suspect I'd have thought they were done by artists who were pretty good at their craft.

I like this kind of post.

Anonymous said...

Then again, maybe artists who paint will go the way of the horse and carriage. - mp

etc, etc said...

This AI "master painter" is far better than the previous one, and reminds me of the work of certain revered contemporary realists who emphasize working from life and Sargent-esque paint handling (will it be given a seat on the next Richard Schmid Panel Discussion?).

I think it's important to recognize that at this point this technology concerns itself only with matters of rendering and execution of a pre-existing photographic image. It does not address the far more sophisticated and complex conceptual and theoretical issues of design and aesthetics, which lead to that very slippery question of "what is art?" that has befuddled humankind since the early 20th century; it will be interesting to see if that question mires down artificial intelligence (and its human progenitors) the way it has the non-artificial kind.

HKP said...
This comment has been removed by the author.
Industrial Serendipity said...

Don Ketchek came closest to the point that I am going to make.

When you play chess against a computer, you're playing merely an extension of the programmers who programmed it. Every decision that the computerized chess program makes, it was programmed to make, the actual decision was made by programmers a long before the game began.

Let's say you have a machine that was programmed to do an immoral act, steal the Hope Diamond, for example. Morally the machine is not and cannot be responsible for that act, the person who made it perform that act is responsible for it.

So yes you can make a computer program that can create art; but as shown here by the fact that THEY are credited, the art was truly made by the programmers.

I suggest that the program itself is a work of art, not a producer of art.

- John V

James Gurney said...

Keep in mind that the input doesn't have to be a still photograph in the conventional sense, nor does the output have to be a painterly image. The output could be a re-rendered photo your downtown with the utility poles and wires removed, or your street turned into a yellow brick road, or the pedestrians on 5th Avenue turned into Warner Brothers characters.

And the input could be a dynamic capture with a moving camera, a moving subject, cinematic mo-cap data, infrared data, x-ray data, Google map data, or a combination of the above. The image parsing and painterly rendering tech outlined in the post is in its infancy and will combine with other technologies.

What's significant to me is that the computer is applying a semantic intelligence to the visual scene; it is learning to see--and the results of this seeing will change the way we humans experience the world as visual beings.

bill said...

In this, and the previous discussion, there are only a couple of people, Jan for one, who seem to be grasping the fact that the argument does not lie in the here and now but the future. If there is the possibility of creating an "artificial intelligence" (poor choice of terms in my opinion) then this "machine" will be able to think and reason, learn and make decisions based on experience. In essence a brain, maybe not human but a brain. We are not allowing for the fact that this could be the natural order of things. We humans in our history of development create a new species so to speak. Then this species will have the ability to think and reason and possibly feel its own kind of emotion.

JNLeDuc said...

Do computers make happy mistakes? Many inventions have come to pass by way of error.

kev ferrara said...

The authors of the computer program assigned the computer to use a hierarchy of as many as 700 different brushes for each of these forms, with various settings for opacity (depending on whether it's painting a cloud or a rock), stroke direction, dryness and wetness, and, of course color.

Art is poetry. The brushes and settings (explained above) are the whole extent of the "art" here. Everything else was mechanical.

C said...

I'm a proud Luddite. Art deserving of the name will never be made by a machine. Scientists interested in solving the problems of sophisticated image processing are welcome to play their little games, and obviously the results will gradually look more and more like paintings to the untrained eye, but the real game is elsewhere.

Andy said...

Other than the textural strokes in the sky of the second "painting", these looked computer generated to me. That might not have been the case if I didn't have the original photo to compare with.

Before I read your exposé, I found myself squinting at the screen to see if anything was out of place in the "paintings". The ducks were too perfectly copied for my liking.

Other than the separation of regions that you mention, there is no artistic licence employed in either painting. Everything is in exactly the same position and proportion as in the photos, right down to the last duck.

In the end, we still have a picture of a street with someone looking out of the frame. Nice enough, but would you have painted it this way, even from a supplied photo?

But those textural strokes in the sky had me thinking.

António Araújo said...

>Art is poetry

....and poetry too shall become automatic...though certainly not yet. :)

Machines beat us at chess and at jeopardy. Soon they will beat us at many other things.

Now it is mostly at games. Why? Because winning is more clearly defined. Perhaps the hardest thing at making a computer that beats us at art will be to prevent the human artists from sulking into a corner, changing the rules, throwing a tantrum, or claiming they just don't feel it is "real" art. Not worth the trouble. One could rely on Kasparov to lose gracefully and actually be fascinated rather than offended by his mechanical opponent and the "little games" that hardworking, creative scientists play.

By the way, I fully expect that one day a machine will be better than humans at proving theorems and generally doing science and math. I am not threatened by that. I can still do it for my own fun. Like I can draw a picture for my own fun. And other fleshbots can still praise my pictures if I do well: "not bad....for a human"

It it really so bad to imagine the possibility that we may create something that beats us at everything? Won't it be like having a child and wishing him to prosper?

It may never come to pass. It may be impossible, for some unknown reason. Or it may just be impractical. That would be a shame.

Fleshbots are too competitive. We are more baboon than bonobo. Frankly, lots of what is taken to be so creative about humans is pretty predictable. There's a constant desperate race to be the first to implement a more or less obvious idea whose time has come in art or maths or science or poetry - if you don't do it then another 20 guys will do it the next week, so rush to put your name on the theorem, art gimmick, patent. It is tiresome. There is a way of enjoying the making of art and science that has nothing to do with being competitive and trying to prove what a unique special snowflake you are. Maybe when computers finally beat us at the creative rat race we'll finally learn to relax in defeat, and enjoy the process of creating in a purer, less desperate way.

Yeah, I know - dream on! :p

dzart said...

The comparisons to chess and other games are false, as they work on a set of limited and strict rules which are far easier for a computer to interpret than a human; there are no such boundaries in art.

I think the perspective of this whole argument as been deceptive because we have automatically attributed the creation of these artworks to a machine/software, but I think the real artist are still human.
Instead of brush and paint, they've used metal and maths.

Machines, no matter how complex, are still just tools to be operated by humans.

If/when the day comes that a machine will be addressed not as 'it' but as 'he', then it will have the potential to truly create.

kev ferrara said...


A camera is indifferent to everything, thinks nothing, feels nothing, desires nothing. When a machine can enjoy a sunny day, then we'll talk.

Have a good new years all!

bill said...

And that is my assertion Kev, that one day we will be able to build a Data with emotion.

James Gurney said...

I'm enjoying the conversation about the "soul of the machine"— well argued on both sides. The fact that this issue comes up in the context of this technology reminds me how central visual perception is to our conception of consciousness. A traditional camera is a wonderful device, but it doesn't "see." The fact that these machines begin to exhibit the capacity to recognize objects and faces and to "understand" what they are endows them with some of the most fundamental capabilities of the human experience and it "puts them on our radar" (to use a reverse mechanistic metaphor).

António Araújo said...


I think there are 2 different arguments here, at least, so I'll separate them:

1-the way I see it, we are also machines, but we feel, think, desire.

Sure, we don't have a (man-built, non-carbon-based) machine that can do that yet. But it seems reasonable (though not certain) that one can be built, and these are first steps in that direction. Here are a few others, more interesting in my view:

"How to Grow a Mind: Statistics, Structure and Abstraction"

Anyway, people (Hofstadter, for one) have argued that you can't have a really *good* (rather than serviceable) AI poet (or even translator) without a full-blown AI. I tend to agree. But, like him, I also think we can have a proper AI. And I do think that we are very complex machines, but still machines. Hence (assuming that to be true) consciousness and feeling, whatever it is, can be had by machines because we are proof of it. (by machines I mean we are deterministic gizmos that take inputs and churn out outputs by the same laws of physics that govern rocks or computers or whatever "matter" you care to name)

That still doesn't prove that we can build such machines, but it makes it reasonable.

2- a very different argument is this: whatever the merits of the first argument, we can come to a point were machines, that do not feel, do not think, etc, can be programmed in such a way that they do make paintings or texts that to the majority of the buying public satisfy their working definition of (serviceable) art (whether they satisfy your personal one or not). We may even come to a point were the machines pass a Turing art test - meaning that you yourself cannot distinguish, in a blind test, whether that painting was made by a rather commonplace plein air painter or by a machine. This, I think, doesn't require an AI. And, whether it is art or not, if machines start turning out a lot of art-like "product" fast and cheap, a lot of artists will be affected by it.

And you can argue that the painting was painted by the programmer and not the machine, but that doesn't change the fact that people that like being creative with sable brushes would be losing jobs to people who like being creative with algorithms, just like people who used to animate by painting cells now either lost a job or adapted to animating by manipulating digital puppets. So this work, in this very concrete sense, is not irrelevant to artists, I think.

Also, consider that very few of us will ever make a work of art that is all that transcendental. Run-of-the-mill, formulaic, working-man art is what most of us will do (and nothing wrong with it, from christmas cards to novel covers to batman cartoons), and taking those jobs away will really hurt people who live of them and love to do them. Formulaic art gets an undeserved bad rep - as Leonard Cohen says, "I know we are not new". Just because somebody fell in love just like you a million times before and falling in love is formulaic, it still is the best thing in the world, and feels new each time *you* do it.

Apart from the jobs, also consider that if the unthinking machine passes the blind test in drawing the 90 percentile of what goes on at conceptart or pleinairwhatever dot com, it really may say something about how formulaic we humans actually are. And I do think we humans are more formulaic and simpler than we think, so that aspect makes me curious to see how close to our achievements we can get with unfeeling unthinking algorithms - and I want to measure that distance by blind tests on *the results* and not on arguments on how the word Art should be defined, because if we start arguing definitions then we can always argue it in such a way that humans always come out ahead.

By the way, happy new solar-cycle, fellow soon-to-be-obsolete flesh-bots! :)

etc, etc said...

The comparisons to chess and other games are false, as they work on a set of limited and strict rules

The same applies to algorithmic composition of music; a uniform, intervallic 12 tone division of pitch forms the substratum of music, and from there, intervals, scales, chords, progressions, etc., are layer upon layer of structure that progressively limit choices in a systematic, methodical way. I find it ironic that the futurists here are claiming that some people don't get their vision of an AI brave new world, when they seem oblivious to the fact that visual art has no such universal and practical intervallic substratum, are utterly clueless to the difficulties of aesthetic certainty, and the way sci-fi/fantasy genre (Space: 1999, anyone?) obviously informs their opinions.

WWick said...
This comment has been removed by the author.
WWick said...

While we may first be attracted to a painting's imagery, and then by some stylization or optical effect in the brushwork, looking further into an artist's backstory - their intent, feelings, struggles, limitations, and circumstances - can provide a far deeper appreciation of the work. Thus informed, and becoming intimately familiar with an artist's style, one can also come to recognize the work as a performance persevered in paint, with his or her very being (call it soul, DNA, spirit - whatever you want) inextricably locked up in those preserved motions.

Learning that an algorithm was responsible for imitating a certain style of brushwork selected from a pull menu is not much of a backstory and pretty much short circuits any admiration one might have for human performance. As a technological innovation, it's mildly interesting, and may have it's uses, but merely manufacturing a pictorial effect is not as interesting as coming across a rare talent.

António Araújo said...

etc, etc,

if we want to reduce music to a limited game just because of the wave nature of sound, then please consider that objectively, images are also decomposable into a combination of waves - that is just a fourier decomposition. So not much difference there.

Also, consider that all the images that you appreciate on the printed page or on a computer screen consist of a finite set of points with a finite number of possible colors. (Yes, the combinations on canvas are larger, but in practice you can't tell the difference after a certain resolution, and illustration in particular is judged by how well it does on rather crude approximations of the originals). So both "games" are well approximated by finite games. Both games, however, have a huge state space. We have been making pictures since the dawn of humanity and we still keep coming up with new ones, all of them distinguishable on the finite set of points of a computer screen.

And no, my position on this certainly does not come from movies - it comes from working knowledge of actual AI, formal systems, and considerations of stuff like logic, bayesian probability, and how Godel's theorem applies to machines and humans. Anyway, good sci-fi movies (of which there are very few) are usually derived from good sci-fi novels and good sci-fi novels usually come from people who actually understand the issues. So don't hate on poor old HAL2000, we'd probably have a better world if kids still dreamt of being astronauts, like we did in my days, rather than pop stars (and we'd probably have feet on mars by now, and many other neat things that are delayed by tight budgets and small imaginations).

António Araújo said...

>clueless to the difficulties of aesthetic certainty

I'm so clueless that I don't even get what you mean by this phrase :)

Do you mean it is hard to determine if a work is aesthetically pleasing or not? If that is your point, I agree. But that is why I said I wasn't much interested in discussing definitions of Art.

It is however very easy to make a blind test and check if a specific sample of humans of a certain society and a certain time finds a certain picture aesthetically pleasing if it is unaware a computer did it. I was careful to say that this won't guarantee that kev or you or me may find the same picture to be worthless. But it means something. Neither you nor kev nor me have a monopoly on taste.

I'm not trying to diss on artists, you know? I draw every day, and it is one of the most pleasing things in my life. And illustrating is even a part-time job for me, so I'm coming from a position of deep admiration for art. I don't see any of this position as being in any way degrading to artists, or humans. I was careful to state that I see mathematics as being just as vulnerable -or more- to being made by machines, and I don't see anybody raising a fuss about that. Why is that? Perhaps you think that doing math is just a silly game that involves no artistry? I assure you that in maths there is also "difficulties of aesthetic certainty": true, one generally knows if a proffered proof is or is not a proof, but one can spend ages discussing whether a proof is beautiful or important. I think mathematics is just about the deepest human art you can imagine, in fact deeper than visual art (although I personally get more of a daily kick from drawing) and yet that too I think will *probably* be one day accessible to a man-made machine.

I will add, however, that an AI will probably have its own aesthetics. Our notion of beautiful is probably not an absolute determined by geometry but an accident of evolution, so if we want a machine - even an intelligent one - to make art that pleases *us* rather than *it*, then we will probably have to make sure the correct aesthetics his wired in. But again, that is the case with us too - just substitute evolution for the programmer.

By the way, when I try to place a comment here, blogger keeps asking me "prove you're not a robot". That is my whole point. At the moment I cannot really prove I am not a (rather well made) robot. Blogger is too easy to fool.

etc, etc said...

if we want to reduce music to a limited game just because of the wave nature of sound, then please consider that objectively, images are also decomposable into a combination of waves - that is just a fourier decomposition. So not much difference there.

You are in WAY over your head.

António Araújo said...

Sort of apropos of issue 2 above: the industrial revolution did not begin with machines but with the division of labour, i.e., with using men as machines.

The other day I was reading an article on those chinese art factories that turn out reproductions of old masters, in which you have have certain workers specialized in emulating a specific old painter or even a specific detail of his paintings, and I was reminded of that fact. Once you get division of labour too thinly figured out, you are begging to be replaced by a machine.

Ok, you'll say that is not real art. Tell that to the people who buy it. I'm not discussing definitions.

António Araújo said...

>You are in WAY over your head.

A taunt is not a very informative way of discussing an issue. We are just starting the year. Good will?

No, I do get your point - that music, of a certain kind at least, is ready made for formal systems. It is a point well know, and the reason why we have lots of seminars on automated generation of fugues and stuff like that, or even of the specific style of Bach if you want, and we don't get such seminars on visual art. That kind of "ready-made" structure makes the state space easier to choose from - and it is also the reason why maths will be easier to automate than certain stuff though it is by no means less deep.

So I do get your point.

Do you get mine? How are you not fundamentally a robot? And if you are, and you can choose from the state space of possible paintings, why can't I make a different type of robot that does the same?

ps:Now go and tell musicians that they, unlike visual artists, are replaceable by robots and see what you get as a reply.

James Gurney said...

Antonio, can you explain what a "state space" is in this context? I'm not familiar with the term.

António Araújo said...

Anyway, I'll leave it at this: people who dream the dream of strong AI keep coming up with new useful and grand things: cars that drive themselves, speech recognition, medical expert systems, you name it, and even toys that emulate certain aspects of art making.

They make my life more interesting.

The other side has been grumbling for decades about what cannot be done, watching the undoable be done, and then retracing new borders that this time can't be crossed (they did use to say that chess was out of bounds, remember?).

So we'll see. I keep saying *probable* and *reasonable* and *arguable*, not certain, perhaps you haven't noticed. Perhaps strong AI can't be done, and perhaps an artificial artists can't be made. Perhaps. But I do know what is the most productive viewpoint and the most interesting bet, and it is not to grumble about how ineffable and wonderfully unique we humans are.

António Araújo said...

James, I just meant the space of all possible objects of a certain kind.

For instance, if you have a screen of 10 pixels by 10 pixels (hence 100 pixels total), and you can only put a black or a white dot in each location, then you can have 2 possible states in each pixel, so 2^100 possible paintings you can do. That would be your state space. You can imagine it as the space of points where each point represents one of the possible pictures. And the question would be to partition that space into regions that correspond to what we would call actual valid paintings and regions that we would just call nonsense paintings.

Now, the question often how to choose the right way to parametrize such a space. Sometimes by clumping the points together in smart ways you can make the problem of navigating the state space (that is, choosing the interesting regions) easier.

António Araújo said...

Rephrasing it: the question is how to find structure in the state space. Sometimes it is obvious and sometimes it is not, but the structure may still be there for you to find.

For instance, I've worked on classifying curves of a certain kind, and these curves are defined by numerical parameters. Each choice of the parameters defines a curve, like each choice of the set of 2^100 pixels defined a painting.

Now, at first all you had was the state space of all these (infinte) curves, and it was a big amorphous jumble where we wanted to match certain curves with each other (meaning, certain matches were good and certain matched weren't, like certain pictures among the 2^100 might be good and others not)

Turns out that if you looked at that state space in a proper way the "good" points were not disordered but actually clumped together into nice geometric, understandable objects, that "lived" in that state space.

I use the term in a rather loose way. It actualy comes from dynamical systems.

kev ferrara said...


I could write for days answering you. Let me just run through some bullet points. This is going to be random, but each point I touch on turns out to be a pandora's box of problems re: Art and AI.

In researching aesthetics, I came across and catalogued 16 different distinct definitions of beauty. The simplest of these, I believe the computer has already pulled off. Mimesis, for instance.

The most compelling of these 16 "causes" of the beautiful I don't think can be quantified. They can only be qualified. This is a central point. Poetry is not quanitification. It is not mimesis. It is instead qualification.

Nor can experience be quantified. Because experience consists of sensations which, despite the evidence we see for them in the brain, have no material existence. The sensation of experience transcends meat, epiphanies being the sensations that transcend the most brain... It is lightning and chemicals moving between meat. (Give this point its due airing in your mind, I think it is rather a profound issue which those in thrall to Scientism seem never to grasp.)

This all relates, quite directly, although it is not yet understood, to the fact that the longstanding project in nominalist/analytic philosophy to do away with all metaphysical elements in language has come to an end in failure. It turned out to be impossible to get rid of some/any kind of notion of abstract concept or truth in linguistic analysis. All that work and propaganda for absolute materialism... all for nothing. Our basic understandings of all things according to abstractions that transcend fact turns out to be more fundamental than fact. (Fascinatingly, any Idealist could have explained this to you 120 years ago.) Which is also a way of saying that without induction, which is by nature an imperfect method of apprehension, no thought or understanding is possible at all. So while a computer can win at jeopardy by sheer number crunching deduction after its scientists essentially quantified the bulk of factual knowledge and shoved it inside it, its ability to deduce this thing called truth from experience is impossible. Truth can't be deduced except in a completely controlled circumstance, unrelated to life. Truth must come as a realization. Which is a sensation of conceptual unity.

Now, in my view poetry and truth cannot do without one another. (There really is no such thing as prose, because we are always speaking in short hand about the world unless we are creating an actual physical demonstration.) And if art is poetry, and machines can't do truth (because, one, they can't actually have sensations, and two, they don't spontaneously create symbolic metaphors for these sensations through a combination of imagination and innate form-translation skills, and three, they can't have epiphanies that prompt synthetic symbolic recapitulations of several seeming disparate sensations into a thrilling isingle mage of truth)...then machines can only do fake visual poetry, not the real deal.

If you make a machine that can do all of the above, then you have made a machine that can feel pain and love... and that is a living organism, not a machine. I don't discount that possibility. But that seems more like genetic engineering than AI.

(sorry for the extreme concision of the above. not going to download a book's worth of info online.)

The rumors of my obsolescence have been greatly exaggerated.


António Araújo said...

(Of course, it would never be a good idea to look at your state space as the 2^100 possible paintings. It is just too messy and unstructured. You'd begin by looking for ways to partition it into interesting classes - the state space of all single figure drawings for instance :)) - and even here I'm just probably being silly :))

But here is another example, maybe clearer: take a stick figure (rigid limbs, rotating hinges on the usual places) and put it on the plane. Now imagine the state space of all possible positions of the figure - each point of that space corresponds to a possible position of the figure. What is that space like? You can imagine it as just the set of all possible pictures. But you can also imagine it as a numerical space with several dimensions - each numerical dimension corresponding to the value of the angle of rotation of a hinge, and a further two dimensions to represent the position of a reference point in the plane, say the center of the torso or something - since any possible position of the figure is defined by first placing that single point and then rotating the hinges in all possible ways that don't break the figure apart.

Now you could ask questions like: what regions of the state space correspond to figures that are stable (won't fall down under gravity). Or even: which regions correspond to stick figure poses that are "beautiful"? Are these regions just a mess of points or do they form beautiful, ordered structures in the state space?

Does this make any sense?

kev ferrara said...

Antonio, As far as I can tell, after years of research, conceptual thought and art are intimately tied together; Art is the most complex language the world has ever known. Music (as it is generally practiced) is child's play by comparison. ... linear in time, particular in notes, 12 tones mathematically pre-determined, steady in meter, etc. Music starts off already quantized. Art is the other side of the world from that. A completely nonlinear, organic, recursively associative, wholly free-ranging form, where about a dozen wholly distinct classes of symbolic abstraction can be superimposed and comingled in meaning. I don't think a single scientist understands even the half of it. And part of that reason is because the language of the scientist is empoverished from the start. THIS IS ANOTHER SOURCE OF SCIENTISM - THINKING THAT ORGANIC SYMBOLS CAN BE QUANTIFIED.

kev ferrara said...

Antonio, all that stuff about Monkeys typing shakespeare is bunk... a failure of epistemlogy.

For each attempt to type the manuscript, or paint the vermeer... the odds of success are zero. And with each new effort the replacement is full. So no matter how many attempts, every one will be a failure..

It is a simple matter. Your scientism has a strong faith component. Or don't you see that? ;)

etc, etc said...

Good will?

You complain about taunting and then proceed to taunt? O.k; I'll waste time for both of us and play along: Are there within state space (or any given visual field) naturally occurring mathematical relationships that create a harmonic or otherwise meaningful effect the way there are in sound waves?

António Araújo said...

I'll answer later, but just two points:

>But that seems more like genetic engineering than AI.

Not genetic, but I do get your point: you'd be creating a thinking, sentient being, whether in silica or carbon it doesn't make much difference. The point is that by creating it you'd be learning something about how we carbon-based organisms work.

Since etc etc brought sci-fi into the question, remember the blade runner movie? You had these AIs that walked around looking exactly like humans. You wondered why they were even called machines. The reason was that people wanted to keep using them as servants. When we make a true AI, we will be definition no longer have a simple machine that we can boss around with moral impunity. That's why I said I'd be happy to go plein air painting with it.

You would wonder why make such efforts to make an AI out of chips when we can make one so easy by just having 15 minutes of fun and waiting 9 months. :) The thing is that we'd be approaching AI "from below", and so we can experiment and mess around and learn from the process with impunity - starting from above (from an already functioning intelligence) you can only learn by picking it apart, and that can't be done decently and morally, of course. Starting from below, you can reverse engineer, and construct instead of decompose, so you'd only get pangs of conscience if you succeed (as all distopian movies remind us) and then you could just release it to live its own life.

And now I'm strictly into sci-fi territory. :)

The point is that you learn a lot even if you fail.

> Which is also a way of saying that without induction, which is by nature an imperfect method of apprehension

Induction is more or less understood. That is what Bayesian probability is about, and it is the big thing now in AI (deduction, of course, is even weaker and also better understood). But I will agree that induction all by itself cannot get anywhere. You need, among other things, to have presets - goals, preferences, utility functions, etc, that are usually expressed by emotions in humans (a strictly emotionless rational human being would never even move - what for? all outcomes are equivalent without inate preferences). Most important, you probably need even more stuff we don't even dream about.

(but these things are taken into account, you know - even the unknown. People in AI are not as crude as you may imagine)

One notion in vogue is that we come with preset modules, that provide us with a subset of basic structures - lots of stuff, from Chomsky's language modules to basic physics modules (i.e., basic imperfect modules of how objects are supposed to behave) to even moral modules, and, I suppose, basic aesthetic ones, that for instance make us look at certain faces with more interest than others. So we'd need to endow the AI with such basic modules - he can't infer them or deduce them from experience. Evolution did it for us.

Of course, this is just talk up until we actually *build* such modules. But the progress already made gives it credibility. Chomsky's stuff seems to be credible (I'm not a linguist, so I don't know) and is certainly unexpected. The Bayesian probability stuff is a huge leap forward in understanding induction (and I teach that, so that one I can vouch for) miles apart from what we had just a few decades ago (and still beyond the head of most classical statisticians that are stubborn as mules and cling to the ugly statistics we were all taught at school). So, given that we do have continuous progress, who knows where to draw the line?

António Araújo said...

>Antonio, all that stuff about >Monkeys typing shakespeare is >bunk... a failure of epistemlogy.

I was careful to say you would never take the stare space to be the full 2^100 possible drawings. No structure. That would be the equivalent of the monkeys typing at random.

The whole point is that the monkeys have already typed. It took millions of years. Meaning, evolution happened. And evolution built modules in our brains that navigate state space in smarter ways than just typing at random.

All that AI is about is trying to figure out what those ways are, and what structure do they impose on complex state spaces that you couldn't possibly navigate through random monkey business.

I find it funny that you guys use the example of the structure of music. It wasn't always obvious that music was so structured! You use the example of a state space that we can now see the mathematical structure of to convince me that those other state spaces that we cannot (yet) see the structure of are hopeless?

Rather, I see it (like chess, like jeopardy) as an example of a state space that we managed to find the structure of and therefore a hope that the other ones can be understood at least as well.

Do you realise how much time it took to make that "obvious" structure of music obvious? It started at least with Pythagoras (or before)! The structure of induction? We are beginning to understand it now, and it started again with the greeks (or before).
We keep going.

António Araújo said...


I don't know what you mean by that. Organic just means carbon based. What are organic symbols? Also, rationally understood doesn't mean necessarily quantified, but I take it you mean it as a figure of speech.

>It is a simple matter. Your >scientism has a strong faith >component. Or don't you see that? ;)

Nope. I don't get that "scientism" thing. What I think is that we should follow the leads that we have. If we do hit a wall and we can't go beyond it I'm fine with that. But if we have interesting stuff in front of us, if it keeps working, then we keep going. And if we dream, with dream about interesting outcomes.

Having "faith" would mean that I knew, I just knew, that it had to work. But I think no such thing. What I think is that I see no fundamental reason for it not to work, and above all I see extremely high probability of finding interesting things even if it doesn't go as far as strong AI. I think it is interesting and fun and productive, so why not go for it? I see no such benefits in investing in unconclusive arguments of why it simply cannot work.

Mathematicians and scientists don't work because we have some master plan to "quantify" everything. We do it because we have fun and we like to learn. Let the limits lie where they may - when we hit them, we hit them. What I find curious is why some people make a point of placing limits here or there precisely because we *don't* know where the limits are.

By the way, we are aware of actual limits - heisenberg, godel, light speed, strong dependence on initial values - and we don't have a fit because of them. We investigate the limits with as much fun and interest as we investigate the possibilities. But why insist there are limits where they haven't been found yet and where progress keeps getting made? Is there a faith component in *that*? ;)

kev ferrara said...

Beyesian probability calculations are still completely dependent on what kind of inputs you put in and their quality, and how the problem is understood by the user, and it doesn't have anything to do with emotion or sensation.

When I say induction, I don't mean in the statistical sense. You aren't getting this. I mean an induction of truths which are predicated on sensations which have been innately understood as emotional symbols in the context of all the other emotional symbols in mind, which have been built up over time into an entire structure of recollectible or sublimated experience. Put simply: This ain't math.

Which leads me to mention another problem of Scientism; a belief that just because any particular phenomena has some degree of analyticity, it is enough to define the phenomena. The result is that non-quantifiables are ignored. This is not good science.

My personal belief is that conceptual symbolism is its own language with its own wholly organic, and complicated rules, which are not, by nature, quantifiable. If you understand how art works, you see immediately that Math is completely inadequate to the task of representing it. And then if you note the similarities between how art works and how thought works, then epistemological humility really sets in.

kev ferrara said...

Organic just means carbon based...

Another form of scientism... thinking that words that come into science from common usage, actually started in scientific usage.

A bit frustrating having this discussion, because you are heavily invested in scientism. You show all the signs. Mainly because you have not really delved into epistemology. And you haven't yet come to terms with the failure of analytic philosophy to do away with metaphysical conceptions.

Read an Edgar Allen Poe short story and think about how it works. Then try to write one.

Aright, gotta go for now.


António Araújo said...

etc, etc, I didn't mean to taunt you. If I sounded like that, I apologize. I always keep it civil in this blog.

If you took the "can you prove you are not a robot" as a taunt, I meant no such thing. I meant it as an actual question: "can you show that humans in general are not machines".

The point is that if we are machines and we can handle the non-obvious state spaces then why can't we build OTHER machines that do it just as well, only built from different components, with chip rather than flesh?

Put another way: do you, or kev, have any reason to believe that there is something in meat that makes it fundamentally superior to silica as a building block of intelligent machines? Or do you just think that humans aren't smart enough to build such machines? Or do you think I am wrong to consider us machines at all? And if so, how are we fundamentally different?

About your question: No, I don't know how to break apart that state space, and nobody does *today*. That doesn't mean it can't be done. Whatever structure it may have, it will be a "natural structure" and seem obvious once/if ever found, just like the structure of music seems so obvious to you in the 21st century but certainly did not seem obvious to the homeric heroes. And yes, I have granted several times, music is simpler in some ways. So what? Some have to be simpler than others, doesn't mean the others won't be found. I keep asking what is so wrong about trying to achieve things, while progress keeps getting made? It is not like trying something stupid like perpetual motion. Why is it stupid? Because it has been tried over and over, it never worked, and no progress was ever made, and further there are strong theoretical reasons to think it can't be done. All of that is absent in AI, and that is why we should keep going and why optimism is more reasonable and productive than pessimism.

António Araújo said...

>Another form of scientism... >thinking that words that come into >science from common usage, actually >started in scientific usage. I just stated I didn't get what you meant and I told you what I understood by the word. You are free to define it as you want, but you have to tell me what you mean.

I tried to put it another way: why is meat different from silica?

We have different languages. When I use a word like "state space" because I am used to it and James asks what it means I try to define it as simply as I can. All I am asking is that you do the same so that I can understand your point. Please understand it is equally frustrating from this side. I see you multiplying terms with each post and I have to try and understand what you mean each time. It all seems like vague words to me, but I do try to understand you, and I do try to find the meat behind the words I don't get.

Also, you keep insisting about the scientism thing. I don't get that. I learned some tools and that of course shapes my way of thinking and speaking. You want to call it an -ism, go ahead, but I am open to using any tool to understand what I can, it doesn't have to be maths. I could spend the rest of my life *not* doing maths, if you must know. I could spend my whole life doodling and I'd be happy. But we are discussing AI and you want me to take maths out of the picture? How do I do that without changing the subject? I simply don't know any other tool that works for AI half as good. And your tools seem to be about proving that we cannot do stuff instead of proving that we can, so even if I got them I could only use them to argue the negative case. But you are doing that already, so what's the point? :)

António Araújo said...

>Beyesian probability calculations >are still completely dependent on >what kind of inputs you put in >and their quality, and how the >problem is understood by the >user,


> and it doesn't have >anything to >do with emotion or >sensation.

Who said it did? I said the same myself, and I said an inductive reasoning being (*in this sense of the word*) wouldn't go anywhere. I said it would need goals, or emotions, or something like that.

If you use the word induction in a way that includes emotions, then we are saying the same thing. I'm using the word induction in the way it is usually used in logic, or even in the dictionary, but I'm ok with whatever definition as long as you explain it.

>When I say induction, I don't >mean in the statistical sense. >You aren't getting this. I mean >an induction of truths which are >predicated on sensations which >have been innately understood as >emotional symbols in the context >of all the other emotional >symbols in mind, which have been >built up over time into an entire >structure of recollectible or >sublimated experience. Put >simply: This ain't math.

Ok then. Defined like that, I agree (I said the same thing above! We need pre-defined emotional structures, or something like that).
So we are just differing on words again. Except that: I'm not sure all that can't be made into math. And don't accuse me of scientism again: I'm not sure it CAN be made into math either. I am just keeping an open mind about it. I ask what do you call the "-ism" that makes *you* so sure it can't be put into math. I am keeping both possibilities open, and going with the constructive one until it proves a dud.

António Araújo said...

TLDR: the way to find the limits is by bumping against them, not by arguing with vague words about where they must lie.

etc, etc said...

No need to apologize; as far as I'm concerned, it's all lighthearted fare. Sorry if that wasn't apparent.

It may well be that some years from now AI artists will pump out original artwork in any periodic style or imitate any artist on demand or even create unique mashup combinations. It's just that I feel any enthusiasm should be tempered with a healthy appreciation for the fact that at this point in time no human intelligence can even begin to attempt to do that, and that it's far more complex than chess or music.

António Araújo said...

> It's just that I feel any >enthusiasm should be tempered with >a healthy appreciation for the fact >that at this point in time no human >intelligence can even begin to >attempt to do that, and that it's >far more complex than chess or >music.

Agreed on all counts. I don't know that it can be done. I just think it is not unreasonable to think it may.

And now I am off to sleep and to dream of electric sheep ;)

kev ferrara said...

Put another way: do you, or kev, have any reason to believe that there is something in meat that makes it fundamentally superior to silica as a building block of intelligent t machines?

Firstly, intelligence is not foundational to human thought. Imagination is the foundation. And the imagination runs an emotional program. And in this program of organic symbolic processing is where the conceptual digestion happens. If you try to create intelligence without creating the processor, you have fenestration without a building.

Secondly, the brain isn't just meat, its also chemicals and electricity. And what makes each of these 3 elements (meat, chemicals, electricity) so necessary is their fluidity of form and promiscuity of interaction with one another.The brain is like the process of evolution housed in a skull , predicated on apprehending reality symbolically and built of meat, chemicals and lightning. (Whoah!)

Furthermore, given the necessity of chemicals and electricity for cognition, it would not surprise me in the least that each has ranges of symbolic qualia. The organic symbolism of the mind is its own unique system of information evaluation which is far more complex than the forms of math. Math is merely one poetic method by which the symbols of the mind can be organized to appreciate the world. Math is formalized thought. Thought is not organicized math. When you appreciate that, you appreciate the degree to which the problem is intractable.

António Araújo said...

> If you try to create intelligence >without creating the processor, you >have fenestration without a >building.

I agree with that. I assume that intelligence will come only at the top of one big pile of cognitive modules. I'm actually more interested in consciousness than intelligence. That is what really makes my (meat-)head spin. :)

>Secondly, the brain isn't just >meat, its also chemicals and >electricity.

When I said "meat" of course I meant the whole biochemical package. So you do think that there is something essential about the building blocks themselves. I wanted to be clear on that, so thank you.

>Math is formalized thought. >Thought is not organicized math.

That may very well be true. Or it may not. I dunno.

> When you appreciate that, you >appreciate the degree to which >the problem is intractable.

I think your contentions are reasonable, I just don't know why you treat them as certain. The fact is we are ignorant about it. We don't know if meat (etc) is essential or not, and we don't know whether or not math is isomorphic to human thought. We just don't know because frankly we don't know enough about either maths, brains, biochemistry, or intelligence. The interesting thing about trying to get AI is precisely that we may learn a lot about all those things even if we fail. It it the process of discovery that matters more than whether we can achieve strong AI. We are sure to achieve *something interesting* on the way.

So, I think it is reasonable to think it may be impossible to get strong AI because "meat" is essential and/or because math is not isomorphic to thought. But, to paraphrase etc etc above, I'd temper your enthusiasm for the powers of philosophy with a healthy appreciation that it has often in the past proved short on predictive power. Words are limited, they have too many holes. Wasn't it Kant who "proved" that the mind could not conceive of a geometry where the axiom of parallels was violated? Yet we soon had non-euclidean geometries. Philosophers of the word juggling type have been setting boundaries for math since Zeno (and probably before that) and those word-made boundaries keep retreating further and further. I'm not saying they'll retreat forever, but only that I would put little trust in any specific word-formed barrier to hold. The language of Math may be limited, but ordinary speech is notoriously so. As I said before, I'll believe we reached the barrier once I bump into it. Is that unreasonable?

kev ferrara said...

Math is formalized thought. Thought is not organicized math

I took me years to come to the above insight. Give it a lot of consideration.

On Kant and words: Kant was a pioneer, but was wrong about a bunch of stuff and I hold no brief for him. (C.S. Peirce is a fellow who much better understood the terrain.) And I am very critical of the ability of words and text based language to convey thought and to properly symbolize the world. Text systems are the worst symbolic systems we've developed... but they have become ubiquitous because they are so easy to build out and use.

Painting is a much subtler and deeper way of communicating thought than text, but it is unbelievably hard and time consuming and it can't hope to match the breadth of scope that text-based language manages so easily.

Chomsky and his ilk proceed from text based language as their source material, which I think is a big mistake because it is such a sloppy, limited symbolism with such awkward and changeable rules. The field of linguistics has a lot of problems which bleed over into the AI field. Part of my issue with scientism is due to the alliance made in analytic philosophy between the math people and the word people to the exclusion of the idealists, the conceptual people. It was these analytical/positivist fellows who used politics to cast out metaphysics as useless and mad, only to have their project fail without it after all the originators went to their graves as heros of the movement. (And now we have 3 generations of people who have been needlessly trained to react against the existence of truth. And this is effecting the clarity of mind of untold numbers of people trained through academia.)

And I don't believe in the innate cognitive module theory. I think its just another way for nominalists to pretend the thought problem does not require a metaphysics built out of meat, chemicals and lightning interacting promiscuously.

António Araújo said...

>Math is formalized thought. Thought is not organicized math

kev, I get what you mean. People have thought about it. "The things you can think in mathematical terms may not be identical with the things you can think". It gets discussed over and over (take Hofstadter's "Godel Escher Bach" for instance, for a popular delivery, or a number of others). The only difference is that you state it as if you knew it to be certain, while for most people it is a hypothesis. Where you get your proof I just don't know.

Stating a thing as being true does not make it true, no matter how elegantly you state it. It is a hypothesis.

>And I don't believe in the innate >cognitive module theory.

It's not for believing. It's a working hypothesis. It is for testing. You are free to come up with a better one and test it too.

> I think its just another way for >nominalists to pretend the >thought problem does not require >a metaphysics's a hypothesis. It is for trying to solve something. If you think you can solve it with metaphysics, then you try that instead. Nobody will stop you...

Really, you seem to believe that people get a lobotomy or something when they take a phd in maths. Sorry, but it isn't like that. We aren't part of a cabal and we aren't hell bent on some grim quest that only validates us if we accomplish some far-off objective of mathematizing the whole world, or something. It is not like that. We have fun working on problems we enjoy (if we're lucky). That's all.
Sure, we may sometimes think of crazy goals like strong AI or "curing cancer", but that is not what we usually discuss. We discuss small problems we enjoy and can actually handle right now, like making a robot that paints landscapes from photos, or navigates a room, or whether a mutation on gene XYZ is related to a specific cancer, or whether some algebraic curves are related to others. We won't die if the whole world isn't "mathematizable". We won't care. There is no master plan.

And we don't all come from some sinister factory, complete with pocket protectors and taped up glasses. We are a varied bunch: my thesis advisor is a bit of a mystic who just loves metaphysics and keeps going on and on about Jung and his yoga classes; I spend most of my time drawing and being a dilettante on 100 things instead of working on my math; we don't spend our time locked up in labs either nor do we have anything against the flesh - we are healthy human beings who have sex lives, climb mountains, and occasionally jump out of airplanes.

We're just dudes, man. :p

We certainly don't pay homage to some devious cult or political cabal and I don't quite get what the maths department had to do with kicking out any metaphysicians. I don't get what that cabal is supposed to be, who the "heroes of the movement" are, what they failed at achieving, and really, what you are talking about right now.(put up some links or something!)

We are dudes solving problems that interest us. I don't get what grates you so much about that. If you think we are solving the wrong problems...what stops you from working on the right ones?

You're just baffling me now. I promise, I never kicked a metaphysician in my life! :D

James Gurney said...

Fascinating discussion, you two--I'm learning a lot from both of you. Suppose we grant that the human mind and the computer mind are fundamentally different, and that the computer may—or may not—eventually match at least some of the human mind's most subtle accomplishments. It already has surpassed the human mind in a lot of categories (such as computation speed), of course.

Doesn't that leave us with another question: How will the symbiosis between these two kinds of minds aggregate around art? Keeping in mind that computer programs are themselves works of the human imagination, will you both grant that some forms of imagemaking that will be consumed by humans may be accomplished by computers at a level that simply cannot be matched by humans? I'm thinking, for example, of those abstract music visualization programs, which blow away what human animators were able to create by brute force in Fantasia.

Or there might be computer-guided responsive home companions (kind of a super souped up Furby) who can tune into the mood of the home user and crack really funny jokes based on the day's news, or make up a song and teach the human companion to sing it in harmony. There are already robotic stand-up comics that respond to the audience and actually keep an audience laughing. Isn't that art of the most subtle and human kind? What can we expect in 50 years if such technologies continue on their course?

Whether a computer can match a human's style of representational painting is probably not the highest use of AI--as many have pointed out, it's a relatively trivial exercise. My point is that it doesn't really matter if robots eventually think and perform things that we're good at. They are on the verge of becoming the kind of companion minds that as humans we can relate to in ways that we're only beginning to imagine.

António Araújo said...

ps: It come to mind that Bertrand Russell used to say about some philosophers "of course that is just a kind of madness", or something to that effect. In the cases he referred to, I remember that I sort of agreed (in a not too serious way - rather than madness I'd say silliness). Is that the kind of thing you are referring to? Is the "great project whose heroes have failed" a reference to Russell's Principia mathematica?

By the way, Russell was mostly being humorous. His take on Nietzsche was hilarious; I had to laugh, and I was a big fan of Nietzsche at the time (that faded somewhat along with adolescence).

"His general outlook remained very similar to that of Wagner in the Ring; Nietzsche’s superman is very like Siegfried, except that he knows Greek. This may seem odd, but that is not my fault."

António Araújo said...


sure, I guess that computers really will shine not in emulation of us but where they can augment our possibilities or complement them. I'm glad you called it "image making" rather than art, as this makes it safe for discussion; and in that respect we already have examples, mostly in 3D, where computers really help to make huge group scenes, for instance, that would be really hard to do by hand. Finite automata are great for that, with flocking algorithms, physics simulations and the like.

And yes, data visualization is a huge deal.

> (kind of a super souped up Furby)

Now you scared me! :D I'm about to defect to the Luddite side of the fence! Please just don't bring up clippy! :D

>There are already robotic >stand-up comics that respond to >the audience and actually keep an >audience laughing.

I haven't seen this. Do you have a link?

>They are on the verge of becoming >the kind of companion minds

That is it right there!

James Gurney said...

Antonio, Agreed!
> "I'm about to defect to the Luddite side of the fence!"

Is it possible to be both a Luddite and a technogeek? Because that describes me. My ancestor Goldsworthy Gurney was attacked by Luddites and had to give up his steam carriages in 1830. I'm part of that tradition of engineering, but I'll also be the last guy standing with paint and hog bristles.

Re: Silicon-based comedy:

P.S. Love your work with Urban Sketchers.

bill said...

I am going to agree with James. This has been a great discussion. It usually has merit when Kev is involved. Antonio I must say I really enjoy your viewpoint here.

kev ferrara said...

Antonio, you can know the truth of that first assertion by thinking about how information and meaning are encoded in linguistic communications... versus what kind of information is contained in a mathematical expression. Quantization of information is a highly lossy operation, firstly because there are only certain aspects of understanding that are quantizable. Everything else is qualitative. And secondly, in order to mathematize some thought, we must consider all the non-math aspects as noise to be ignored in order to hone in on the mathematical signal... or some mathematical signal we recognize. (math having its forms.)

So firstly, we must begin with a quantizable thought. Then we must strip away all connotation, the specifics, and aesthetic aspects until the thing is reduced down to a mathematical essence.

Now what equations or formula or algorithm should you begin with to reformulate my thoughts of my grandmother? How do you begin to calculate and remodel the dim memory I have of her voice? The texture of her hair? The smell of her perfume?

You see, math begins without the substances that build out the thought.Whereas a mathematizable thought, already has the quantization built into it, or else you wouldn't be able to distill it down and find the mathematic foundation within it.

Cognitive Module Theory: If you think about the implication of mirror neurons, you don't need cognitive modules. Nor is there any evidence for cognitive modules as such. So why believe in it?

kev ferrara said...

Hi Bill!

António Araújo said...

James, I think you sure can be on both sides of the fence at once. I feel the love-hate thingy for computers quite often. I used to love to program my own games when I was a kid and at the same time I hated the little fiddling with parenthesis and whatnot and all the bug hunting that comes with programming the dumb machine that never guesses what you mean. I prefer to think about programming rather than do it, most of the time! :D

Same with maths, really. I love beautiful maths, but I cannot stand repetitive and careful number (or symbol) crunching. I'm the kind of guy that cannot balance a checkbook because I get distracted midway. And I always have errors in my calculations. Sometimes I just want to throw it all out the window. But then I see a really pretty piece of math and I get all fuzzy :)

And regarding art and computers...I love the concepts, and admire all the work, but I'm the kind of guy that in photoshop just uses the brush and the color picker. Cannot stand fiddling with filters and all that. Layers are my limit. But that doesn't mean I cannot appreciate when other people do complex digital work. It's a bit like scientific illustration - I learned it, love it, admire the people who do it, but I know it is not for me. Or perspective. I teach it, but when it comes to my drawings, I just wing it. I feel both the need to know it, and the need to not rely on it. Same with color theory. There's an appreciation for precise knowledge, and a pleasure of not feeling tied to it while actually drawing.

So yes, I understand the feeling of being a techogeek Luddite :) - and oh, yeah, I'll be there to fight you for possession of the last box of real, dirty crayons! :D

>My ancestor Goldsworthy Gurney was attacked by Luddites

Damn, you have the coolest family tree ever! :D

>P.S. Love your work with Urban Sketchers.

Coming from you that really warms my heart! :) Thank you so much for being kind to my doodles. :)

By the way, speaking of machines, art, and urban sketching, I started carrying an android phone for night sketching about a year ago (before that I used a nintendo ds). It is really cool for making color studies when you have no lights around or you can't really deploy your gear (I usually carry watercolors and a clip-on reading light, but sometimes it is just too fiddly). And you only use the brush and the color picker on the phone app! Simplicity rules! :D

Thanks for the link to the video!

António Araújo said...

kev, I apologize but I am over my alotted time today, I have some work to finish urgently, and I don't want to answer you thoughtlessly. I'll answer tomorrow when I can parse you properly.


António Araújo said...

Of course "mathematizing an emotion" (and I think you mean the emotions you get from your memories more than you mean your memories) won't be writing an equation for it. There is this weird notion of what mathematics is. (by the way, maths isn't all number crunching and equations - you can make maths with pictures, for instance! maths is about playing games with rules; and about the creation of interesting rules also, but I diverge into dangerous territory; trying to define maths is like defining Art - maths is what maths does)

Let me play a game around your question.

Suppose I come up to a computer and tell him: I think I can make an equivalent of you in a ticket tape. The computer might be sceptical; he might say "look, what I do depends on how my electrical circuits fire, it is all about silica and electricity meshing together interestingly", and also "how can you ticker-tape the delightful heart-warming (or CPU-cooling) sound of my fan going off?", or the way old memories fade away from my RAM when I reboot, or the way broken jpegs still linger in my hard disk in spite of having their references erased? And the computer might be right, except that we happen to know that in fact he isn't, at least in a sense: all computers are equivalent to a mathematical turing machine that can in fact be modelled (in principle, although horribly inefficiently) by a ticker tape on which we do simple operations. Of course the ticker tape won't have a fan to delightfully whirr (the analog of the grandma smile in this little game) but it may contain the process that would command a fan to whirr, and, in that sense, there is an equivalence between the two. In a computational sense, it doesn't matter if the computer is made out of chips or made out of ticker tape or made out of pebbles - some people have fun, in fact, creating turing machines within larger systems, be it with physical objects, be it simulated inside computer games themselves.

Now of course I don't know - and that is the whole question - if the coming together of chemicals, carbon, and electricity has something fundamentally different that changes everything. Maybe it *is* fundamentally dependent on the building blocks in a way the computer isn't dependent on chips and electricity. Certainly "meat" is hugely complicated. Also, there is the notion of equivalence - our "computer equivalent" might be quite underwhelming in many regards even if it can be made. You can imagine the computer looking at his computationally equivalent ticker-tape and simply not accepting that the sense in which it is equivalent is the important sense. He might simply refuse to recognize it as an equal even if it could make the same calculations: "it is just mimicking me!". Again I am back to my previous answer: you have to try it out and see both how far you get, and what you get when you go that far.

kev ferrara said...

I think you mean the emotions you get from your memories more than you mean your memories

No, the point is and was that such things are all tangled up. It is common form of thought to have a visual memory running like a movie in the mind, tied with a memory of smell , and talking, and along with an emotional tone that is non-specific, etc. Such a thing isn't something you can build with math because it begins as a synthetic mental event. It isn't organized or arranged out of some logical necessities, some necessary sequelae from first principles, if you know what I mean. It isn't built from some diagram, blueprint, or rule set. It is built from synthetic imprints, if anything, with symbolic emotional values.

The brain has an unbelievable capacity to take imprints. It is impressed by all five sense all day long. Our experience is constantly being synthesized from this incoming data. And this data is analog, unpredictable, and everywhere. The ability of meat to respond to electricity and chemicals, to touch and light, to warmth and cold, to smells and sounds... this is quite the sensitive substance.

it may contain the process that would command a fan to whirr, and, in that sense, there is an equivalence between the two

No, I don't agree with your characterization of that as equivalent. A command is not an encoded event. And I disagree with your overall comparison. You keep neglecting that the brain doesn't just compute, it computes in sensations. Thought isn't just 0s and 1s flying by a mechanical sensor that only picks up 0s and 1s. Thought has meat, chemicals, and electricity interacting. The idea of computational equivalence isn't the issue, because sensations have qualitative content. And art is a qualitative business.

WWick said...

James has a point. Let's put aside the argument whether a machine can possess a human-like mind and just think about what kind of mind it can have and what sort of art it could produce. It may not be able to reformulate Kev's thoughts about his grandmother, but it could be fed an awful lot of useful data - images, video, documents, maps, GPS data, medical records – even DNA samples and all that might one day say a lot about someone's future grandmother. Hooked up to a Vangobot with a plethora of sophisticated filters to choose from, I have no doubt such a machine could produce a decent painting, at least one that many people define as decent in the conventional sense. In fact, hundreds of randomized versions could be produced, limited only by the amount of money one would have to spend on materials. And that's not counting sculpture via CNC machines or 3D printers, or other modes of artistic expression such as video, holograms, or 3D graphics.

Does a painting produced this way suffer for not having been modulated by the complexity of human thought? A computer may be comparatively dumb, but a painting is even dumber in that it can't think it all, feels no pain, senses no joy. No matter what goes on in the human mind, the human painter must express feelings by way of visual codes, and in turn, one has to in some way be visually literate to feel it. To what extent that those visual codes can be aped by a computer may depend on who's looking. But if one sees art as a communication between one human and another, the machine-produced art is simply a dead end. If we come to love robots like we do our fellow humans, well, that's another story. I'm not holding my breath. I must say though, I do like Antonio's observation that Google is already asking us to make that distinction in the comment box.

kev ferrara said...


1. I don't understand how you plan on converting a bunch of data into art that "says a lot about someone's future grandmother." Says a lot of what? Data points? Is art a spread sheet to you? And how are filters going to help?

2. Filters, sophisticated or otherwise, don't make art. Otherwise every kid who plays around in photoshop is making art by applying unsharp mask to a photo. In order to understand (or should I say, believe) why applying a filter doesn't make art, one must have a logically coherent understanding as to the difference between art and not-art. And it has been the mission of a million culture vultures to destroy the ability to make such a distinction. It is my mission to destroy these culture vultures because they are perpetuating cultural ignorance.

Does a painting produced this way suffer for not having been modulated by the complexity of human thought?

The issue is not just complexity, but the brand of complexity. You need to spend some time thinking about how a work of art can encode thought and emotion. Yeah, there's codes... but they're built of completely understandable concepts. Concepts and conceptual artifacts we have innate facility for. This is an enormously deep matter that is getting very shallow attention here. You cannot get away from the question of human encoding. And it keeps tracking back to the problem of sensation. Thought is built of sensation and sensation is emotional. So how does sensation build thought?How does sensation become knowledge?

Until the centrality of that question sinks in, we're all just talking past each other.

Art is not a filtration. It is a synthesis.

António Araújo said...

WWick, exactly, that was my point 2 above. However kev defines art, people will respond to what people will respond. Whether a dumb PC can make art that convinces people is an empirical question.

> It is my mission to destroy >these culture vultures

I won't stand in the way of a man with a mission!...;) But I do wonder what you'd do if the PC fooled you in a blind test. I mean, look at that landscape with the duckies. If a friend came up to you and said "I just did this" would you scream "robot!" immediately (if you didn't have the photo to compare to)? And would you say it was all that awful, or would you just pat him in the back and say "keep working, you'll get somewhere"? Come on, it's a nice little trivial plein air painting like so many others made by fleshbots.

Back to issue 1, WWick's comment just reminded me that holograms are a pretty good analog for some of the aspects of how our brain stores information, memories, sensations, etc, all in a jumble that isn't located anywhere specific...and yet holograms are pretty amenable to a mathematical understanding and engineering production.

Also: You keep saying that some things are qualitative, as if that put them beyond maths, but in fact much of maths is qualitative. Please do not mistake maths for accounting. Math is not just about numbers.

As for synthesis, I can have a zillion different sensors on a robot, pouring data into an artificial neural network that responds by firing artificial axons, and the pattern of firings that results is a synthesis quite analogue to what you describe. Only it isn't happening with real neurons, it isn't using proteins and all the other juice. Again all that remains is to claim the juice is essential. So we are back to the same argument. Although we could probably make a neural net that uses proteins, by the way. But then I bet you'll say we miss something else.

You say the brain computes "in sensations", but in a final analysis, at a lower level, the brain is a bunch of neurons and other stuff performing physical processes. Somehow, from all of that, sensations arise in our mind. If we create analogs of the physical stuff that happens in the brain, who knows what can arise from those analogs. Sensations of some entity? Maybe yes, maybe not. Again, same argument.

Also, sensations are the superficial layer. There's a lot going on in the brain that can in no sense of the word be described as "sensations", so I really disagree with the contention that the "brain computes in sensations". Just think of one of those famous patients of Oliver Sacks with brain damage that have sometimes no awareness of complex processes that the brain keeps executing. Those cases really speak for a modular mind, by the way, and for the conscious mind as a pretty clueless outer layer.

By the way, we don't start with sensations, we start with a few cells replicating themselves in a puddle and somehow after a few months those dumb cells engineer a being that has a mind. Seems hopeful to me.

Anyway, I fear I've been rephrasing the same argument for the last couple of rounds. I think that from my side at least there's not much to add, and I have a deadline coming up, so I think I'll call it a day.

Thanks for the discussion, Kev (and James, and everyone else), it's been fun. I'll keep peeking once in a while but I probably won't say anything more for a few days at least. Cheers!

kev ferrara said...


I'm feeling your last post was a bit more rhetorical than your previous ones. I understand you need to disengage from the conversation, but some of the points shouldn't stand and I feel they should be addressed whether you answer or not.

I am not moving goalposts, as you accuse. If you want to believe that a photograph is qualitatively the same as a Velasquez, your argument has already been won to your mind.

If you want to believe that taking a photograph into photoshop and running a few filters on it makes it a work of art, your argument has already been won to your mind.

No, I wasn't fooled by Vangobot's renderings into thinking they were done by a human. I did think they looked nice. And I would not have been surprised, outside of the context of this discussion, to have found out they were by a human. However, it is obvious that they lack compositional thought and emotion, so regardless of their author, I wouldn't have had much to say about them.

I prefer to take your argument at its most meaningful, its most powerful. Which is to say, that a machine will one day be able to create art equivalent to the best creative works of art man has made. Not the cheap junk.

The main requirement of this question is to try to derive the qualities present in the best works man has made. And then think about how a computer might compete.

The "master" integration of a mind's sensations is a personality, a worldview, a pattern of thinking, which has a unique mood... a signature of a soul, its own vocabulary of conception. Nobody knowledgable who looks at a Mort Drucker drawing will mistake it for an FR Gruger drawing. Instead of glibly dismissing such a point, ask yourself why not?

Fyi, the entire brain is functioning by sensations, whether we feel these sensations or not. Just look at the architecture.

Again all that remains is to claim the juice is essential. So we are back to the same argument. Although we could probably make a neural net that uses proteins, by the way. But then I bet you'll say we miss something else.

Put your finger in water. Lift it out, blow on it. Put it in a light socket. Put a match to it. Put it against a vibrating tuning fork. Squash it between the pages of a book. Amazingly sensitive this meat-stuff, ain't it? Same piece of meat feels the unique qualities of every different thing you put against it.

Roberto said...

WOW! Bravo! Bravo! Well done… and with sincerity and respect too!
(I just finished catching up on ‘The Journey.’)
While I’m no TechnoNerd, as a street-wise muralist and dilettante autodidact I am very much enjoying the conversation and well-made points.
As to a much earlier part of the discussion…

It seems that what the camera started to do, the computer will inevitably accomplish. Not just animators and illustrators have felt the bite already, but also background and set painters as well as billboard artists and sign-painters (all highly skilled and talented artisans) have gone the way of John Hennery’s mighty hammer. That said, there will always be a place at the table for Leonardo, Buonarroti, Von Rijn, Matisse, Du Champ, Ai Wei, and all those talented shaman/artist/tricksters who flatter, entertain, and challenge us by reflecting back to us our humanity with genius. (As for the rest of us… Que sera sera, let us dance and sing while the sun still shines! Would you like fries with that?)
Point 2: Prove you’re not a Robot.
‘Sentient meat does-not-equal sentient silica.’
As you have noticed, Blogspot in its infinite wisdom, asks us to prove we’re not robots, not that we’re not zombies. A zombie is a Human without sentience (or a fleshbot), so I would say a zombie would be the meaty equivalent to a non-sentient-robot.
Zombie = fleshbot = Robot.
While there may not be any robots contributing to this blog (yet), there may be many zombies. (I’m suspicious of all those ‘anonymous’ posts, and many of the pseudonyms;) Which reminds me:
@ Kev: your link to your website on the blogger I.D. gives this message:
[Reported Attack Page! This web page at has been reported as an attack page and has been blocked based on your security preferences.
Attack pages try to install programs that steal private information, use your computer to attack others, or damage your system. Some attack pages intentionally distribute harmful software, but many are compromised without the knowledge or permission of their owners.]

Back to it…
A sentient robot would be, in many ways, far superior to a human in terms of strength, ability, sensitivity, computation, memory, etc. etc., while not being a ‘living’ creature in the same sense as an organic, mortal-critter… In essence a wholly separate category of being.
Therefore: Sentient meat (does-not =) Sentient silica.
(Hey! Check it out. I’m doing math with concepts!) ;p

Point C:
What is not at all trivial are the implications in real time, or near future time, of these developments in robotic technologies for use by the rich, powerful, and possibly not so well intentioned sentient-beings, guvments and guvment type military agencies, and weapons manufacturers. The potential for abuse and misuse is staggering when you start considering nano-technologies, drones-as-surveillance, and drones-as-weapons. You don’t even need to get close to strong AI or sentience for the implications to be staggering! Once we stick our fingers in that mess there’s no turning back. Resistance is futile! (We’ll soon be looking back from a brave new world order at the quaint and naïve days of plein-air painting with our furbies.)
Qe sera sera, let us dance and sing while the sun still shines! -RQ