What problems can artificial intelligence more easily solve, and what problems are far beyond its reach?

Posted by

In a previous post, I touched on a topic that raises many questions, namely when and what professions will be replaced by artificial intelligence. It is a race between humans and the practically unlimited capabilities of computers, which can perform logical and mathematical operations (that is, basically almost all cognitive operations) incomparably faster than humans. If we add to this the fact that they can learn using deep learning mechanisms and that they can work in parallel on a given task that is complex and complicated, then man is without a chance.

However, there is one area of problems, including organizational problems, that remains beyond the reach of a simple application of, for example, GPT Chat. These are problems whose solution is not a simple logical function of the components. I will now give two examples, the first strictly defined, where the end result is easy to calculate, and the second, where the number of end results is virtually unlimited.

Example 1 – together with friends we go to the cinema

You’ve probably been in a situation more than once when you’ve tried to organize with your friends to go to the movies or to a party. Suppose there are 4 of you friends. Everyone can go (state 1) or not go (state 0) to the cinema. When there is only one person – Adam – the decision is simple, he can go or not go. If he is to go with a person named Eva, 4 possibilities already appear in the results, only one of which means that they will go to the cinema. This happens when Adam and Eve want to go to the cinema at the same time. When there are 3 people, the possibilities are already 8, when there are 4 people, the options are up to 16. And so on. For the decision-making situation and the result – we will or will not go to the cinema – see Figure 1.

Figure 1 Options for going to the movies with friends

Despite the fact that the number of options grows at a rate of 2 to the nth power, all the outcomes are easy to determine, we know what can happen and when. The only question is who wants to go and who does not want to go to the movies, that is, what are the stable inputs.

This type of problem is very easy for artificial intelligence to solve, one might even say trivial, even with a large number of input variables.

Example 2 – how do you feel about the team in which you work

Let’s assume that you work in a team of 3 people. You and two co-workers. You need to determine whether you like the working atmosphere of your team. Hopefully, after reading Example 1, you have guessed the complexity of the task before you… And for several reasons.

First, your well-being will not be measured on a nominal 0/1 (no/yes) scale. You may feel satisfaction with your job, but… Or satisfaction may not be complete because… Or you don’t like working in a team, however, one element is positive that….

Secondly, the result “do you like the work atmosphere of your team” does not at all have to be calculated based on formal logic, that is, as in Example 1, based on the logical product. Actually, it is not clear what would be the mechanism of calculation and what actions would have to be performed… So there is no “formula” to calculate the result.

Third, there is the question of what values such a result should take? Is it just 0/1? Or perhaps a whole spectrum of values from full satisfaction to full frustration with the work in this team?

As you can see, this problem is very difficult to describe, let alone to apply any inference mechanism. At this point, we can only talk about the probability distribution of the results, if we somehow deal with the first cause above.

These types of tasks are impossible for artificial intelligence to solve without some approximations and operating on the probability distribution of possible outcomes. And therefore the solutions are not necessarily accurate.

That’s why people who, in a certain profession, constantly face problems like in Example 2, can be (for now) reassured that artificial intelligence will not take their jobs. For now…

See more on YT: Polski twórca ChatGPT o jego kolejnej wersji | Szymon Sidor