What Are the Limits of Artificial Intelligence?

At the very start of this discussion, it can be said that when a number of data and factors come together AI (Artificial Intelligence) is obviously much better for human intelligence. However, it is an established fact only humans are capable of thinking in a logical manner and can thus distinguish between which AI advice is useful and which one is worthless.

Limited and unlimited

The thing with AI is that it is limited and unlimited at the same time. It is unlimited simply because it is immensely superior to the physical learning capabilities of the human brain. The powerful computers are capable of performing a greater number of operations within small fractions of a second.

How is it superior?

This means that machine thinking provides people patterns that they are either unable to recognize or recognize after the passage of an amount of time that is too long to be accepted. However, such thinking is limited because a computer is only capable of detecting patterns. Only humans can recognize the logic and sense behind it.

Not really intelligent but highly fast

Irrespective of all this it can be said that such patterns can lead to insights that thinking, which is solely logical in nature, is unable to achieve.

This is because of the amount of data involved in the process as well as the complexity of the same. This is actually the quality of AI that marks it out from the previous forms of digitization. The best illustration of this difference would be the supercomputer Deep Blue built by IBM (International Business Machines Corporation) in 1996. This giant computer beat the then world champion in chess Garri Kasparov. However, it needs to be pointed out in this context that Deep Blue was not really intelligent. It was just so fast that it was incredible.

How did it function?

It only calculated the probabilities for the next few moves on the basis of the situation of the game at that point in time as well as the logic of the game that had been programmed into it. It started the next game with a clean slate. It did not really learn anything from the earlier game. On the other hand in AI the computer is not logically programmed. Rather than that such system works on the principle of trial and error. It basically re-sorts or sorts out what is unsuitable and enriches what is supposed to be left from the process.

How does AI learn?

In this way, it permits human beings to recognize connections and draw conclusions as well as make predictions without knowing what the basic logic is in these cases. In that sense, it can be said that the way AI learns is similar to the way in which a kid learns. She or he knows that a stove should be avoided but does not why it is so. She or he does not know that it is because of heat. This is why she or he is not in a position to transfer that logic while using any other thing made of iron.

It is only with age and growing experience that a child learns how to think in a logical manner and apply this logic to other aspects of her or his life.

New situations baffle AI

There are plenty of things that people would be able to do in a particular situation that AI would not be able to do. Deep learning, which is the most dominant technique in AI, is incapable of interpreting natural language. AI also gets baffled by situations that are new to it. It is unlikely that these shortcomings would be solved any time soon.

What is the immediate future of AI?

It is expected that in the days to come deep learning will create an AI that would generalize and reason about the world in an abstract way. It would not be able to automate activities that are ordinarily performed by human beings. All the advancements that have been made in deep learning have happened because of pattern recognition. These are basically neural networks that memorize classes of things and when they encounter them, later on, you can more or less rely on them to deal with the same. The problem with this situation is that most of the interesting problems of cognition are not really classification problems at all.

General naiveté of AI supporters

Experts say that people are naïve if they believe that if they take deep learning and scale it 100 times, add more layers and 1000 times more data a neural net would be able to do anything that a human being is capable of doing. That is not true at all. There are also experts who are right in pointing out that AI is not the end all and be all solution. It is just one of the many tools that are available to humankind. They feel that new approaches are necessary to evolve AI.

Why are new approaches needed?

Without it, AI risks reaching a point beyond which lie all the problems that cannot be solved by pattern recognition. These experts feel that AI is greedy, opaque, brittle, and shallow. It is greedy because it needs huge amounts of data for the purpose of training. It is brittle because it has been seen that in most cases when AI is provided scenarios that are different from the examples that were used in order to train it such a system is unable to contextualize the same and breaks down. They are opaque because they are not like traditional programs.

Conclusion

This is in the sense that unlike those their parameters can solely be interpreted in terms of their own weights in a mathematical geographical system. This means that they are black boxes whose outputs are beyond explanation. This obviously raises doubts about how reliable they are as well. They are deemed shallow because they have been programmed using sparse innate knowledge and have zero common sense about human psychology or the world in general. These limitations imply that AI may not be able to reach the heights that a lot of its supporters have envisioned for it.