Understanding Human and Artificial Intelligence
|Datum:||22 september 2020|
|Auteur:||Team Industry Relations|
Understanding Human and Artificial Intelligence
Artificial intelligence is in the spotlight in today’s world and it gets more attention than human intelligence. Still, human intelligence and artificial intelligence are closely related. People use metaphors of the human brain to explain artificial intelligence and there are many similarities regarding the architecture of both. In professor Niels Taatgen’s group, this relation is turned around. They work on a cognitive architecture called PRIMs that puts artificial intelligence to work to achieve a better understanding of the human brain. Understanding human intelligence is a huge challenge, but improved understanding holds great potential for psychology and AI research alike.
Interaction between human and artificial intelligence
The progress made in AI research in recent years is impressive. AI technology is practically implemented in every facet of our lives. At the same time, the current successes of AI are based on restricted problems with a lot of data. As Taatgen, professor in Artificial Intelligence and chair of the board at the Bernoulli Institute for Mathematics, Computer Science and Artificial Intelligence, puts it: “A lot of data and restricted problems; this is the winning combination right now.” In contrast to us humans, AI cannot properly transfer knowledge. Having a board game night with an AI that is trained in one game is tricky - it might beat you in chess, but for every other board game, it has to start learning from the ground up.
At the same time, there are many unsolved mysteries about human intelligence. Usually, psychologists use experiments to get better insights about human intelligence, but those cannot be scaled. Taatgen approaches the matter from another perspective that brings the scalability of AI and the architecture of human intelligence together. The project is called PRIMs, derived from primitive operations.
Taatgen developed PRIMs as a cognitive architecture. That is, a computer simulation of human cognition helping to understand the cognitive strength and weaknesses of our species. He says that an AI which is built on that structure does not have the goal to exceed human capabilities. It will not beat the best chess players in the world and it will not be able to detect anomalies in millions of lines of customer data. Instead, PRIMs seeks to find a bridge between human and artificial intelligence where one hand washes the other. The project crosses faculty borders. He works together with a neuroscientist, a psychologist, two linguists, and two AI researchers.
Potential outcomes and applications of the PRIMs project
Overall, there are two possible applications of the cognitive architecture: first, improving AI and second, improving human and social aspects of our lives. You might wonder, how can we improve current AI technologies by limiting the model to what humans can already do? It is often overlooked that modern AI technologies are essentially based on the human brain. Neural networks, for example, draw inspiration from human brain cells. Taatgen exemplifies this, saying that Leonardo Da Vinci failed to build a plane with flappy wings - still, in the end, a plane has wings like a bird, although steady. This is to say, AI and the human brain are not intended to function exactly the same, but some aspects, like wings, will likely be similar. It is not only about building an AI that is just like the human brain, but understanding our intelligence to understand computational intelligence better.
PRIMs in practice
In psychology, experiments are used to gain insight in the building blocks of intelligence. With PRIMs, it is possible to simulate how these building blocks work together to produce intelligent behavior. Taatgen says, it is almost like having a programming language for the human brain. For example, he was involved in a project about multi-tasking where the group wanted to predict if two tasks could be done at the same time. Of course, one could run an experiment with all possible combinations of tasks, but is that really efficient? Instead, they developed a model capable of predicting an outcome like ‘this combination probably goes well together’ and ‘this combination of tasks will collapse’.
Another application of PRIMs is in education. Cognitive tutors are intelligent tutoring systems based on a model of your intelligence. They assess what you know, what you do not know, and what you are capable of. Since capability is so individual, there is no one-fits-all solution for teaching and studying. An architecture like PRIMs helps to map the cognitive abilities and skills of a student to personalize the subjects and timetable.
While Taatgen’s research is theoretical, there are countless applications of it. Overall, there are so many things we have not figured out about intelligence, be it human or artificial. Taatgen’s research shows that the two forms of intelligence should not be treated as separate fields - understanding one helps the other.