We are far removed from been able to build anything like a dog in terms of brain capacity

Professor Luc Steels

They have been experimenting with a souped-up model of their popular robotic dog, Aibo, to find out how a robot like this can learn language.

"The main thing that we have found, which of course any mother would be able to tell us, is that you need a lot of social interaction," said Luc Steels, of the Sony Computer Research Laboratory in Paris, France, and director of the Artificial Intelligence Laboratory, at Vrije Universiteit Brussel.

"You cannot just show an object and say, 'ball'. You actually need to encourage and correct, and set up situations for scaffolding the learning," he told the BBC programme, Go Digital.

Brainy Aibo

For the experiment, Professor Steels and his team used an Aibo with twice the normal brain capacity of the model in the shops, and used a state-of-the-art speech system.

Slow process of structured learning

By teaching Aibo its first words from scratch, they were hoping to gain an insight into how children learn to associate objects with words and then attach certain meaning to those words.

Three types of objects were used, a red ball, a yellow puppet called Smiley and a small Aibo imitation called Poo-chi.

The team then tried to teach Aibo about what each of these objects was called. So, an instructor would take the red ball and ask the Aibo what it was, then repeat the word ball.

The robot dog would then store the image of the red ball and associate it with the word "ball".

Routine learning

A teaching session would unfold like this:

• Human: What is it? (showing a ball)
• Aibo: Smiley.
• Human: No, listen. Ball.
• Aibo: Ball?
• Human: Yes.
• Human: Is it Smiley? (showing a ball)
• Aibo: No. Ball.
• Human: Good.

Computers today in terms of brain capacity would be more like a slug

Professor Luc Steels

But Professor Steels realised they had a long way to go before they could develop a truly intelligent Aibo.

"Computers today in terms of brain capacity would be more like a slug so there's no comparison to human brain power," he said.

"We can only simulate certain parts of it. For us, the most important part is to do experiments to learn more about learning.

"We are far removed from been able to build anything like a dog in terms of brain capacity," he said.