Could information technology become an engine of unemployment, automating roles that once used scores of human workers?
Some writers and academics argue this is already the case, blaming communications technologies and automation for the falls in both wages and the number of men finding jobs in the US over the past 40 years. The problem of workers being displaced and wages being driven down will continue to get worse, they argue, as increased computing power and better software empowers computers and robots to take over more roles in factories and offices.
The net result, argues MIT economist Erik Brynjolfsson, could be a widening of the already sizeable gap between the earnings of employees and employers, and between college graduates and the less educated.
A sensible response to this shift, Brynjolfsson says, is to reassess workplace roles to find tasks particularly suited to people, and have humans and computers work alongside, rather than against, each other to complete them.
This notion of intelligence augmentation dates back to the early days of computing, when engineer Vannevar Bush coined the term to describe an intellectual symbiosis between man and machine. In 1945, Bush wrote about a future where an associative store of all books, records and communications called the memex would aid human recollection, a concept that today is embodied by the World Wide Web.
The power of man and machine working in unison was at the heart of a speech about intelligence augmentation by Ari Gesher, engineering ambassador with Palantir Technologies, at the recent Economist Technology Frontiers 2013 conference in London.
"The idea is to have a very well defined division of labour between the computing machines and the humans," he said, spelling out the complementary skills of men and computers.
"Most of AI is statistics. Any time you need to do this kind of statistical processing - be it figuring out how to target an ad, give recommendations to someone on Amazon or figuring out how to segment a voting population - computers are magic. They can really come up with very good robust answers to those kind of questions. These statistical methods basically depend on the characterisation of data remaining the same.
"We [also] know what humans are good at. It's making hypothesis, writing poetry, dealing with things like incomplete data. Recognising patterns that are similar to other patterns that have been seen before but are not the same."
The online game Foldit provides an example of how to exploit the relative strengths of humans and computers when it comes to information processing, he said. In the game players fold computer models of proteins to help scientists gain insights into their real-world structure. Computers can take the brute force statistical approach but human pattern recognition skills enabled by the brain's visual cortex has allowed people to devise solutions to Foldit tasks that computers have been unable to match.
By being aware of these relative abilities, and matching people and machines to the right tasks, you can outperform machines or people acting on their own, Gesher said.
"The idea is to do everything you can to remove the friction at the boundary between man and machine. Offload as much as possible onto the machines and bring in the injection of human insight into the system."
How to beat a chess grandmaster
The power of human-machine collaboration was demonstrated by two unranked amateur chess players in 2005, he said. The pair took part in a Playchess.com freestyle chess tournament, where individuals can team up with other people or computers. Using custom chess software running on three laptops to analyse play these amateurs were able to win a competition that featured the Hydra supercomputer and several grandmasters.
"They understood the problems of chess well enough to know how to communicate with the computers to get them to do all the right work," said Gesher.
In this instance the deciding factor in who was victorious wasn't the ability of the individual humans or computers to play chess, but how effectively the human and computer chess players were able work alongside each other, he said.
"The grandmasters knew a lot about chess but they didn't know how to use the computers as effectively as possible, how to leverage them to win."
The reason that perfecting the interface between man and machine can pay such dividends is that that increases in computing power have outpaced our ability to exploit them, Gesher believes.
"In 1960, $1,000 would get you one calculation per second. Today that number is somewhere around 1010, so you're talking about nine orders of magnitude in 50 years," he said.
"What's special about this? It's never happened before. This exponential growth inside two or three human generations is completely unprecedented in human history. We're still figuring out how to use those machines' effective power.
"[Therefore] small changes in the friction at the interface boundary, in how we offload work to computers, can lead to huge gains in the work that we do."