Rumination

Grounding in the real world is an essential trait for a competent person in any profession. And yet, we superimpose the world with our own existing algorithms which have sometimes gone haywire.

Rumination is like a large language model gone amok. Do you want to find the truth? Do you find the truth? Just look at the evidence, look at the real world, look at what is happening. Ground your experiments in the real world. Do not keep generating ideas in your brain that do not have a basis in reality. I realize how CBT-coded this reads, but sometimes the most obvious things are hiding in plain site.

Brains (and language models) are constantly lying to people. The real question perhaps will never be to get a better ‘language generator’ or an ‘idea generator’ but to develop neuro-symbolic algorithms, that take the generation, quickly test it out in the real world, learn a lession, and then are able to generate some truths about the world. This has to be the loop. Of course, this implies the existence of continual learning, we will have to come up with proxies like using memory and such; maybe there are other ways of simulating (or even inventing) continual learning.

However, for humans, the situation so far is a bit better, we are able to do continual learning. We have no excuse to not be interating with the real world.

So, what are the best kinds of interactions for human subjects?

1. Interacting with information-rich real world mechanisms.
2. Interacting with predictable real world mechanisms.
3. Interacting with mechanisms that are healthy for the humans.
4. Mechanisms that maximize learning for the humans.

This implies for example, reading a lot of good high quality research papers and books, as opposed to getting the learning form YouTube shorts. This involves doing experiments, figuring things out for oneself, keeping an open mind, and doing it frequently and every day. What real-world interaction does for humans is that it refines our internal mental models (that i now refer to as mental turing machines, or mental algorithms).

This is the real model that separates the experts from the novices. Of course, everyone interacts with the real world, but what kind of interaction?

A non-good interaction: a human already has a pre-conceived notion of the kind of shots to play in a game of tennis, they do not update their beliefs based on the evidence at hand. They had once interacted with the game, learnt a few algorithms and then use it ad nauseum.

A good interaction: a human is playing tennis, they realize the opponent is stronger than them, they start observing and asking what is it that the opponent is doing that they aren’t? Can they quickly replicate it?