Generative AI is going to change the world, and change how we do product design. Many vague AI-as-a-panacea promises have been made, and many checks cashed with the promise of essentially easy access to unlimited, perfect knowledge. I have heard it described as a question of when rather than if. Bold claims. How can we come to an understanding of what generative AI is capable of, what it’s role may be, and what the trajectory may look like over time?
There are things that computer models are excellent at and which our brains are not. In a sentence, I would say that computers are materially different in terms of processing and/or finding trends in large sets of data. As noted elsewhere, I have used neural network models to process information contained in fluid flow simulation results. The neural network was trained on the data set and was serving as a useful surrogate model in no time at all. If my brain would have been able to do this, it would have taken me many months, at a minimum. This is a clear case where the neural network was serving a uniquely valuable function; basically serving as a data-processing extension of my brain. In the future, could a hyper-logical AI have guided the design process as well as carried out the mechanics of the work – basically, could it have replaced me?
Perhaps the nearest milestone is for AI to replace me at a level which is equal to my performance, and it could probably do it cheaper and that’s fine. But, is there a realm of hyper-logic, heretofore not accessible to our brains, which deep learning is going to reveal and put us all to shame? Computers beat humans at games (chess, go, etc.) all the time these days it seems. Is the computer really better or more-or-less equal? If we saw AI showing a behavior, could we learn from that behavior and advance through that, so that the AI was just acting like an ultra-smart mentor, but one that we can keep up with once we see it in action? In the case that AI does beat a human, where might the AI’s advantage come from that would allow that to happen?
In terms of replacing white-collar jobs, we can conceive of various milestones for AI: 1) Smart Buddy Use– AI as a supplement to human workers brain capacity and expertise, 2) Worker Replacement– AI making decisions about design direction and also carrying out the computational work of the task, or 3) AI Revealing Unique Super-Human Logic– where recommendations made by AI are essentially beyond our understanding, and which we may or may not be able to understand after it is shown to us (basically, AI has the same relationship to us as we do to monkeys). We’re currently firmly in Phase 1, but could we move to Phase 2? Is Phase 3 even possible?
How to get Behavior You Don’t Understand
Speaking as a person who is well versed in getting non-understandable behavior out of, well, any system, really, I know how to do it. How you do it is ignore, gloss over, or otherwise fail to understand the functions of the components of the system. Time and again in my life the message comes across: the fundamentals are always important. It is possible that AI can track and reason with more and more complex fundamental rules than us. But I would note that evolution saw fit to take us to the current state that we have, and not further (so far). My experience making predictions in the past suggest to me that it is quite possible that a super-human level of reasoning is not valuable, because of huge non-linearities and effects coming out of the blue to change the situation (see: COVID).
But, isn’t that the point of this AI thrust? To get behavior that we humans don’t understand? Either an idea, code snippet, design of a part, or process that we do not a priori understand and wouldn’t have been able to come up with ourselves? So that humans are relegated only to verifying that a solution from AI is correct? To provide real blockbuster unique value, it would have to have a kind of hyper-logic or access to unique abilities which are different in kind from our own. Otherwise, it’s just a smart buddy who you can collaborate with and borrow expertise from. I don’t have a doubt that, in time, we will take this technology to its full potential. But what the full potential is has yet to be seen.