Most humans are not good at chess. They can't write symphonies. They don't read novels. They aren't good athletes. They aren't good at logical reasoning. Most of us just get up. Go to work in a factory or farm or something. Follow simple instructions. Have a beer and go to sleep.
What are some things that a clever robot can't do that a stupid human can?
The "baseline humans" you describe have been historically described in the media industry as "the lowest common denominator" (LCD).
The LCD is the broadest possible audience for content, traditionally for network television shows. (Before the age of cable, there were only 3 to 4 networks and all video content was broadcast over the airwaves--no way to specifically target audience segments so content had to appeal to the LCD.)
Because captchas have to be solvable by the LCD, but trick bots. As long as captcha are viable, by definition they will always be something that baseline humans can do better than AI.
First Question
To treat this question in a scientific way, because I think it is a reasonable enough question that draws on the realities of postmodern culture in post industrialized societies to be treated scientifically, we should define some things.
The most difficult is intelligence, which is the realm in which smartness, cleverness, and stupidity reside.
Let's go through the list.
The question is fine until it devolves into the dismissal of the intelligence that people apply to their method of earning income, which for many if not most people requires more than just getting up, commuting in, following some simple instructions, and going to sleep mildly inebriated. Let's replace this last part with this.
If we define intelligence as the union of these things, for simplicity's sake, we have this (which is subject to change as AI develops).
We must, to interpret the above list correctly concede two things:
Second Question
What are some things that a clever robot can't do that a stupid human can? These are a few, but they are of particular importance from certain perspectives.
I would not dismiss these human propensities as irrelevant, even from a scientific perspective. I would also not dismiss the possibility that these things are beyond the capabilities of silicon based entities.
Some fields that humans are born with advantages:
Fast and precise image processing ability. Even the stupidest human can tell the edge of two different objects precisely, e.g. which part of the image is a dog and which is a cat.
Fuzzy learning ability. Humans don't need to see all kinds of cats to identify a cat. As long as we see some cats (real ones or pictures or videos) we can identify a cat easily.
Reasoning. Current machine learning methods are mostly statistics-based high-dimensional model approximation. Instead of finding a solution or a pattern, I have never seen any AI entity can generate new ideas based on current facts.
Abstraction. Now GANs and other AI techniques can create vivid drawings. Yet currently I cannot find any model that can do abstraction drawings. E.g. human can doodle a cat from a real pic of cats, while AI currently can't do that.
There are more of these kinds that human born with skills in their genes because of millions of years of evolution. While I believe in the future we'll have better AI entities with better algos to defeat the advantage of humans eventually.
The thing that comes to mind is a new-born, when you said "the stupidest human", and it already has some basic “survival instincts”. It will avoid pain, consume food, and quickly learn to distinguish "safe" and "dangerous" conditions and people.
We have computer programs that can learn chess and calculate the optimal move in a split second, but isn't playing chess is a bit pointless. Merely being able to play a board game is of little value from a survival perspective, industrial perspective, or economic perspective.
There are programs that can do things that are very helpful for the modern world, but as far as I know, they just don't have survival instincts. A self-learning robot; left in a forest with all the tools it needs to generate power for, build duplicates of, maintain and defend its self; probably wouldn't be able to learn how to do so in time to ensure its survival. Our current self learning programs would need to be able to identify when it has succeeded or failed to improve its survival odds. A child of two may learn fast enough to survive if the conditions are not too severe and non-toxic food and some form of shelter is nearby.
A financially poor, marginally educated person with lower than average aptitude working at a farm or factory might not be able to play chess well, but they would definitely be able to tell if someone is murdering someone else, and know to flee and seek the authorities. A robot that can play chess would not.
Furthermore, humans can continue to learn when separated from the problem by thinking about the problem. The ability to construct arbitrary models and run thought experiments is currently unique to humans.
That said, I do hope that we will soon have programs that well replicate the human mind, and demonstrate some of the aspects of what we call consciousness.
I do not know the precise definition of intelligence, but from lots of people I have interacted with, they regard people as intelligent on a particular field, if and only if:
split second correct decisions in a situation in that particular field.Let us see where AI have succeeded in this case:
These are the few famous cases. If we examine carefully these cases we see that computers are outperforming humans only due to:
So AI's are actually workhorses, working without fatigue and without any limitations. Human brains do not excel in the field of decision making or speed. Here is a comparison of What makes animal brain so special?
Human brains excel at creativity. We can learn how to make symphonies. Can an AI do the same? Possibly with correct programming. Much of our intelligence comes from its distributed nature. We learn from other peoples mistakes, we improve it. Large number of humans combined with record keeping has made this possible. Although scientists like Tesla, Einstein, Newton, Feynman discovered Calculus on their own, think of the possibilities of new inventions had they been made aware that Calculus already exited and a lot has been done to develop it? Check this: Swarm intelligence vs Normal Human Intelligence.
So our intelligence and experience comes from the huge source of information rather than huge source of personal resources. As of now we can think of abstract concepts which an AI cannot (i.e. we can create new things, not new artworks or music by mixing things up as an AI does, but a new thing completely).
For example, it has been seen if you keep many deaf babies together and isolated they develop their own form of sign language, completely unique. Points to note here are:
So although machines might be performing well due to their algorithmic complexity and immense power they still have some catching up to do to be compared to even stupidest humans.
Main problem is we do not yet know the capacity of a brain. Some people can perform exceptional feats with their brain when the need arises. Some one did this during WW2 to find his family: Grandmaster plays 48 games at once, blindfolded while riding exercise bike. But how is this suddenly possible? No one knows until we have uncovered our own mind fully.