What is Imitative AI?

A dumb question, I am sure you are thinking. It would not be the first (yes, there are dumb questions. I do not get smart just in time to ask a question) from me, but I do think that the marketing around imitative AI has obscured critical aspects of what the technology is — and is not — that are worth discussing. I have talked about many of these issues in separate posts, but it is worth seeing an overview, I think.

First, imitative AI is not intelligent. It does not think or learn in the way that a conscious mind does. It is a set of algorithms that turn information into a set of statistics and then use those statistics to probabilistically determine outcomes. That is why it is impossible, per the scientists at leading imitative AI firm OpenAI, for these machines to ever stop hallucinating. And that is why what it produces is merely imitative — it can only produce, at best, a mash-up of what it has in its training sets. That is also why it has a tendency to produce output that bends toward plagiarism. It is an impressive technology in some ways, but it is not intelligent.

Imitative AI is a normal technology. This means that it is like almost every other technology ever produced — it is going to take away some jobs (most likely programming and customer service; translation was already in trouble do to normal machine learning based so I don’t think it really falls under the imitative AI blast radius) but will come nowhere near to replacing all of science or medicine or white collar jobs as has been claimed for it. In fact, it likely will have less impact even in areas it is best suited for given the aforementioned hallucination problem.

Imitative AI is also a marker of the sickness of our current version of capitalism. Having a normal technology that has a well-defined set of use cases that can, at a minimum, be used as a loss-leader (imitative AI programming will likely survive the bubble only if Google, Apple, Microsoft, etc. wish to use their models as an enticement to use their programming platforms and languages) should be fine. It is instead seen as a crisis. Wall Street demands that industries always ensure that the number goes up. It is no longer sufficient to run a profitable business that has a long-term future. You must run a business that constantly grows and eventually collects monopoly rents. Anything less is seen as a failure, no matter how profitable you are. The imitative AI boom, premised on the idea that it can replace most to all white collar jobs, is a sign of how broken our economic system actually is.

Imitative AI is a health crisis. Imitative AI has not only encouraged people to take their own lives, it has actively helped them to do so. People who use imitative AI are susceptible to a form of psychosis. Designed to keep people engaged, the chatbots flatter and encourage people, even when they are wrong or drifting into fantasy. Best case is some wasted time and a little bit of embarrassment. Worst case is a full blown mental health crisis leading to isolation or physical harm of themselves or others.

Imitative AI is, perhaps most importantly, a tool of authoritarianism. One of the central preoccupations of authoritarianism is the destruction of a shared reality. Imitative AI makes producing almost realistic videos and sounds much easier, allowing disinformation to flood the information environment. Imitative AI has been shown to degrade the skills of people who use it, affecting their mental cognition beyond just the specific skills. Imitative AI appears to be driving a decline in actual learning among students. Such a lack of skill and thinking ability is helpful to authoritarianism. It is also being used as a convenient tool to evade responsibility. The system produced the output. Pay no attention to the owners and creators who programmed it. A lack of accountability for the wealthy and powerful is another hallmark of authoritarianism.

Imitative AI is, to use the cliche, many things. Many of those things are unequivocally bad, and many more are mixed at best. It is important to remember, though, that imitative AI does not have to be all or even any of those things. We can prevent it from destroying livelihoods. We can make its owners responsible for its outputs. We can keep it far away from education. We can do all of these things and more. The bad uses are not inevitable — no tech is. Otherwise we would be having this conversation in the Metaverse, looking at our NFT collections, wondering where our legs have gone, before watching our 3D televisions. We just have to recognize both what imitative AI actually is and that we can, collectively, control what it becomes.