On Sun, Sep 20, 2026, 5:14 PM Linas Vepstas <[email protected]> wrote:
Instead of tackling free will head-on, I'm asking "what else is there that > co-occurs with it?" and "can we make progress on one of these other > co-occuring factors?" > Yes. We tend to think of randomness, but it's not that. No program can predict its own output. No program can run a copy of itself to predict what it will do next and then do the opposite. Not you and not a thermostat. As I said, the source of the illusion of free will comes from internal positive reinforcement of actions. This signal evolved to motivate you not to die, which increases your likelihood of passing on your DNA. The alternative is to feel like your actions don't matter. Optimal reinforcement learning in the AIXI model is not computable. That strategy would be to use Solomonoff induction to model the environment by finding the shortest program consistent with past input, evaluate the result of possible actions, and choose the one that maximizes reward. However, the first step of finding the shortest program that produces some string is not computable. If it were, you could write a program of length n bits that searches for the first string that cannot be written by any program of n bits or less. So the brain uses an approximation of rational reinforcement learning. It makes you more likely to repeat actions that were rewarded in the past. This is usually good enough but not always. To illustrate the difference, your choice to take or not take a drug should rationally not depend on whether you have already tried it. So what happens if you reward all actions regardless of what choice you make? https://pubmed.ncbi.nlm.nih.gov/17958712/ In this experiment, children or monkeys were offered a choice of 2 different color but identical tasting M&Ms. Then they were offered a choice between the unchosen color and a third color. They found that the rejected color was more likely to be rejected again. The reinforcement signal detected in this experiment is the source of your illusion of free will. -- Matt Mahoney, [email protected] ------------------------------------------ Artificial General Intelligence List: AGI Permalink: https://agi.topicbox.com/groups/agi/T9e51fc70bbc4e9fc-Md40d0c66082e8df3bcea0c06 Delivery options: https://agi.topicbox.com/groups/agi/subscription
