IDF Day Zero: Computers that feel - and squeeze the cheese

Andrew Chein, director at Intel Research, has taken the stage at the International Press Day of the Intel Developer Forum in Shanghai, and he's in full-on future mode.

Computers, he says, are very good at complex analysis and optimisation. Practically every company in the research communities are using computers to design and visualise things we couldn't manage any other way: from genetics to global transport networks, only big iron has the power to crunch the numbers and present the results in ways we can use to engineer answers.

Look at a different part of the world we're part of every day: humans do things every day that computers aren't very good at. Laughing and joking – it would be wonderful if computers could understand that. Learning – not just rote, but tutoring throughout life. Interacting – computers that could recognise not just the environment they're in but how the humans in it were acting and feeling, in order to make useful suggestions.

There are a couple of vivid demonstrations: a video of a robot hand with electric field sensors that can feel things before it even touches them. It delicately wraps itself around a human arm without making contact - then closes firmly on a red string bag of Baby Bell miniature cheeses. Perhaps I'm a bit sensitive to scrotal subtext, but I suppress a tiny wince.

I'm more comfortable with a display of image recognition working orders of magnitude better than I've seen before – rapid-fire identification of toys, and of particular individuals waving against a background of other people doing other things. Such feats take around four teraflops of processing and ten kilowatts of power, says Chein: getting them into a handheld device taking under a watt is going to be hard, but Intel thinks it can do it. Half of that progress will come from new algorithms, and half from better hardware.

All this is being rolled into the wittily – and scarily – labelled ESP programme, for Everyday Sensing and Perception. If these smart, feeling, empathic machines are to be part of our world, he says, they'll have to get 90 percent of their actions right 90 percent of the time, and work across the whole range of human activities. Anything less and they'll just be annoying.

Chein really warms to his theme and the briefing takes on something of the tone of a revivalist meeting. By equipping computers with sensors monitoring everything around them, and inference engines capable of correctly deciding what that data means, it'll be the machine equivalent of the Cambrian Explosion.

If you're not heavily into paleontology, that reference bears some explanation. For around three billion years, life on earth was stubbornly unicellular – basic bacteria, plankton and algae. Then, around 600 million years ago, a huge surge in innovation produced a massive and still badly understood burst of huge numbers of most peculiar creatures. Thirty million years later, another enormous reshuffle produced the basic structures from which all modern complex lifeforms have evolved. The key happening was life getting a body: Chein's big idea is that by integrating computers with real life as experienced by humans, a similar burst of innovation will happen for IT. That's pretty big.

Phew. You've got to love Intel when it gets this sort of bee in its bonnet, even if it proves a little too divorced from reality to really catch hold – as happened with Pat Gelsinger's Radio Free Intel in 2003, when he predicted that every chip Intel made would talk to other chips via radio, building on what “we are pursuing with the ultra wideband [...] will be a critical step in enabling the next generation of communications in the future, a common UWB radio platform for a wide range of CE, PC, and mobile uses". Five years on, and UWB is far from common.

Perhaps in recognition of previous evangelism, Chein is careful to outline some of the many problems. Energy efficiency, even just for the sensors by themselves, is far from adequate. Sensors don't produce neat, unambiguous data: cleaning and sorting out real-time real world information is a massive and badly understood task, even before you try to make sense of it.

But as he said, if you can make an application that reliably delivers fun, you've really got something. I thought briefly of Douglas Adams' Sirius Cybernetics Corp., whose slogan “Your Plastic Pal Who's Fun To Be With” so deftly parodied the gap between marketing gloss and grotty reality: I'd be surprised if Chein hadn't got that in the back of his mind too. And if he hasn't read The Hitch-Hiker's Guide To The Galaxy, he should. Soon.

The show closed with yet more hyperactive enthusiasm, this time from Intel Fellow and radio guy Krishnamurthy Soumyanath. He got down and dirty with some of Intel's new techniques for putting lots of radio smarts onto a single chip capable of very high performance and extreme flexibility at very low cost. I'll revisit what he said at a later date: he touched on lots of very important and highly technical aspects of what makes radio work, each one of which deserves a decent airing. In brief: there are lots of ways to do cheaply in digital what has been done in complex and expensively analogue ways before, and as each bit of the radio transfers from analogue to digital it becomes easier and makes more sense to integrate it tightly with the existing digital circuits. Moreover, if you get really clever you can digitally compensate for the shortcomings of the remaining analogue bits to make them very high performance without having what would normally be the associated very high costs.

Thinking about it, this sort of approach only happens when you have a bit of pure engineering enthusiasm pushing away at the sensible – Radio Free Intel still lives.

But while all this future stuff is fun, it's what's happening today that matters most. Tomorrow is the first full day of IDF, when today's technology takes centre stage. Can't wait.