Why AI Can't Replace the Human Thought Process: The Essential Feedback Loop

AI tools will progress very fast, it is impressive how much they have evolved in the past few years. First slowly 1 then all of a sudden, making LLMs mainstream around 2022. Brains will not evolve that fast. So as much as we would like for AI to cover up for the gap brain to machine, this is something that will not happen during your lifespan – not even by the time your children grow old –.

There’s a fascinating take that Larsen presents in this post: “Imagine how cool it would be if you could imagine something and a computer would do that for you”. He is implying it would be amazing, as I’m sure most people think in this AI driven times. But there is a gotcha: brains don’t work that way.

While your brain is thinking, it’s doing so in real time there is a feedback loop where you’re not just thinking per se you are acting based on those thoughts and instructions your brain is sending to your muscles. In parallel you are receiving input from your actions, processing it, and pivoting your thoughts in different directions based on your observation.

Driving is a clear example of this process2, but it is far from the only process where your brain operates like that. If you sit down and attempt to write a Novel, your brain acts in that exact fashion. How fast you can right it and the quality of the output depends both on your writing speed and your thinking speed. The exact same pattern applies to reading. Where fast readers are fast thinkers, people who can quickly imagine what they’re skimming.

If your brain is the limit to how well you can write, or read, where does that leave AI workflows that aim to replace the process in between? From thought to paper, from thought to project, from thought to action.

Instructing an LLM to perform an action does indeed fast forward a good part of these activities, but it bypasses the most important step for your brain: the feedback loop. Experimenting, seeing how your actions unfurl in the real world, how they look. Appreciating the shades in the colors of your design. Thinking through that scene you just wrote for your script. Typing out the if-else workflow in your function. All of those small actions are irreplaceable for your brain. All of those bits are what makes up your thought process. Once you move to dumping your very first clouded thought into an LLM, you will get an equally blurred outcome. There is a lot of work involved in distilling a thought in order to end up with a pure idea. And this is something that only the human brain can do for now.

https://x.com/gregpr07/status/1985825522035949588Larsen and Gregor talk about an idea

Footnotes

  1. First mathematical models dating back to the 1950s with the first multi-layer networks developed in 1986 ↩

  2. In fact driving as a computational problem is almost fully solved with FSD ®. ↩