Gen AI: Finding a Middle Way

In a recent toot, Sara made reference to “intellectual property theft” in the context of generative artificial intelligence (Gen AI). This is one of at least several well founded objections that I’ve seen a lot lately. Like many people, I’m still trying to work out a consistent, rational position on Gen AI. This post is not that by any stretch, just some thoughts.

I’m not sure about the term “theft” here. It may be true sometimes but it fails as a blanket characterization of Gen AI. Consider a few hypotheticals: ^[1]

• Suppose I am an aspiring artist who has studied Pablo Picasso extensively. I can recite facts about his life and work and I can recognize (and identify and describe) his paintings. I can even recognize cultural references in television shows or everyday conversation. I can answer my friends’ questions about Picasso with some degree of authority.

• In the course of studying Picasso, I have learned to understand his form and technique. I’ve similarly studied many other artists’ work. Over time, these explorations have helped me to hone my own personal style. When a teacher gives me an assignment, I can draw on my experience to create something new. As is often the case, the influences are sometimes apparent.

• A friend asks me to paint a portrait of him and his partner in Picasso’s signature Cubist style. This is not the type of work I usually do, but I agree since we are good friends. The result is obviously not a Picasso, but the stylistic resemblance is unmistakable.

• I realize that I’m pretty good at painting in the Cubist style, so I take my show on the road and begin selling paintings at art shows. Everyone always comments that they just like Picassos. I make a lot of money.

In each of these four scenarios, I’ve leveraged knowledge gained through study. I did not pay Picasso or his heirs for access to this knowledge. ^[2] Did I at any point along the way commit intellectual property theft? If so, where is the line? If a Gen AI model does essentially the same things, are the boundaries different? If so, then why? ^[3]

Does it matter how I acquired my knowledge and skill? What if I had studied at university? What if I was self-taught through watching public television and visiting public libraries? What if I had paid another Picasso expert? Note that in all of these cases, money changed hands in one form or another.

It occurs to me that a key factor is profit. If so, who is profiting? In the first scenario, no one is profiting. One person is asking a question of another person who happens to know the answer. I use Claude this way a lot. ^[4]

There are some genuinely bad possibilities, and we should of course be cautious about those. But there also are some applications that could make the world better. We should not be quite so quick to dismiss or (worse yet) to condemn Gen AI outright. We need to keep moving forward. The world has a way of doing that with or without us anyway.

There seems to be a continuum ranging from “obviously acceptable” to “obviously unacceptable” in the scenarios I contrived above. But there also seems to be a great big, fuzzy gray area in the middle. And finding our way through that vast middle is where the struggle will be.