Posts tagged with "iq"

After listening to the EconTalk podcast episode Nature, Nurture, and Identical Twins (with David Bessis), I read the essay David Bessis wrote, Twins reared apart do not exist, which was the subject of the episode. Both were informative and helped me understand the conversation between hereditarians and blank‑slatists. But something else caught my attention in the essay.

Bessis begins with three illustrations of potential values for the heritability of IQ, 30%, 50%, and 80%. He maps genetic potential against actual IQ for each percentage. In this post I’m not addressing questions about IQ, its measurement, or use. What sparked my interest is the distinction between potential and actual, and whether that distinction adds to the conversation on what differentiates human thinking from AI thinking.

It seems to me that genetics does set limits on the range of capacity each human is able to achieve in thinking, but it is not a predetermined number. Other factors have significant influence on where each human ends up in this range. Therefore, there is some capacity the individual human can achieve, but there is also some measured value designating where they currently are.

This leads me to believe that there are many factors that influence how the brain thinks. It’s not about crunching data about a question and ending up at a result. Instead there’s a lot of seemingly unrelated data accumulated over a lifetime of experiences which mingle in the brain, impacting pathways on the way to producing the thought.

How does AI compare when mapping genetic potential and actual IQ to machines? Is it accurate to say that their training is their genetic potential and then reinforcement learning is their actual potential? I don’t think it is. In humans their actual potential continues to be shaped from their interactions in the world. The experiences humans have influence us. For AI, once the model is released it is fixed. Additional information can be provided to them which influences their generated responses, but their intelligence is locked-in; their genetic potential is still their actual intelligence.

In the end an AI’s intelligence is hereditary rather than blank‑slate, and that leads to a very different form of thinking. AI thinking is not human thinking.