Artificial Intelligence System Imitates How Children Learn

 Artificial Intelligence  

A research team claims they have created an AI program that can learn how to solve problems in many different areas. The program is designed to imitate certain aspects of children’s cognitive development.

Typical artificial intelligence programs lack the versatility and adaptability of human intelligence, otherwise known as general intelligence. For example, they cannot come into a new home and cook, clean and do laundry.

In artificial general intelligence (AGI), which is a new field within AI, scientists try to create computer programs with a generalized type of intelligence, enabling them to solve problems in vastly different areas.

In August, "exceptional contributions to the AGI field" earned a team of researchers from the University of Gothenburg and Chalmers University of Technology the Kurzweil Prize for the second straight year.

"We have developed a program that can learn for example basic arithmetic, logic and grammar without any pre-existing knowledge," says Claes Strannegård, a member of the research team together with Abdul Rahim Nizamani and Ulf Persson.

The best example of general intelligence that we know of today is the human brain, and the scientists’ strategy has been to imitate, at a very fundamental level, how children develop intelligence. Children can learn a wide range of things. They build new knowledge based on previous knowledge and they can use their total knowledge to draw new conclusions. This is exactly what the scientists wanted their program to be able to do.

The system Strannegård and his team created is called O*. O* uses arbitrary symbolic domains, including arithmetic, logic, and grammar and can start from scratch and learn the general laws of a domain from examples. The main learning mechanism is a formalization of Occam's razor.

"We postulate that children learn everything based on experiences and that they are always looking for general patterns," says Strannegård.

Related articles David Hanson Wants To Create Genius Machines  Ben Goertzel At TEDx Hong Kong  James Barrat on Our Final Invention

A child who for example is learning multiplication and who knows that 2 x 0 = 0 and 3 x 0 = 0 can identify a pattern and conclude that also 17 x 0 = 0. However, sometimes this method backfires. If the child knows that 0 x 0 = 0 and 1 x 1 = 1, he or she can incorrectly conclude that 2 x 2 = 2. As soon as the child realises that a certain pattern can lead to incorrect conclusions, he or she can simply stop applying it.

"We postulate that children learn everything based on experiences and that they are always looking for general patterns."

The child can in this way create a large number of patterns not only in mathematics but also in other areas such as logic and grammar. The patterns in a certain area can then be combined with each other and make it possible to solve entirely new problems.

O* works in a similar manner. It can identify patterns by itself and therefore differs from programs where a programmer has to formulate which rules the program should apply.

In their paper, the researchers conclude:

At present, the main heuristic of O is to confine the search for solutions to the cognitive model. This heuristic is unsophisticated; nevertheless, it is sufficient for reaching human-level performance in certain domains. It can potentially be combined with more traditional heuristics to improve performance further. 

We find the approach to AGI as proposed in this paper promising.

"We are hoping that this type of program will eventually be useful in several different practical applications. Personally, I think a versatile household robot would be tremendously valuable, but we’re not there yet," says Strannegård.

In the video below, Strannegård explains O* at the the Seventh Conference on Artificial General Intelligence (AGI-14) in Quebec City.

SOURCE  University of Gothenburg

By 33rd Square

Embed