What is artificial intelligence?

Artificial intelligence is the subject of great hopes but also fears, with demands for greater regulation increasing as we speak. All the while, we don’t even have a universal definition for it.

Although artificial intelligence and its commercial and societal potential have been discussed for years, there still is no clear definition for AI itself. Rather than a singular phenomenon or method, AI involves a broad group of different research problems and technologies that are used to solve them.

On the one hand, AI can be understood as a research field that explores the intelligent-appearing behavior of artificial systems and devices. In practice, then, this involves automating intelligent-appearing activity with computers. On the other hand, AI can be approached by describing it as code residing in the memory of a machine, conducting tasks assigned for it, in a way that appears intelligent. These tasks may have been assigned by a person, or they may have been learned by the AI itself.

Whichever way we approach artificial intelligence, it is important to note that it is not only AI that lacks a clear definition, but also intelligence itself. This is why we prefer to speak of intelligent-appearing behavior, rather than simply of intelligent behavior.

If ever human-like artificial intelligence was developed, we would call it strong AI, as it could be applied to all kinds of problems. Current AI systems, however, are neither intelligent nor conscious like humans. Most AI researchers don’t even try to understand or imitate human thinking. Instead, they accept that artificial intelligent activity can be based on completely different mechanisms than natural intelligence.

Current AIs are all so called weak/narrow AIs: they each concentrate on tackling one particular problem. One filters out junk mail, the next recommends interesting Youtube content. Each works in a way that makes it possible to address the particular problem at hand. Because of this, a breakthrough in one area doesn’t necessarily translate into progress in others.

A bit like learning is part of human intelligent behavior, machine learning is an important aspect of AI. Machine learning refers to processes where AI experts adjust a computer program with the help of examples – often, this means vast masses of data – in a way that eventually enables the program to complete the task assigned to it. In practice, by showing a machine example images it can, for example, be made to understand what separates a particular smartphone user’s facial features from those of other people. On the basis of this, it can then reliably identify the user with the phone’s own camera.

At FCAI, we develop next generation AI methods that strive to be more precise and trustworthy, as well as more efficient than current methods in terms of data, energy consumption, and cost. Machine learning is at the center of the research. A special focus is on developing AI that helps its user, so that the narrow but powerful artificial intelligence augments the general-purpose – but in many ways limited – human intelligence in the best possible manner.

Petri Myllymäki is FCAI’s vice-director and professor at the University of Helsinki. Arno Solin is assistant professor at Aalto University and the coordinator of FCAI’s research program R3 on data-efficient deep learning.