Using the internet to harness the wisdom of the crowd

Collective intelligence is a growing trend that seeks to exploit the computational power of millions of users

You have probably done it but maybe you didn’t realise. Or maybe you did it on purpose, but it was a game. What is it? Collective intelligence, or “human computation”, is a growing trend that looks to harness the wisdom of the crowd to solve problems.

Today, enormous computational power is distributed among millions of users, and the internet offers a means to connect it, explains Prof Barry Smyth, professor of computer science at University College Dublin.

One early initiative that looked to harvest that power was “SETI at home”, where people could sign up to run code on their computers during downtime to analyse radio telescope data from space in the search for signals from alien intelligence.

“The hope was that your computer would be the one that would find signs of extraterrestrial life,” says Prof Smyth. “That network of home computers made SETI into one of the most powerful supercomputers on the planet at the time.”

Yet it wasn’t quite collective intelligence because the human didn’t have any input other than signing up. Enter the “CAPTCHA”, a puzzle that presents a series of distorted letters and numbers and asks you to type them into a box. You’ve probably encountered it while buying tickets, and it ensures you are a human rather than a “bot”, because computer algorithms aren’t good at working out the symbols.

“There are literally hundreds of millions being solved every day,” says Prof Smyth. “So people started thinking of ways in which we can harness that ability for other tasks, and that really started collective intelligence as we know it now.”

One initiative, reCAPTCHA, has been helping the likes of Google to improve the accuracy of projects to digitise books and periodicals. The software that converts images of the written word to text isn’t perfect, so reCAPTCHA presents unrecognised words (scanning errors) as a second word for us to decode in our CAPTCHAs.

“So now modern CAPTCHAs present two distorted words, one of which is known to the computer, and the other is a scanning error,” explains Prof Smyth.

“The human must attempt to decipher both to sign up to their free email account or register with their social network. As millions of users decipher these scanning errors, their guesses tend to converge on the correct version of the mis-scanned words.

“Collective intelligence helps to better predict what the word should have been. They were able to farm out tens of millions of scanning errors and were able to scan large amounts of text very quickly.”

See Prof Barry Smyth's talk on collective intelligence at dublintalks.ie

Sharing results: Smart 'social' search

Prof Barry Smyth is himself involved in bringing more collective and “social” intelligence to search on the Internet. He’s co-founder and chief scientist at HeyStaks, a UCD spin-out that has developed technology to let friends share search results online and across mobile devices.

In practice that means if you are searching for information and your friends have already found something interesting, you can see and build on the fruits of their labour.

“We are trying to figure out a better way to rank search results,” says Smyth. “The results that we are promoting to the top of the results list are essentially the product of the collective intelligence of other people who searched before you.”

The technology grew from research at the Science Foundation Ireland-funded CLARITY Centre for Sensor Web Technologies – a partnership between UCD, Dublin City University and Tyndall National Institute – which Prof Smyth directs.

The spin-out, which has offices in Dublin and the US, has already secured investment of €1 million in equity funding and is “well on the way” to making social search a commercial reality, according to Smyth.

“From the company’s perspective, they have fully developed the technology now and the next 12 months is going to be about user acquisition and the commercialisation of it proper.”

More generally, he reckons that the scale of collective intelligence is now becoming more apparent.

“If you think about some of the major endeavours of human effort, like building the pyramids or the Panama Canal, you are talking about tens of millions of hours of human effort,” he says.

“That is just a tiny fraction of the effort that is available online.”

All the play for: Games with a purpose

Games offer another way to engage humans and get them to help solve computational problems, notes Prof Barry Smyth, who describes how a group at Carnegie Mellon University came up with a game called ESP to improve how images are tagged and indexed online.

“You play the ESP game in a browser against an another unknown player you are paired with,” he explains. “Each of you is shown same picture and the game is essentially to guess tags that apply to the picture. You don’t see the other player’s guesses but if one of your tags match then you score points and move on to a new picture and keep going until your timer runs out.”

Players are motivated to pick meaningful words, because they are more likely to match what the opponent was typing in, notes Smyth, adding that the game itself paid off too.

“Google bought that technology and have used it for a number of years to improve their image search engine,” he says.

Shapes of things to come: Foldit to solve it

Another game with a purpose, Foldit, has been making strides by using collective intelligence to help solve scientific puzzles.

The game, developed at the University of Washington, involves players manipulating three-dimensional shapes that correspond to protein structures.

“The Foldit people decided that humans are really good at manipulating three-dimensional objects – we can twist and contort them into different shapes – and that is the type of thing that is happening when proteins are folding,” says Prof Barry Smyth.

“Meanwhile, scientists knew some of the general constraints that dictated how a protein would fold into a three-dimensional structure.”

To play, the user manipulates the proteins on the computer, moving the parts into more efficient arrangements to earn higher scores. The solutions ultimately give scientists hints to work out real protein structures in biology.

And it works: foldit helped to solve the structure of an enzyme that helps an Aids-like virus to establish an infection. Knowing its structure could inform new therapies, but the puzzle remained unsolved in science.

Within days the Foldit players were able to provide information to scientists to model the enzyme structure, and groups of gamers were listed among the authors in the resulting paper last year in Nature Structural Molecular Biology.