Comment: Chimps can do something that AI can't: reason
By F.D. Flam / Bloomberg Opinion
For something so admired, so synonymous with merit, the concept of intelligence is remarkably poorly understood. Our society operates on the assumption that people with greater intelligence deserve access to better schools and better jobs. Many people believe that animals with higher intelligence deserve to be treated more humanely, or at least not used for food. Our tech leaders obsess over comparing human intelligence with the latest AI systems. Many claim that once these systems surpass us in intelligence, they will quickly enslave us, destroy us, or solve all our problems.
How can we compare human and machine intelligence when we can’t decide which species — cats or dogs — is more intelligent? Scientists who study human or animal behavior and brainpower tend to view intelligence differently, breaking it down into abilities they can actually measure. What impressed them was a recent paper providing experimental evidence that chimpanzees can use reasoning to weigh different strengths of evidence, draw rational conclusions, update beliefs and recognize the strengths and weaknesses of their own knowledge.
-snip-
Many animals update their beliefs through learning, said Gopnik, but this experiment demonstrated something new; the animals were not just learning but weighing evidence, both new and remembered. And they combined all that information to decide whether they had good reason to change their minds. That requires what Gopnik calls metacognition, the ability to evaluate what they know and what they don’t.
-snip-
Coincidentally, this month, Science reported that ChatGPT and similar systems are programmed to lack strong metacognition. As many users have noticed, these systems often give confidently incorrect answers rather than admitting they don’t know. The training, as Science reported, is designed to “reward confident guesses and penalize honest uncertainty.”
https://www.heraldnet.com/opinion/comment-chimps-can-do-something-that-ai-cant-reason/
= new reply since forum marked as read
