Logo: to the web site of Uppsala University

Focus on AI and its users

Currently the risks of Artificial Intelligence (AI) are discussed intensively. This discussion focuses mainly on technical aspects or on internal technical processes which may give AI unforeseen super powers. The fears are that AI may become more intelligent than humans, emerge as an independent agent with its own agenda, force us into something we do not want to be, lead evolution in a radically different direction, and possibly transform the whole universe. Although we cannot to rule out these risks, and because we want to do something to eliminate the risks, we would need to look to other conditions than the exclusively technical features, which may be playing an important role in causing these concerns about a very powerful AI.

Let us first see what we expect from AI, and what AI actually does. AI, as it is designed and used today, provides answers to problems and delivers products and services. All our efforts to develop AI tools and systems aim to support their ability to solve problems we humans have, and to satisfy our needs. All interest around AI is concentrated on achievement of goals, solutions of problems, and satisfaction of needs. This is what we expect AI to deliver and we design it accordingly.

If we now turn our focus on ourselves and try to see how we manage to handle our problems we can identify a certain tool we have for this purpose. It seems that thinking is the biological tool for achieving goals, solving problems, and satisfying needs. However, this is not something we like so much to use. For us it is not easy to reason in a rational way, and it is even less easy to control the process of thinking even though we are conscious about the high value of right thinking. For us as persons it is necessary to have the skill to run a rational thinking process and the emotional strength to takeresponsibility for our own decisions. But if we try to reason in a rational way, we need to put great effort in this, concentrate our focus and spend a lot of energy. We have also to postpone making the decision and proceeding to action until we can come to a conclusion and find a suitable answer to the problem at hand. On top of these difficulties is the feeling of responsibility and the anxiety that follows with it when one knows that if something goes wrong there is no one else to blame but oneself.

Rational thinking

Suppose now that we, despite all these difficulties and costs in energy and time, still make the effort to think rationally in order to find the answer to the problem we have in front of us. Will we achieve a satisfying result? The answer is unfortunately “maybe”. So, the optimal solution cannot be guaranteed through the adoption of a rational, philosophical or scientific approach. Not in the real world, compared to an ideal or metaphysical world (see for example the discussion on the application of the categorical imperative in real life). The fact is that in real world non-rational ways of handling are successful very often. This has been studied empirically in psychology of human cognition and it is the so-called heuristic. In philosophy we have the Aristotelian habits, which also function very well.

Why should we then adopt the rational way of handling our problems given all the difficulties and the fact that we may still come to a wrong conclusion and action? When we also know that not so autonomous ways of thinking have a great chance to give us a satisfying solution? So, most of the time we do not adopt rational thinking.

AI can strengthen this. We have already seen the effects of technology on our thinking. For example, during the early times of Internet in the world many of us were enthusiastic about its potential to give us access to all kinds of information, especially information contradictory to our beliefs and certainties. We were seriously expecting it to have a healthy impact on thinking. Unfortunately, that proved to be wrong. Instead of promoting doubt and dialog it allowed people to look for and connect with other like-minded people in order, not to examine and falsify their beliefs, but to confirm what they already are convinced is true. Another aspect worth consideration is that algo- rithms, e.g. in social media, suggest links which are supposed to be in accordance with what the user likes.

Design and application of AI

The design and use of AI follows the same path. We create it in order to give us answers, not to ask questions that dispute our wishes or beliefs. We expect it to give us the products or the services we need. We want AI as a tool for the satisfaction of our needs, as our thinking is, but a biological one. The key point here is that advanced AI is or will be an immensely better tool. Will it then replace thinking?

A great risk is that this will happen. Unless we change focus in the design and use of AI: from the answers, or products or services provided, to the process leading to these results. An AI designed to help us use our thinking in order to find the answers and to produce the things we need, is something different than using AI to deliver the answers and the products directly to us. We can imagine AI as a partner in an ongoing dialog.

First and foremost, we have to be highly conscious about the difficulty of such a plan. For example, we have already seen how Internet is used, meaning that the same may happen with the use of AI. But if we want to use AI as a support tool for our thinking, to help us run the process of thinking in the right way, we have first to ask the question whether this is possible at all. Are we able to accept or at least tolerate something that disturbs our state of mind and makes us doubt our convictions? Especially when we know very well that this same tool can deliver to us the best possible answers to our problems, and give us satisfaction, harmony and happiness?

This seems to be a very difficult plan to design and apply. However, it is pos- sible to think that a compromise may give it a bigger chance. A compromise that is in accordance with the real-life use of thinking. We are never perfect philosophers, scien- tists or democrats. We cut corners and we use heuristics, but still, we manage somehow to reach our goals and to satisfy our needs. Then, what about an AI that sometimes and when there is no other better alternative because of the high demands of the problem at hand, support us to think in a rational way? But also, an AI that most often support us in finding and using the best possible heuristics and habits?