YORKTOWN HEIGHTS, NY—When it came time for Thomas Malone, Director of MIT’s Center for Collective Intelligence, to address the crowd of cognitive computing enthusiasts today at IBM's research colloquium, he began his talk with a quote.
“The hope is that, in not too many years, human brains and computing machines will be coupled together very tightly, and that the resulting partnership will think as no human brain has ever thought and process data in a way not approached by the information-handling machines we know today.”
Malone first read that statement, written by the computer scientist J. C. R. Licklider in 1960, as a college student and he said it inspired his subsequent career researching human-computer symbiosis.
“In the traditional vision of artificial intelligence,” he told his audience, “if people are involved that's cheating. But in this vision if people are involved that's the point.”
In order for us to understand how people and computers working together can achieve more than they can working alone, Malone said, we first have to understand how people working in groups can outperform people working as individuals. He and his colleagues created experiments to measure what he calls “collective intelligence.”
“Individual intelligence,” he said, “predicts a lot of important things in the real world about people. It predicts even life expectancy for individual people. But as far as we can tell no one has asked the question if there's a general cognitive ability for groups.”
To answer that question he and his colleagues assembled about 200 groups of two to five people, and asked them to complete various tasks. “And the answer we found,” he reported, “was yes, there is a general cognitive ability for groups. There's a single statistical factor for groups that predicts how well they will do at a task. We call that factor 'c,' and it turns out it predicts a lot of things."
“The next question is, what predicts 'c?' Now before we did this work we were afraid that collective intelligence would just be the average intelligence of the individual members of the group. Now it turns out that is correlated, but only moderately so. It's not just having smart people that makes a smart group.”
Instead they found that collective intelligence correlated strongly with the social perceptiveness of individual members (as determined based on their ability to judge the emotion conveyed in photographs of human eyes). Secondly they found that equal particpation among group members also lead to stronger outcomes. And thirdly he said, to much excitement in the audience, the number of women in a group also positively correlated with the collective intelligence. “This last result,” he noted, “is mostly explained by the first one. Women on average score higher on social perceptiveness than men.”
“Now what does this mean for computers and people working together? ” Malone asked. “The computers can help by helping the people work together more effectively, and the machines can be more useful to the degree they have social intelligence. Maybe that means the machines have better user interfaces, maybe the machines have better models of the people they're working with.”
Malone has presented people and computers with the challenge of predicting the next play in a football game.
“It turns out,” he said, “that the computers were significantly more accurate than the humans at predicting the next play, but the combination of humans and computers together were significantly more accurate. Humans and computers together were also much better in terms of their reward to risk ratio. I think we've learned that people and machines can do pretty much the same tasks, and the results can be combined in a way that's more accurate than either one alone. ”
He cited another example where machines divided an article writing task among a large group of people. The machines also delineated the way in which other people scored the human writers to determine which written portions should be involved in the cumulative article.
“These articles,” Malone said, “tended to be better in quality than an article written for the same cost by a single person. What can we learn from this? Even when computers cant do any of the basic tasks you're trying you do, they can still provide a very useful function by managing the humans doing these different tasks. And on of the things you may need to do is break up the tasks in unusual ways.”
“I think,” he said in conclusion, “as the world becomes more interconnected, it will become more and more useful to view all the people and computers on our planet as part of a single global brain. And perhaps our future as a species will depend on how we can used that intelligence to make choices that are not just smart, but also wise.”