Monday, August 7, 2017

Who Makes Better Art—Humans or Computers?

As abstract images go, do these images strike you as creative, expressive, and artistic? 

A group of test subjects thought so. They were ranked highest out of a large selection of sample artworks in a blind test.

The samples included artworks created by up-and-coming human artists from ArtBasel together with images created by computers using machine learning algorithms.
"Researchers programmed [the computer system] to study 80,000 WikiArt images of Western paintings from the 15th to the 20th century so that it knew what kind of images have traditionally been aesthetically appealing. But the scientists didn’t want to devise a system that could merely emulate history paintings, genre scenes, landscapes, and portraits in established styles — a machine that truly has artificial intelligence, after all, must be creative. Once the system learned these styles, it then worked to deviate from them."
What surprised the researchers was that the images deemed most original and creative by human observers were created by the computers, not the humans.

And these weren't just computers aping the superficial art styles of existing artists. They were original visual ideas that expert art-watchers had not seen before.

"Researchers tested whether or not these generated works could pass as creative to some people. An object, for their purposes, demonstrates creativity if it is both “novel and influential.” The first question they posed was whether humans could simply distinguish between the computer’s art and human-made artworks. As Elgammal sums up in a blog post, participants believed that the generated images were made by artists 75% of the time, compared to 85% of the time for the collection of Abstract Expressionist artworks, all made between 1945 and 2007. In terms of the Art Basel paintings, participants thought that humans had made them just 48% of the time."
The finding raising profound implications about the uniqueness of human creativity. Is what we do as artists a personal and emotional act of genius, or an optimized set of choices based on available information? If it's the latter, it is something that a computer could figure out.

It also raises questions about what we expect as viewers of art, because, after all, art is really about communication. Does it matter to us who creates a work of art? Do we want to surround ourselves with images that have passed through a human mind or hand? Or are we willing and eager to be surprised and delighted by what our bot-brothers invent for us?
Read the article on Hyperallergic


Sesco said...

I believe viewers of art care not the source of the creativity. If a computer delights me with what it has created, then I personally cannot be disappointed to learn of its origins. I would hazard to say that the bulk of viewers of art do not seem to care whether art is original or from a print of an original if it delights them in some way. Home Goods, Lowes, Big Lots, etc. sell a ton of prints. There may be a smaller faction of purists who understand and feel the difference between a print and an original, but this is a matter of perspective and scale. I suspect that computer programmers will come to compete with painters just as photographers do. Since the invention of the camera I think painters have already felt the sting of this competition. Now they will also feel the competition from computer-generated creativity. I think painters paint for many reasons, not all of which has to do with catering to the decorative public, and I believe there will always be a few who delight us with unique creativity. Computers likely still have a growth curve ahead of them as they become more able to process terabytes of information at greater speed. This implies that their creativity will expand, or improve, as programmers direct the output of such creativity. I'm not as sure about the growth of human creativity. I can see a future where we abdicate our ability to create to machines that 'appear' to do it so much better. Even now apps allow rank amateurs to modify photographs in an artful way. The holder of the phone just chooses a creative effect and says, "Look what I've done!" If computers begin to create that which cameras cannot, then painters will truly be marginalized. Hasn't this already happened?

Ben K said...

The relationship of prints to originals strikes me as being similar to the relationship of MP3 music files to vinyl records. The grooves in the record are an analog response to sound waves generated by a human voice, and the MP3 is a digital approximation of that process. Likewise, an original painting bears an unmistakable record of human touch, one of the things I love best about seeing originals. It's a connection between humans and a chance to get inside someone else's head and literally see a reflection of their internal world. A good print loses much, but not all of this. I don't mind looking at a print when the original is not available.

Computer generated art doesn't come from the same place, though. No matter how much I like the images in this post, I feel a sense of let down when I learn that the connection to the artist isn't possible. I am impressed by the achievement, but not moved.

Sesco, good point about the introduction of the camera. Perhaps this can be a useful tool (much like photography) if used carefully.

James, thank you for writing about this. I didn't really appreciate the importance of the process of creation and an understanding of the history of a piece is to my enjoyment of art.

bernicky said...

There was a wonderful project involving Microsoft and Technical University Delft to create a new "Rembrandt" portrait and the computer did a bang up job. There's no reason not to expect that machine learning can excel in the abstract world as well. Whether or not people care is not likely to enter into it because most people don't care where the art comes from they care what it looks like on their wall and how much square footage it takes up. People who collect will always care because as much as we collect art we also collect stories. I have stories about every artist in my small collection and, strange as it may sound, stories can influence collectors. The story behind Picasso's Nude, Green Leaves and Bust and the affair behind it is at least as interesting as the painting itself and to my mind more interesting than the fact someone paid 106 million for it. You can be assured that whomever paid the 106 million though was more than familiar with the story of Picasso, his affair and his life, because that is piece of what they bought.

Regardless of how good AIs get at creating art there will always be a market for human created art just as there is for hand made clothing, farmer's markets and a host of other things which people see as having intrinsic value because of the human component in their creation. It could be that the market will shrink to the current market size of bespoke buggy whips, but there will be a market.

Black Barbie said...

You might like this

Emanuele Fittipaldi said...

I think that a computer generated art will always produce some kinda of standard pattern, but even though it doesn't there are few things to consider:
1)surface quality Is also an important part of the painting , because the thickness , the way it is layed on the canvas, the material that was employed , cannot be faked by a computer who can only generate a 2d thing.
2)a painting done by a human being , always carry its emotions and mindset at the time it was created, it will reveal how the mind of a human works , its experience , and for me it is the most valuable thing in a piece of art
3) the human nature is very unpredictable, when we make a piece of art we make so many choises , so different that allow us to be unique every time and this Is what makes art so requested.
4) this just applies to abstract , but abstract isn't very much a creative process, it just consist in painting appealing shapes of color , and it sounds like a very basic thing to a computer to accomplish and i am pretty confident that a pc will never obtain the power to create any more than that

arturoquimico said...

I think folks are missing the point here... to me it is about where creativity comes from... without getting into Creator, creation or machine... let me say... my day job is chemistry and I can tell you that many of the most creative developments have come from accidents... like Post it Notes and plastics... creativity can come from inspiration, but as Thomas Edison put it... lots of it comes from perspiration... I'd imagine a great painting often comes from some blah thumbnails and drawings... So what if a computer comes up with something interesting... the artist can humanize and make it better... take it to a higher plane.

Chris James said...

The criteria by which most people, including so called experts and critics, judge what is "good" or creative AE work and what is not are often nebulous at best. So what are we to draw from this? Chiefly that the general viewing public doesn't care the source of art (and thus, at least in this type of work, humans potentially face competition from computer), and secondly that good enough is good enough for the larger section of markets.

What I don't think it does is, at least from a technical standpoint, speak to the potential of computers to equal or outdo humans in art making. So far the examples mentioned in this post and comments are either abstract (where a lot of b.s. can be and has been passed off as brilliant, depending on who you talk to)or journalistic, i.e. creating a portrait in the style of someone else. But what about narrative work? Or anything more than organization of shapes and colors? It's one thing to recreate the look of a Rembrandt portrait (after all, we don't call the many forgers and imitators of that artist creative geniuses), but another to conceive of The Night Watch, from the initial germ of the idea to the choices made along the way to its completion. Intent, meaning, interpretation; selection from a large pool of facts and possibilities and personal experiences. This is what humans call upon to create such works. Will an AI do this? What legible ideas and concepts can it come up with on its own? What personal experience can it draw from that will inform its interpretation and expression of those ideas and concepts? What kind of selection process will it go through in determining what makes it into the scene and what does not?

From my perspective as someone who has little interest in any medium for its own sake, or depictions of things that exist (e.g. the traditional genres of landscape, portrait, still life. I'd rather look at Dali or Giger), I'm not much sympathetic to the fears of the painter when it comes to competing with machines in the market, outside of the sympathy I have for anyone who desires to make a living doing what they love most. People will enjoy what is interesting or beautiful to them, so the key is to make things more interesting or beautiful than the next human...or machine.

Siver Black said...

Did the computer enjoy making the art as much as the human artists did? Maybe this will be the point where the AI machines take over. The machines decide that human civilization should be brought to an end because of inferior art.

Anonymous said...

I would argue that humans still created the art. The input of thousands of humans are what enabled the AI to make the art, since it has no self-aware intellect itself. If we pour paint down a canvas and it drips in random patterns, we don't give gravity and the texture of the canvas the credit. The AI is just a new tool with human intelligence and thought behind it, albeit more people involved and more removed than most art.

People may give it value, but it's more novelty than anything.

scottT said...
This comment has been removed by the author.
Eugene Arenhaus said...

Leaving aside the comment that this may still be considered indirectly human-made (a human trained the neural network to produce a particular kind of images, after all), this says more about the experts than about the images.

Sesco said...

Isn't art wonderfully complex? Peter, perhaps AI will be retired in the future for reasons much like the reason we stopped building nuclear bombs and asked everybody to get on board. We scared ourselves with what we produced. Ben, you said you 'liked' the images, not that they delighted you. There is a difference. When I first saw computer-generated images that approximated the galaxial images of Hubble, I was delighted. I'll never forget first seeing those images, created by a computer. bernicky, perhaps there will be unforgettable stories of disturbed programmers who cut off their ear which will boost the value of the creativity of their program in the eyes of the public. Perhaps I need to learn how to code if I am to stay relevant. But I love to apply paint, just as I love to turn the pages of a paper book. Some of us don't have enough time to stay relevant.

Laura G. Young said...

I for one welcome our "bot brother" overlords. (But what about their supercomputer sisters...?) ;)

Bug said...

This is frightening to an old painter such as myself. I already feel passé, doing straight ahead landscape work that might be called impressionist based. I already feel passed by when the technology of my cell phone is too much to bother with. Now people are falling in love with computer generated images. Most local "art" is music and theater oriented. People are buying canvases that look good in the room because of the matching color schemes. Art history is considered not a precursor but an artifact. But then, I admit to never having had an energy drink. So, I suppose, life marches on. You either go with it, or languish.

Unknown said...

I feel this is a bit misleading. Maybe I'm bias against abstract art but it is relatively easy for a computer to generate it and mislead us because as the paper puts it "Existence of figures or clear subject matter might directly bias the subjects to conclude that such paintings are done by human when contrasted to the generated images which lacks such figures". I feel they should have done this anyways, wouldn't it have been interesting if all the art lacking figures was concluded by people to have been generated by machines? If a changing context changes the result, doesn't that mean the results say more about our psychology than they could possibly say about the AI?

I would have been more impressed if they'd done landscapes or something like that.

Also it's interesting to note how much nosier the top ranked images where compared to the lower ranked images. I think this says more about us and how we like, as the paper explains, images that produce a certain amount of arousal (not too much, not too little) more than anything else. Also because of this I would hypothesize that if people had been shown images with figures, although they would have been recognized as being machine made, I still think they would have been rated higher in creativity (because distorted faces, etc) since they would technically be more novel/original and that is what they were specifically measuring for in their questions. Same thing would have happened with the landscapes imo.

I find it all very fascinating from a programmer's perspective that the AI can be seen as creative, but as an artist I'm often rather irritated by how stuff like this is often reported as if AIs could replace us any moment now. I'll be worried when an AI can create an original work with a story (that it intended and others can understand), when it can design, etc.

Unknown said...

BIG CANVAS Education was founded with a vision to impart quality and affordable education| Best CA Coaching Institute in laxmi nagar, Delhi. Our student’s performance and feedbacks speaks for us. VIEW MORE - CA institute in delhi

A. said...

It's because it's abstract art, not real art. Computers cannot make art. I don't think modern art is art because I don't think anything anyone decides is art is art. Art is more than that. Not just anything can be art.

Unknown said...

If our species had more sense, we would program computers to buy art instead of make it.

I find it ironic that I had to pass a captcha turing test in order to post this comment.