Infinite prep
I think AI's greatest disruption in fields historically reticent to change will be via enabling infinite prep, rather than via automating human decisions. "Infinite prep" is a term that's stuck with me from high school debate. In its original form, it means something like "having infinite time to prepare the initial argument," but over time, I've come to interpret it as "seeking out and exploiting every edge in an activity." Once infinite prep is introduced to an activity, the 'casual' times are over, and the game will become professionalized. I’ve written about competition much more extensively before, and think that piece holds up very very well.
Examples of infinite prep are everywhere. Elite video game players train their mechanics and reflexes on simpler but mechanically intense games such as Quake Live and Osu. Policy debaters write as many evidence cards as possible to support their arguments, and find obscure arguments that are harder to argue against. Elite cyclists clock in very heavy volume, e.g. 20 hours on the bike per week, while elite runners use altitude training to challenge their lungs and maximize their training input.
That being said, infinite prep can be more challenging to implement effectively, especially in fields that require high-intensity human interaction, such as negotiation and conflict resolution. While these types of fields certainly require skill and preparation, they also rely heavily on intuition, emotional intelligence, and the ability to read and respond to other people's behavior. These skills can be difficult to replicate in a purely scripted or rehearsed setting and require ongoing practice.
The end of the anthropocene?
A recent study showed that Go players have improved at a faster rate and become less predictable, since the introduction of AI-based Go bots. These concurrent effects are intriguing -- not only do you see the first order effect, that people got better faster, but you see the second order effect, that people started playing more creative strategies. I believe that we'll see the same effects when AI is applied to other industries too.
In fields such as programming, AI is not yet an elite practitioner, unlike in Go or Chess. Even then, I think the AIs are very useful to learn from. A novice in a field can benefit from being tutored by a B-minus AI, getting up to that level in the first place. I wrote about this effect more extensively before, with respect to how I've been learning TypeScript and front end development.
Practicing programming can be pretty hard. People have said, I don't know even how to get started, outside of practicing LeetCode questions. I think LeetCode is quite popular because it gives instant and accurate feedback, even if it is just limited to if your code passed the test cases. You can easily imagine an "explain my answer" extension to the platform that significantly improves the learning process, a la Duolingo Max.
What's interesting to me is also where we go from just LeetCode problems. Could we generate more interactive prompts / problems and offer similar levels of feedback? For example, when I taught statistics, many of the common statistical datasets are biased towards male interests (eg cars and sports statistics are very common intro examples), and I tried to find more gender neutral datasets (eg babynames). Could we instead generate a problem set or course lecture using a user provided dataset or domain? Most of the core concepts remain the same, and perhaps still written by the professor, but with examples or code samples personalized to the user and their data.
Similarly, LLMs can help emulate human interaction, and allow for practice via repeated simulations. For example, the GPT-4 Turkish Carpet Merchant allows users to practice negotiating a purchase price for a rug. In a similar vein is Negotiation AI, which provides additional negotiation scenarios and specific tactics to tutor. I wouldn't say GPT-4 is world-class at negotiation (the carpet merchant, in particular, is subsceptible to 'grandma attacks'), but it can help people genuinely practice.
In a field like law, AI can support lots of document-related tasks, such as drafting, search, and preparation, but ultimately most value in law comes from human-to-human competition. You find what you think is a loophole, they disagree, you go back and forth. AI can help lawyers by generating arguments and counterarguments to try and walk through different scenarios, and better prepare for the courtroom.
Similarly, doctors and nurses can utilize LLM's in navigating tricky conversations with patients. While a script generated by AI may lack emotion and engagement, employing AI to guide healthcare professionals in understanding possible patient responses and emotions can be highly valuable. Additionally, LLM's could provide scenarios where the patient is lying or leaving out important diagnostic information and train medical staff to find the pertinent info more quickly. You could train it on old episodes of House!
Even if fields such as law or medicine were to restrict AI making final decisions, or even making suggestions, I still think AI will have a profound impact on how we study and prepare for activities. Overall, I think this is a positive development, that we are going to get better and better at, well, everything.
Alas, there is always a tradeoff. As people become ever more specialized and excel in their own fields, it becomes harder to tell if somebody in a different field is good. It also becomes harder to become the best, which can be an alienating experience. My hope is that more people recognize an option to opt-out of the rat race, and that we can create a less inequal society, free of some of these pressures.