The recent decision by Open AI to make a DallE-2 a paid API service has triggered a few people online. The question being, does OpenAI have the ethical/legal grounds to charge money for DallE-2 access when it was trained on other people’s work? Even though I will use DallE as a placeholder everything in this post applies just as well to the myriad generative AI art models that have been released recently (Imagen, midjourney, Craiyon and others).There are smart people on both sides of this issue, so I am to going to steelman both sides to understand them better so that I can make up my own mind about it.
Point : DallE is a creative tool
Everyone agrees : Individual pieces of art belong to the artist who made them. Copying them and passing the forgery off as the original is theft.
However, artistic style is not proprietary. All artists begin as students imitating great artists of the past. Eventually they find their own style but even the greatest art is some mixture of imitation, inspiration and innovation. And so goes generative AI. It does not produce forgeries. It creates pieces of art that never existed before.
More importantly, DallE is not the artist but the paintbrush. If Adobe started simplifying the artist workflows in Photoshop by condensing common sequences of actions into one click actions, for example, change lighting from studio to outdoors to cloudy etc, we would praise Adobe for making their product better. As people learn more about using generative AI to its full potential, we are going to see human creators take digital art to new heights. DallE would be no more than a tool at their disposal, simplifying hours and days of work into prompt engineering. Accusing DallE of stealing is no different from accusing Photoshop or a paintbrush of the same.
Counterpoint : DallE is a copyright violation
You can break the spirit of the law even if you are following the letter of the law.
The common defense is that DallE does not reproduce exact copies of existing art hence it is not a copyright violation. But consider why is that the law, why is copying works of art illegal? What is the line between forgery, fair use and inspiration? Let me propose a framework –
First, the laws against forgery or copyright violation are not protecting a ‘fundamental’ right like life or speech. They are safeguarding culture and innovation. We want artists to make new art because human society seems to like art. Specifically society likes new art, art that is product of its time.
Secondly, copying art would not be a problem if the copied artist was never affected by it. Copyright exists because if we let the copying artist take value/reward away from the copied, we are disincentivizing both sides away from value creation – the copier because they know they can get the rewards without the work and the copied because they know they will get no reward for their work.
So what about borrowing ideas?
The letter of the law starts getting murky, but the principles described above can guide us better. If we rule against borrowing, the original artist benefits. If we rule for borrowing, the borrowing artists benefit. So the burden of proof becomes –
- do the actions of the borrower threaten or harm who they borrowed from, and if so, then
- is the borrowing artist creating sufficient new value to offset this harm?
Note that neither of these questions are about what exactly was borrowed, because that is not really important. All that matters is the incentive structure for creative work.
DallE and its ilk are clearly guilty on both counts. They threaten the very industry that they needs to exist. Borrowing between two humans can be permissible if the market is large enough for both of them to be happy and then some. Picasso is known as the most prolific artist having created 150,000 pieces in his 75-year lifetime. But DallE can create millions of pieces every hour for far cheaper, no artist can coexist with that. If you are only hurting one other artist, you only need to create a little bit of new value to argue for the tradeoff of innovation vs regulation. But if you are going to drown the entire industry in your wake, you better have a strong argument as to why you are making it better not worse.
Consider self-driving cars.
Self-driving cars would collapse the entire driver economy, so it needs to have a strong reason to be. And it does. Drivers don’t “add value” to driving, i.e. new drivers don’t create new ways of driving. Therefore replacing 1 or even all driver with robots does not diminish the value of driving to society, in fact driving will become safer and more reliable. Note I am talking about the harm to driving, not drivers. The harm to drivers is of course real and should be managed. All improvements in technology displace resources and people, and if society values the improvement it should use the gains from increased productivity and quality to transition the displaced into new opportunities.
So, if DallE was the greatest artist in history and replaced all existing artists, that would be fine. But the issue here is that DallE is not an artist. Its output is “original” in a meaninglessly literal sense, i.e. the precise combination of pixels may have never existed before, but that is like saying you invent a new number every time you calculate one more digit of Pi. Interpolation between two pre-existing points in space does not expand the space and DallE cannot create anything that did not already exist. It is copying and mixing existing artist’s work, but in a way that is superficially novel enough to drive the artists out of business.
Some people think they can use DallE like a tool i.e. they can mix their creativity and tell DallE what to make and how to make it and post-process it etc. Sadly that is but a temporary waypoint. Anything distinctive created by human artists can be absorbed by generative AI faster than we can make it. The economic space for human artists will get narrower and narrower until art can no longer be a career or a passion, it can at best be a hobby. Then a day will come when the world is full of copy-mixed ‘artwork’, deep in the uncanny valley where everything looks new but feels old. And society will long for something fresh but there will be no artists left to create it, just a fleet of machines rehashing the past.
My Conclusions
Ok, that was fun to think through. It is easy for me to get lost in the argument and get convinced by what you are saying at the moment even if I was just arguing the exact opposite.
If OpenAI had not gone commercial, DallE would be an unequivocal good; because it would just be a research milestone en route to AGI, which I believe is an unequivocally good objective. In chess and Go, it turned out that creativity was programmable. Artistic creativity may be much harder to learn, but it is possible that it is also programmable. I don’t think we have solved creativity yet; but they have found a potentially profitable local optima with big consequences for artists.
So what will happen? The good/agile artists will move upmarket to high-resolution, complex and commissioned work and will probably embrace the tech. The current generation of generative AI’s work will likely gobble up the low end art industry – stock photography, wallpaper art, magazine/blog covers, logo design. So both the arguments are kind of true in the short term and I think most consumers will be happy with this state.
In the long term however, as AI gets better the upmarket is likely to get smaller, so I can condense the whole set of arguments above into 1 question –
Which will come first – creative AI or the end of human made art?
The techno-optimist obviously wants creative AI to appear before all human art is replaced by AI while techno-doomers expect the reverse to happen. Both sides of the equation also have an implicit “never” option. Trad-optimists might hope it is impossible to destroy human artistry, that no matter how creative AI becomes, human + AI will always have something to offer that society values. The trad-pessimist obviously thinks creative AI is impossible, but artists could still vanish because narrow interpolator AI’s have choked off their livelihood.
However, the long-term outcome evaluation seems moot. The short-term outcome seems to be quite desirable which means nothing can stop society from greedy-optimizing towards it. So I guess we should just hurry up and solve creativity.
Niranjan S 