Mapping the imagination

For researchers at Carleton University’s cognitive science institute, the sky is no longer the limit as they work towards unlocking the secrets of human imagination.

Led by associate professor Jim Davies, the Science of Imagination Laboratory has been working since 2006 to design computers capable of mimicking the way the human brain constructs imagined scenarios.

“All of your dreams and all of the things that you imagine are some kind of reconstruction of things that you have seen before and things you’ve experienced before at some point, put together in new ways,” says Davies.

Using human memory as the foundation for imagination, the team is working on a computer program called Visuo which uses artificial intelligence to imagine scenes. In the program’s first successful test, Davies created a tree and Visuo imagined the rest, placing a patch of grass below the tree and a blue sky above.

However, modeling computers after human imagination has come with the challenge of charting an area of the brain which rarely receives attention from conventional neuroscience.

“People are sceptical of being able to measure people’s imaginations,” says Davies. “They sort of think that there is this wall where you can’t look beyond.”

The first ever computer-generated imagination.

To peer deep inside human cognition, Davies leads a multidisciplinary team of experts who bring knowledge from  neuroscience, psychology, philosophy and computer programming. Davies says that this range of specialization makes his lab uniquely suited for discovering how imagination works.

One member of Davies’ research team is Sterling Somers, a Ph.D. student and philosophy major turned computer programmer. Somers says that a central part of imagination is the “object relationships” which humans use to automatically associate certain images with each other, such as knives with forks and chairs with desks.

“For most of our imagination, despite what everybody says about how creative humans are, we kind of just go off of what we know and what we have already seen before,” says Somers. “Imagining is about taking memories and playing with them to create new representations.”

Capable of mimicking object relationships observed in humans, Visuo draws from a database of 50,000 tagged images and mixes and matches objects to build scenes that it believes are common occurrences.

Davies says that one day Visuo could be used to imagine virtual environments like the interiors of burning buildings so that firefighters or SWAT teams could effectively train. He says it could also be used to create animated movies.

Listen to the podcast to hear Davies speak about how humans imagine sizes.

“In the world of videogames and 3D movies, although it is all done on computers, 80 per cent of the cost is human artists making every little piece of clothing, or face and emotion, and all the buildings and architecture,” says Davies. He adds that Visuo would eliminate those costs by automatically generating original scenes from imagination.

And while Visuo’s current ability to complete only primitive scenes may pale in comparison to its possible uses in the future, Davies is optimistic about the program’s potential. He points out that it took humans nearly 50 years to create artificial intelligences such as Bell Canada’s automated sales representative named ‘Emily.’

“Realistically, I don’t see why it couldn’t be done,” says Paul Thagard, the director of the cognitive science department at the University of Waterloo. He agrees Visuo has the potential to be far-reaching, but says the next step will be to figure out how humans manipulate objects when they imagine.

“If I was trying to imagine something strange like Justin Bieber sitting on Lady Gaga’s shoulders, I can do that because I’ve got visual representations stored in memory and I can apply spatial manipulations to project Justin Bieber onto her shoulders,” says Thagard. “However, I don’t think computers can do that yet.”

Davies says there is still much to be learned about imagination before computers can be modelled to imagine  like humans.

“From what we know so far, we think that this program works the same way the human mind does,” says Davies. “Now we have to get people into the laboratory and see if they act the same way the computer program does. If they do, then that gives us some support for our theory of how the mind works.”

Produced By: Laura Kluz

Front Page and Tree Photo © Jim Davies

Antelope Photo © donjd2 on Flickr