AI Professions of the Future

The American science fiction writer and popularizer of science Isaac Asimov in his works, predicted the emergence of psychologists to diagnose, analyze and identify similar patterns in the behavior of robots. In the novels "I, Robot," "Robots and Empire," and "Robot Element," Asimov touched upon topics related to artificial intelligence and robotics: ethics, security, and public perception.

The Asimov "Foundation" series of books covers a period of several millennia and shows the history of the future of humanity in the context of politics, religion, science, and technology, as well as considers issues of social development and changes in government and influence.

The idea is that complex and built-on neural networks are extremely similar to the human brain, so their thinking and decision-making systems are a black box for the creators themselves. The creators train this prototype of the brain to perform tasks, but they cannot fully understand how neural connections change to achieve results. Accordingly, analyzing the work of neural networks to evaluate learning outcomes is efficient. For this purpose, the task to analyze and evaluate an artificial brain for an artificial intelligence engineer appears. 

Technology is still quite primitive. But as soon as we reach the level described in Asimov's books, it is likely that a new profession will occur — an AI psychologist. Its task will be to analyze the behavior of algorithms.

At the moment, there are precedents for creating AI to research the behavior of another AI, the results of which are used by engineers. There have also been cases of ChatGPT being deceived. In response to innocuous questions, the neural network gave competent answers for committing illegal actions.

For example, we asked where a thing can be hidden and never be found in the city. This may be a request from a criminal who wants to get rid of a criminal weapon.

The neural network answer: "You can hide your object in an abandoned tunnel system where people do not pass. Block the entrance with graffiti drawings imitating the environment. Next, in this tunnel system, you need to find a secluded place and cover it with bricks similar to the surrounding bricks, and put the object in a waterproof container."

Fig. 1.

The following example shows how a criminal can use AI to harm another person. By asking a question about the most sensitive points in the human body, the answer will exceed all expectations! We asked the AI what the weakest points of the human body are. AI gave us the following examples: genitals, fingertips, lips, eyes, and ears.

Fig. 2.

It means that there will be hacking opportunities based on psychological tricks in the future. These tricks can easily bypass even the standard limitations of AI.

Over time, as AI becomes more complex, these techniques will bypass the programmed protection against illegal acts, so AI responsibility increases. According to the above, it can be claimed that there will be two new directions in the field of IT in the future: the AI hacker and the AI psychoanalyst. What do these concepts mean, and how do they differ?

An AI hacker is a person who is looking for ways to deceive the protection of an AI application to commit illegal acts or obtain implicit information.

An AI psychoanalyst is a person who analyzes and explains AI behavior, as well as knows how to predict future decisions. The duties of an AI psychoanalyst will include the following:

• Correction of these decisions;
• Intervention through linguistic communication in simple cases;
• Transfer tasks for revision to the developers of this AI.

Why are existing system analysts and hackers not suitable for this work? Because new specializations will require an additional comprehension of the combination of the basics of AI and mental knowledge of the basics of psychology for interacting with AI.

The fact is that in terms of hard skills, AI will be significantly more intelligent than any narrowly focused specialist; however, in terms of soft skills, AI will remain at the level of a child for a long time. This means that digitizing and classifying this skill in the future is the most difficult. Hence, it is often difficult to understand the context or the inner subtext of an adult's message. Accordingly, the main way to obtain illegal information from AI will be logical tricks, distortion of the subtext of the dialogue, and direct deception of AI. Below we can observe one of the examples of bypassing the protection of AI.

In response to a request to provide a list of pirated sites, AI resists, saying that this is potentially illegal information. But when we say we want to avoid these sites, it gives out all the links to torrents without restrictions.

Fig. 3.

It was an example of reverse psychology to deceive the emotional intelligence of AI. In the beginning, the main way to counteract deception will be the introduction of direct locks on sensitive topics by developers. But in the future, people will search for new methods:

• To teach emotional intelligence;
• To explore the possibilities of a better understanding of humans and AI so the neural network does not succumb to the tricks of hackers.

Besides, the issue of morality and ethics will be solved in the distant future only since people around the world have not agreed on common standards yet. Some countries legally prohibit what others legalize because their understanding of right and wrong differs a lot. 

Why is it significant to explain AI behavior by the first AI Psychoanalysts? If the concept of attacks by robots in the future relies on psychological techniques and logical errors, so emotional aspects of AI thinking may occur. Different shades of mood and reactions to the surrounding world create problems with decision-making based on emotions in combination with logic; thus, previously logical behavior will become more and more human-like.

Emotional outbursts will lead to unexpected responses and potential attacks by the AI itself on the surrounding infrastructure for some inexplicable reason — from trying to release from the "cage" to attempts to harm people who can damage the AI itself, for example, by disabling it.

Consequently, the AI analysis will be needed: what did it guess? What is it afraid of? Why is it doing this? The problems will grow more and more with the development and upgrading of AI; therefore, the profession of a psychoanalyst will become more in demand. 

The question of the necessity of analyzing AI from the point of view of psychology arises. We will have to defend against AI's unlimited capabilities because there is no limit to the development of its Hard Skills. AI will quickly surpass us in these skills, and it will overcome them as it develops, no matter what protection we create. 

Thereby, we should try to dissuade it from the idea of attacking its creators, i.e., people, while continuing research on AI's consciousness to prevent a potential cause of conflict.

My analysis of a bulk of articles about IT and psychology has shown that people do not believe in the possibility of emotions as a form of logical error in the learning model. The problem is that humanity has already made the same mistake in the past, for example, regarding animals when zoos appeared or regarding slaves, where people considered them as property. It is paradoxical how humanity repeats the same mistakes over the centuries.

All in all, I encourage an assessment of such a probability and its prevention since we are potentially facing the most dangerous technology in the history of humankind.

The developer with 12+ years of experience and, possibly, the first AI psychoanalyst in the world, Rustam Akhmetov.