If you were to ask a computer what the square root of 1346086 is, it would probably give to you the correct answer (1160.2094…) in about a fraction of a second. However, if you were to ask that same computer to, say, place blocks in a tower it would fail miserably. So why is it that computers can perform tasks that are incredibly difficult to us humans yet are unable to do things that we find seemingly easy? It was this question that came to be referred to as Moravec’s paradox, named after the Austrian Roboticist Hans Moravec who first formulated it. To understand why Moravec's paradox occurs, we first need to look at how computers learn things.
 
In the early 1950s, there was a theory amongst scientists that the nature of intelligence could be boiled down to nothing more than symbol manipulation, that all human thought was simply manipulating symbols in our mind and if we could automate this process, we could theoretically create a machine with the intelligence of a human. This theory became the foundation for the first artificial intelligence and, as it involved manipulating symbols, it came to be known (rather unimaginatively) as symbolic AI.
 
With the advent of the digital computer, a machine that was able to perform said symbolic manipulations, there was no better time than to start seriously working toward creating artificial intelligence. Soon computer programs were solving word problems, proving theorems in geometry, and even learning how to play checkers. Optimism rose and the money poured in, with researchers claiming that by as early as the 1990s we would have solved the problem of artificial intelligence.
 
It's understandable to think that because scientists had solved the so called ‘hard’ problems, that the ‘easy’ problems, like placing blocks into a tower, would just fall into place. Unfortunately, this is not what happened. Researchers slowly realised that the tasks we perform everyday are incredibly complex. For example, the act of looking at an object and classifying it is something that scientists have only really been able to accomplish in the last two decades. So what could be the reason that humans are able to perform such tasks with ease?
 
One explanation for Moravec’s paradox, offered by Moravec himself, is that humans are so skilled at certain actions due to the way we have evolved. Moravec suggests that all human skills are made possible using biological ‘machinery’ which has been designed by the incredibly powerful process of natural selection. Over the course of time, natural selection has tended to preserve design improvements and optimisations resulting in our ‘machinery’ being incredibly well-suited for tasks we commonly perform, such as motor control. This means that the longer a skill has been around, the more effortless it should seem to us; hence why abstract thought is seen as so difficult, whilst walking is seen as axiomatically trivial. Some examples of skills that have been evolving for millions of years include recognising faces, moving around, judging people's motivations, catching objects, recognising a particular voice, and so on. Some examples of skills that have appeared more recently are: mathematics, logic, and scientific reasoning. This argument can be condensed into the following:
 

1. We should expect the difficulty of reverse-engineering any humans skill to be roughly proportional to the amount of time that skill has been evolving in animals.
2. The oldest human skills are largely unconscious and so appear to be effortless.
3. Therefore, skills that appear effortless are likely to be difficult to reverse-engineer and skills that appear difficult may not necessarily be hard to reverse-engineer.

 
Overall, Moravec’s paradox shows us that whilst computers can perform certain tasks that we would find normally difficult, they struggle with tasks that we find easy. This is a result of the way that computers are constructed, being able to perform given instructions with exact accuracy but not being able to do anything else. On the other hand, it is our human ability to interpret and predict that has meant that we are able to perform such tasks with ease. It has also meant that we are unable to carry out instructions with absolute accuracy.