As societies around the world become more intertwined with technology, the interaction between humans and machines can become complex or conflictual. However, artificial intelligence that understands how humans think, can offer a solution to this challenge, says Assistant Professor of Cognitive Science Jussi Jokinen.
“Can machines think?” asked Alan Turing in 1950 in his renowned article Computing Machinery and Intelligence. Turing’s answer to this question was a thought experiment known as the Turing Test: An intelligent, thinking machine must be able to behave in such a human-like manner that a human cannot detect whether it is a machine or a human being.
What kind of qualities and abilities would a machine have, then, for it to be considered intelligent? I noted, as a researcher of cognitive science, the Turing test is inspiring, as it forces researchers to define intelligence and ask what human intelligence actually is.
To be “human-like,” machine intelligence should include cognitive phenomena such as forgetting, occasional logical errors, and spending time on complex calculations, just a human being typically does.
For this to be credible, the machine needs to have a theory about human memory, logical deduction, and mental arithmetic. Formulating this theory of human intelligence in a manner that it can be written into a computer program constitutes a basis for my research in the field of cognitive science.
Aiming at artificial intelligence that understands human thinking
What concrete benefits could there be from a machine that implements a theory that simulates human-like intelligence? The answer, in short, is understanding, which is a key component of social intelligence. Human social intelligence can be defined as the ability to get along with others, an outcome that requires people to understand each other.
For example, a person may notice that his/her colleague is in bad mood and begins to think what events could explain this observation. A prerequisite for such consideration is a kind of a naïve, pre-scientific theory of how people’s minds work. A computer—unlike humans—does not have any innate ability for such consideration.
The rapid—perhaps even accelerating—technological evolution and increasing use of artificial intelligence causes an alignment problem between machines and humans.
A machine cannot cannot facilitate human goals unless it understands how it can succeed in this task. Solution to this alignment problem is a machine that does not necessarily think like a human being, but possesses a theory of how humans think.
An intelligent car directs the driver’s attention
Jokinen points to an example from his own research to underscore the point. In recent years, semi-autonomous cars have been regarded as an increasingly common everyday demonstration of the possibilities of artificial intelligence. In practice, however, many obstacles still hinder the development of fully autonomous cars. Perhaps the biggest of these pertains to the alignment problem.
People’s ability to drive safely in a highly complex traffic setting consisting of other vehicles and pedestrians is possible because of human social intelligence. Self-driving cars lack such intelligence.
In my research, I have investigated how a self-driving car could take into account how much a driver adapts to the different stages of automation.
Using a computational model of human intelligence, I was able to predict how a human’s actions during driving will change when using a lane watch. The model predicted how a driver’s observation of the road decreases due to the lane watch, while the eyes are fixed longer times on something else, like a mobile phone, for instance. This behaviour depends on how much the driver relies on automation and how demanding the simultaneous interaction task on the phone is.
This kind of model-based intelligence can be transferred to a self-driving car, which provides it with an elementary ability to understand a human driver. An intelligent car “notices” if the driver is relying too heavily on the safety brought by the lane watch and prompts the driver to pay attention to the traffic as well.
In the future, it is increasingly important that technology is able to predict human thinking and behaviour correctly.
Turing posed his question about machine intelligence more than 70 years ago, yet it is still impossible say for certain whether a machine can ever think like a human being. However, researchers can learn and benefit from Turing’s question by asking what makes humans intelligent – and if this intelligence can be formulated into an exact theory that could help a machine understand its user better.
Jussi Jokinen works as an Assistant Professor of Cognitive Science in the Faculty of Information Technology at the University of Jyväskylä, Finland.
Subscribe to the JYUnity newsletter
Get latest articles from The University of Jyväskylä’s stakeholder magazine into your email. You can cancel your subscription at any time.