I started a YouTube channel to analyze AI developments from a general tech perspective. This article is based on one of my recent videos, where I took a deep dive into Moravec's Paradox—the frequently cited adage that tasks difficult for humans are easy for AI, and vice versa.
Here is what I found:
Moravec's Paradox has never been empirically tested. (Many AI researchers, including pioneers I know and respect, repeat it as fact, but that doesn't mean I'll take it at face value!)
It is actually just a description of what the AI community deemed "worth doing." It does not predict which problems will be easy or hard for AI.
It comes with an evolutionary explanation that I find highly suspect. (There is a long history of AI researchers making up stories about the human brain without any background in neuroscience or evolutionary biology.)
Moravec's Paradox-style thinking has led to both alarmism (about imminent superintelligent reasoning) and false comfort (in areas like robotics).
To adapt to AI progress, we don't need to predict breakthroughs. Because the diffusion of new capabilities takes a long time, it gives us plenty of time to react—which we often waste, only to panic later!
You can watch the full argument here, or read on below.
Every week brings new claims about AI progress. How do we know what's coming next? Will AI predict crime? Write award-winning novels? Hack critical infrastructure? Will we finally have robots in our homes folding laundry and loading the dishwasher?
What will AI progress mean for your job? What will it mean for the fabric of society? Dealing with these uncertainties is overwhelming. If only we had a way to predict which new AI capabilities would be developed soon, and which would remain difficult for the foreseeable future.
Historically, AI researchers' predictions about the progress of AI capabilities have been pretty terrible. We don't really have any principles to explain which tasks are easy for AI and which are hard.
Except, we do have one—Moravec's Paradox. It refers to the observation that it is relatively easy to make computers exhibit adult-level performance on tasks humans find difficult, like math and logic, but difficult to give them the skills of a one-year-old when it comes to things we find effortless, like seeing the world or walking.
This concept comes from Hans Moravec's 1988 book Mind Children. Moravec was, and still is, a robotics researcher. He wrote:
it is comparatively easy to make computers exhibit adult level performance on intelligence tests or playing checkers, and difficult or impossible to give them the skills of a one-year-old when it comes to perception and mobility.
In the early days of AI, researchers focused on chess and other reasoning tasks because they were thought to be among the hardest tasks and a reflection of human uniqueness. Interestingly, if you want to build a robot that can beat a human grandmaster, deciding which move to make is actually the easy part; the hard part is physically moving the piece on the board. Today, this is widely known, so Moravec's Paradox seems highly intuitive.
If Moravec's Paradox were true, its implications would be staggering. If we wanted to know which AI capabilities might emerge next, we would only need to look at how difficult they are for humans. Consequently, scientific research would be automated before folding laundry, and so on.
But here's the catch—Moravec's Paradox has never been fact-checked. Even with videos getting hundreds of thousands of views and TED talks repeating it as fact, that hasn't changed. When I dug into the evidence behind this so-called paradox, I found something surprising.
In this article, I will discuss why the theory and evidence behind this paradox do not hold up. Next, I will explain why simplistic predictions about what is easy or hard for AI have misled AI researchers and tech leaders. It has led to alarmism on one hand, and false comfort on the other. (Which is a paradox in itself.) Finally, I will answer the question: if we cannot rely on Moravec's Paradox, how should we prepare for AI progress and its impact?
The Evidence Behind the Paradox is Unreliable
How would we test Moravec's Paradox? We could take a sample of real-world tasks, determine how difficult they are for humans and how difficult they are for AI, and plot them on a graph. If we get a result like this, the paradox would be confirmed.