A 'Babelfish' could be the web's next big thing, says AI expert

Ford Prefect and Arthur Dent (played by David Dixon and Simon Jones) in the BBC's adaptation of The Hitchhiker's Guide to the Galaxy, by Douglas Adams. Photograph: BBC

Though the idea of the “Babelfish” - a thing able to translate between any two languages on the fly - was created by the author Douglas Adams as a handy solution to the question of how intergalactic travellers could understand each other, it could be reality within 25 years. At least, that is, for human language.

Prof Nigel Shadbolt, a close associate of the web inventor Sir Tim Berners-Lee, says that the idea of automatic machine translation “on the fly” is achievable before the world wide web turns 50.

Shadbolt also forecasts that future changes to the web will mean people will be “connected all the time” to medical diagnostic systems – but also that search companies including Google and China’s Baidu may face challenges as web use shifts from the desktop to handheld and mobile devices.

Having first used the web in 1993, via an early version of the Mosaic browser while on a visit to Canada, Shadbolt now thinks that it opens up huge possibilities for artificial intelligence systems built by connecting computers across the web - so-called cloud computing - that will be able to enhance daily life.

“I think we’ll have a sort of Web 3.0 and smarter systems,” he told the Guardian. “With more cloud computing you can imagine that there will be more machine translation.”

The growth of the web and the addition of artificial intelligence systems has already led to automatic translation such as Google’s and Microsoft’s text translation services. But Shadbolt foresees the exponentially growing number of computers connecting to the web providing far greater power.

“A ‘Babelfish’ system will be closing in on reality because of the large-scale resources that will be available to do entire voice translation on the fly,” Shadbolt said. “I would be surprised if in 25 years we haven’t got enhanced Bluetooth-based translators that you can just put in your ear.”

Machine translation and transcription presently struggles with real-time speech. IBM’s Watson system, which won an episode of the quiz game Jeopardy! against the two best human players of the game, had its questions delivered by text (at the same time as the human contestants heard the questions read out). So far no machine has demonstrated an ability to carry out real-time transcription of conversations, though the growing power of cloud computing - and computers generally, which roughly doubles every 18 months - could bring that within reach in just a few years.

Update: Microsoft demonstrated a system in 2012 which can do almost real-time translation between English and Chinese with a lower error rate than before: “rather than having one word in 4 or 5 incorrect, now the error rate is one word in 7 or 8,” noted Rick Rashid, Microsoft’s chief research officer, in a blogpost at the time.

Separately, the OECD calculates that by the time the web turns 50 it could be linking 50bn connected devices worldwide, and that within 10 years people might have as many as 50 internet-connected devices at home.

Shadbolt thinks that “lifelogging”, collecting details about what one sees and hears automatically through an internet-connected system, “will be a big future of the web - people will put more and more of their life on the web - and technology and businesses will be around that will monetise that.”

The growing spread of the mobile web and of web-connected devices which generate their own information and link to each other though will require shifts. “Google uses links between web pages to build its search index,” he explains. That index means that a search for a phrase will rank a series of web pages based on the perceived “authority” of pages containing that phrase, and of those pages which link to them.

“But it gets problematic if people or machines make links that just don’t make sense in the normal way,” Shadbolt says. “Maybe there will be other kinds of search as the ‘machine web’ comes along.”

IBM’s Watson proves that you can get “quite deep integration” with the wealth of information that’s online to answer a broad range of questions effectively, he notes.

“I foresee a world in which there will be augmented intelligence, where we are made smarter by our connection to the web.” He notes that there will be downsides - “cyberterror and cybercrime” - in which there will be an “arms race around security”.