The large language models are trained by using information available on the Internet. So people expect the new AI to know everything, to be a sort of ultra-Google. In my opinion, that is the wrong way to think about the latest AI. People who are limited to that expectation are going to doubt that there is any “killer application” for ChatGPT and its relatives.
I don’t think of the latest AI is an augmented version of Google. I think of it as a computer that has learned to communicate on human terms. It is a computer that does not have to be programmed by coders. I think of this ability to communicate with humans as the superpower of the latest AI’s. Accordingly, I think that three more compelling applications might be robotics, mentoring, and animation.
Robotics
Think of work that is physically tiring, somewhat repetitive*, and mostly unpleasant. This is work that you would pay a robot to do.
*If it were perfectly repetitive, a robot probably already would be doing it.
Given a loaded dishwasher, washing dishes is perfectly repetitive. But loading the dishwasher is only somewhat repetitive.
If every mouth were identical, the job of a dental hygienist would be perfectly repetitive. But in practice it is only somewhat repetitive.
Today’s labor-saving machinery has fixed behaviors that are designed. Tomorrow’s labor-saving machinery will have flexible behaviors and can be trained. The ability to be trained will enable tomorrow’s robots to adapt to tasks that are only somewhat repetitive.
My bet is that as engineers learn to connect robots to the new AI, it will become much easier to train robots. It will be possible for a non-engineer to train a robot using ordinary language. That will result in a spectacular increase in the utility of robots. This will be one of the compelling applications of the new AI. But it will not appear until more work is done to build the AI-robot interface and until new robots are designed to be trained using that interface.
One possible problem is that the attempt to create a verbal interface with AI’s may prove difficult. AI’s have learned to “converse” by reading text. They may not be able to talk to humans in a way that seems natural, and they may have a hard time processing natural speech.
Mentoring/Coaching
We bounce our problems off of other people. They might be friends, therapists, coaches, mentors, etc. A lot of what goes on in a coaching or therapy session is that we our nudged into figuring things out for ourselves.
My sense is that the new AI’s are already surprisingly good at playing the role of mentor or coach. People are already using them that way. I imagine that they could get even better. For example, an AI mentor or coach could be trained to imitate a successful role model. Instead of going to see an average therapist or coach, you could have a session with one of the absolute best.
Does a young man benefit from sitting in an audience of hundreds watching Jordan Peterson? What if he could have a one-on-one consult with a Peterson=trained AI?
As an investor, do you see yourself following the advice of Warren Buffet? What if you could have Buffet’s opinion about how you should be selecting a under your circumstances?
My simple essay grader was a convincing proof-of-concept for me. If I were a high school teacher or college professor, I would be happy to use an AI to grade student papers. That would address one of the most tedious, time-consuming parts of teaching.
But more generally, I think that one-on-one relationships between AI’s and humans are going to be important going forward. Everyone can benefit from friendly coaching.
Animation
This one I am personally unlikely to get involved with, either as a producer or a consumer. But it seems to me that to the extent that people want to create and consume entertainment in the form of short videos, the new AI could make production much easier.
Instead of having to set up and film a scene, a creator can simply tell an AI to create the scene in an animated form. Depending on how the relevant law develops, your videos could include characters who very closely resemble actual people.
If augmented reality and virtual reality hardware improves, this could get quite extreme. More and more of our experiences could be fictional, created by other people. The philosophers’ brain in a vat might become more than just a thought-experiment, with the virtual world taking precedence over the real one.
I do not guarantee that robotics, mentoring, and animation will all be enabled by ChatGPT and its relatives. But I do guarantee that if the only use you can imagine for AI is as a search engine, you are not thinking hard enough.
But I don’t think an AI mentor will fulfill our desire for mutual respect, or as you say in Three Languages, high self-regard; not even if carefully programmed. What will happen to those unaware of this? Will they become ill seeking respect from an AI?
“As Adam Smith pointed out, we have a desire for high self-regard. In part, we want to be recognized by others as being admirable. Moreover, each of us has what Smith called an ‘impartial spectator,’ or conscience, which makes us feel happier when we believe that we are acting in a way that others will regard highly. Following group norms is a way to please the impartial spectator.” Pg 45, The Three Languages of Politics
“I would be happy to use an AI to grade student papers.”
Today’s teachers would have to develop a grading script that is entirely different than yours. Does this essay compellingly relate the author’s sense of victimhood? Does it use the proper opaque woke jargon? Does it reflect the urgent need to tear down the existing system and create a new future that is free of constraints? Does it dismiss all dissent with ringing declarations and emotive slogans that are devoid of substance?