On the Irony of Expecting AI to Reason Better than Humans
Reasoning, that precious jewel in the crown of human cognition, has always been more of a cheap imitation than the flawless gem we like to imagine. We put reasoning on a pedestal, telling ourselves that our decisions are born from logic and reflection. Let’s face it: in reality, we humans are a messy brew of prejudices, emotional impulses, and wishful thinking. And now we dare to demand that AI “think” logically as if it were the flawless heir to something we have never truly mastered.
Large language models are often criticized for not mastering logical thinking. Do we master it?
David Hume once posited that ‘Reason is, and ought only to be the slave of the passions,’ acknowledging our strong emotional underpinnings. It’s ironic that we now expect LLMs to embody a platonic ideal of rationality when the human mind often operates on the basis of fallacies and assumptions.
We have established logic rules — syllogisms, formal inference systems, and the like. But like the rules of a game, they can be bent when emotions or convenience come into play. We should pause and reflect: how many of our decisions today were guided by a strict chain of irrefutable arguments?
Was it the systematic, deductive process of Aristotle?
Or was it a casual mental shrug supported by confirmation bias and selective memory?
And yet, we expect AI to untangle the mess of our inputs and respond with divine clarity ;-)
By asking LLMs to reason better, we are outsourcing a job we were never qualified to do. It’s like blaming the calculator for a miscalculation, forgetting that we entered the wrong numbers. Perhaps we hope that by expecting flawless reasoning from it, AI will fulfill our promise to human rationality — a promise we have never kept.
We ask machines to reason without flaw, forgetting that we are masters of tangled logic.
The real irony is that we are disappointed when it doesn’t turn out perfectly, conveniently ignoring the fact that our thoughts weren’t either.
Murat
This might be of interest. I created a podcast with NotebookLM based on my book Beyond the Algorithm. The result is quite impressive.