Scarcity and Machines

All,

AI is being adopted by P123 so perhaps it is a topic for the forum. Here is an article in the Journal of American Medical Assocition specifically about machine learing in medicine: Proactive vs Reactive Machine Learning in Health Care

Given my history you may think I will advocate for AI in this post. To put it simply, not really. In fact AI scares me a little. Okay I am lying–it scares me a lot.

First, something from the article that may be useful to investing: They site reinforcement learning as one of the most successful methods. I have discussed the explore/exploit tradeoff (Thompson sampling) in another thread: Algorithm for Finding the Best Trading Strategy Quickly. This algorithm (or one like it) is being discussed as “successful” in he JAMA article.

But the more general point in the article and the one that is (potentially) most troubling is: "Their reinforcement-learning system (known as Eva) allocated scarce…"

So, here is the thing: Eva is making medical decisions. There is no other way to look at it. In this example a diagnostics decision. And there is “scarcity.” But AI is, without a doubt, also making treatment decisions in the presence of scarcity.

Sticking with something that I know about, there is scarcity in medicine everywhere now. With Covid, there was scarcity of masks. There was scarcity of monoclonal antibodies and now a scarcity of Paxlovid (an oral medication for Covid). Scarcity of ventilators, ICU beds, testing kits…

Arguably, at least at times, government is helping with the scarcity. Government is purchasing Paxlovid, for example. It could be argued that government is trying to increases the supply.

Also it is true that while the scarcity exists, the resources are being rationed.

The truth is everything in medicine is now rationed–requiring prior authorization through the insurance companies. Generally done through a computer.

I won’t continue to make the case about medicine here. But I think this is a general trend toward the rationing of scarce resources that is affecting our economy in general.

For example, the rolling blackouts that can occur in California are a type or rationing. More and more, AI will be in charge of managing the scarcity (and rationing). Again, it is already here in medicine.

I read an article that said Facebook will do well going forward, basically because any (every) new startup will have to advertise through one of the large tech companies. Facebook claims to have the “fastest AI supercomputer in the world:” Introducing the AI Research SuperCluster — Meta’s cutting-edge AI supercomputer for AI research

Facebook (Meta if you prefer) has a computer with, a mind-boggling, 22,000 GPUs. All to support using reinforcement learning (and AI in general) for decisions about presenting ads to you.

All I am adding here is that AI will not be employed just for advertising in the future. AI will be used to manage the “scarcity.”

One can probably invest in that. And maybe use a little reinforcement learning yourself. But it will certainly continue to have an effect on the economy, or at least how the economy is managed. It has already happened with your healthcare which is not a small portion of the economy.

[color=firebrick]If this is all too much just call your Robo Advisor for guidance. “I promise by the time you’re done…, you’ll feel right as rain.”[/color] (from the Matrix)

Best,

Jim

Jim,

A very good post on the subject on AI.

As you have pointed out, we need to use some methods (like reinforcement learning) to address the increased chance of overfitting with machine learning in P123.

By the way, I have been following a list of the fastest supercomputers in the world.

Here is the link :

https://www.top500.org/lists/top500/2021/11/

Regards
James

All,

The JAMA article refers to this clinical study as an example of something that is not an entirely bad idea IMHO: Randomized, Embedded, Multifactorial Adaptive Platform Trial for Community- Acquired Pneumonia (REMAP-CAP)

It is only in the context of government rationing (where politics might play a roll in a Biden, Trump or potentially worse future administration) and the physician’s judgement is being overridden for economic reasons alone that I become concerned.

I do believe in (and use) some of these techniques for investing. As you can imagine would be the case for Covid and pneumonia, some of these techniques are finding favor when one cannot make use of a normality assumption (or prefer a method not requiring any distribution assumptions at all) and the situation is constantly changing. The jargon can be simplified for those wanting to use this for practical investing.

BTW, with all the power it has, Meta’s supercomputer has run 3 weeks when training some data while I shutdown my Mac if it has not finished over the weekend. Institutions have resources that one cannot hope compete with (where direct competition or arbitrage opportunities exist).

Best,

Jim