If you try to hit a mosquito with your hand, it will fly away.
This means that even a mosquito knows what a hand is, what damage it can do (crush), that getting crushed is a life hazard and that life is precious that doesn’t return once lost (and is worth saving).
A mosquito understands what your hand is, what getting crushed is, what a hazard is, what life is, what loss (of life) is.
In other words, a mosquito got the correct definitions of all above.
Without having correct definition(s), one cannot correctly identify or understand what something is.
Let me give you an example.
Suppose you’ve never heard about apples and that I’m trying to teach you to recognize an apple when you see one.
If I gave you, say, this defition of an apple: “an apple is something round and red.” Then you’ll also consider a red ball an apple — which is wrong.
Suppose that I then improve upon my definition and say “an apple is a round & red object that is edible”.
Then you’ll reject a yellow apple and consider a tomato an apple — both of which are mistakes.
Hence, my definition is not yet concise & correct.
Suppose I then say “an apple is a fruit that is usually round and tastes like apple”, then this definition will not only exclude all objects that look like an apple but do not taste like an apple, but it will also correctly identify, say, a genetically-engineered black apple.
Even if you eat a dish or drink a beverage made of apple, you’ll still know that it contains apple.
Let me clarify that “tastes like apple” does not constitute a recursive (circular) definition because the apple’s flavor is not an apple but it’s a chemical that can be completely defined by a chemist independent of apples.
Assuming that the last definition was perfect, you’ll now correctly identify an apple when you see one, even when it’s altered.
The mosquito had correct definitions when it decided to escape your attempt to crush it.
Now let us see whether a computer can “see” anything.
Suppose that we use a camcorder to record a session between a patient and a physician.
Now if a human is shown that recording, she’ll correctly understand what’s going on. She will also identify doctor’s clinic, the doctor, the patient, any objects on doctor’s table, the chairs, the walls and any other object known by common people.
So it’s safe to assume that there was nothing wrong with the recording.
But what if we show this recording to a computer? Will the computer be able to understand anything? Will it be able to identify any objects?
No. The computer can record the video and play it back but it will not understand anything — and it’s terribly difficult to teach it to identify simplest objects.
And it’s identification performance will always remain awful. Compare that to a tiny mosquito’s!
Another serious issue with the computers is that they can’t learn about new objects.
Many domestic animals can identify common household objects and they can usually also understand many abstract “objects” such as “death”, “food”, “injury”, “love” and even “work”.
If you’re, say, a software developer and got a cat, then the cat will initially not understand why you’re wasting your time on a computer — it’s not food.
But after a while, the cat will understand that your computer is actually a tree whose fruit (money) you use to provide for your food (as well as cat’s).
1. The ability to “see” things is dependent on our ability to construct definitions.
2. To achieve AI (Artificial Intelligence,) we need definitions, not a neural net to compare (a stored image) to another (that we expect the computer to be able to identify).
3. Even a mosquito with less than a milligram of gray matter knows what is “food” and thus, can find it.
It’s not because its brain is more powerful than a supercomputer (or even an ordinary smartphone’s processor). It’s because it uses different algorithms.
4. A computer equipped with a camcorder is capable of vision and hearing — provided we let it eat from the “tree” (algorithms) of vision & hearing.
5. The computer lacks no intelligence. Intelligence is the number of questions one can solve in a given time. An ordinary person can solve 100 questions/hour. A computer can easily solve millions and billions.
6. We first need to solve the problem of computer vision and hearing. In other words, we first need to build an electronic eye and an electronic ear.
And for that, we’ll also need to build an electronic brain and a database so that the computer will then learn to see and hear.