Machine psychology: A bridge to general AI?

Artificial General Intelligence (AGI) has been the Holy Grail of artificial intelligence research since the 1950s. So far, humans have not succeeded in creating artificial intelligence that can solve intellectual tasks like humans. But some scientists believe this could happen as soon as the next five years.

One of them is Robert Johnson, who recently defended his doctoral thesis in computer science at Linköping University. But unlike the pessimistic future scenarios about general artificial intelligence that often pop up in pop culture, he believes it can benefit humanity.

“Yes, I’m sure of that! It changes everything we do and is not just a continuation of ordinary artificial intelligence – it is something completely different. I think AGI will be used in a completely different way than artificial intelligence is today. To have a broad impact on society you can create a new type of agent, such as a virtual researcher or a psychologist, but it can be many more types,” said Robert Johnson.

At the same time, he sees the challenges facing technological development in a turbulent world. AGI can be used for a variety of purposes, such as creating divisions in society.

“This is a technology that we really have to be careful with. On the other hand, I also think that general artificial intelligence can help fight some of the destructive developments in society. It might help us humans become more caring. I’m open to that. AGI can help us evolve in this way as well.

But developing artificial intelligence at the same level as humans is a huge challenge. Different researchers have tried to solve this problem in different ways. Some believe large-scale language models like ChatGPT are the way to go, while others suggest simulating the brain. The path Robert Johnson chose to take could be called a principle-based path. This meant that he tried to identify important psychological learning principles that could explain intelligence and then apply them to computers.

While studying for his PhD at LiU, he worked as a clinical psychology teacher and researcher at Stockholm University, where he was also an associate professor, a background he drew on in his thesis “Empirical Research in Machine Psychology”.

“I use the principles of modern learning psychology to solve problems of learning, thinking and intelligence. Then I use a specific form of adaptive artificial intelligence, which is a logical system in which I try to implement the psychology of learning,” Robert said Johnson, who is now earning his second Ph.D.

This logical system is called a non-axiomatic reasoning system (NARS) and is designed to operate on the fly without complete information and with limited computing power. It provides flexibility that is important for dealing with problems that may arise in the real world.

The combination of NARS and learning psychology principles constitutes an interdisciplinary approach that Robert Johansson calls machine psychology, who was the first to propose the concept but now has many more participants Get started, including Google DeepMind.

The idea is that an artificial intelligence should learn from different experiences throughout its life and then apply what it learns to many different situations, just like humans do as early as 18 months old – which is Something no other animal can do.

“If you manage to do that in a computer, then you’ve really solved the mystery of artificial intelligence at the human level. I think psychology will be a potentially key science for general artificial intelligence,” Robert Johnson said.

It remains to be seen whether humans can successfully create human-level intelligence within five years. But Robert Johnson says there are many other aspects to consider beforehand.

“We live in a society with laws, regulations and moral premises. It is necessary to take a stance on how to view the rights and obligations of these agents. Maybe AGI is just a program you run in your browser, but I think if it is conscious, It’s still some kind of life.