Skip to main content

Moravec's paradox

In the 80's some radical artificial intelligence and robotics researchers came up with a controversial idea: what's easy to us is actually harder for the machine than those things that are hard to us. In fact, the early AI research already displayed prominent capabilities in areas of complex logic uer even today the most fine motorized robots are no better at moving than toddlers.

We as humans are a product of millions of years of R&D. Our physical skills are so finetuned that we hardly even need to think about them. Walking on just two feet is actually really, really hard. But we do it without a thought. Best bipedal robots are still far from the grace and elegance of the human body. We are also good at detecting patterns, emotions and paying attention. Another things that have been challenging to "teach" to the machines.

Complex logic problems are one example of problems where we are not that good. Sure, with the help of tools we have been able to push our limit further. But still, even the simplest modern computer can easily outperform us with such challenges. While our motoric and attentive skills have had all those millions of years time to evolve the higher level "intelligent" functions are quite new to us. Even our language skills are mere 200 000 years old, barely out of beta. Higher logic, mathematics and such have existed only for a few thousand years. We are just getting started at those, so it's also still easy to turn those skills into "thinking" programs.