If Steve Wozniak is worried Microsoft is now more innovative than Apple, the root cause for that concern undoubtedly lies within Microsoft’s network of research labs. Dotted around the globe, from Redmond to India and Asia via the UK, these university-style research institutions are the quiet engines behind innovations such as the Kinect depth camera which translates human movements into computable gestures, and Xbox users’ movements into gameplay.
Another notable Microsoft product that its research arm has played a substantial role in developing is the Bing search engine — with researchers knuckling down to crack problems such as how to compute relevance and design the auction mechanisms underlying search advertising. Microsoft Research has also helped to improve the reliability of the Windows OS via the development of Microsoft’s Static Driver Verifier (which addresses the problem of trusting third-party software – and has made the Blue Screen Of Death a rarity, where once it was a running joke).
From the outside looking in, Microsoft’s research labs look like the jewel in the crown of a corporation founded ice ages ago, in technology terms, helping to ensure that, despite being the grand old daddy of tech — with a former sales chief for a CEO — Redmond continues to be a huge force to be reckoned with in many of the spheres in which it plays.
The labs are “the far seeing eyes of Microsoft,” says Andrew Blake, lab director of Microsoft Research Cambridge, giving the insider’s view. “Our job is to be a cauldron bubbling with ideas and the ideas are there to be plucked out at the right moment,” he tells TechCrunch.
“It’s sort of intrinsically difficult to predict what’s going to be important, so that’s why you have the cauldron bubbling, because let a thousand flowers bloom, let’s just see what happens. You genuinely don’t know what the outcomes are going to be.”
Microsoft spent a whopping $9.8 billion on R&D in its 2012 fiscal year but Blake says the labs account for “a small fraction” of that. “We don’t publish our budget but it’s a small fraction of the total spending on research and development,” he says. “I wouldn’t know how to spend [$9.8 billion]!”
On a press visit to Microsoft’s Cambridge Research lab, we are shown a glimpse of the huge variety of research projects bubbling away underneath the quiet corporate facade however modest its budget: from projects using machine learning to harness the power of big data to make better predictions about the Earth’s climate; to research into new user interface mechanisms that blend the real and the virtual so you can ‘hold’ a 3D ball of pixels in your hand; to a PhD project recycling Kinect components to fashion a wrist-mounted glove-less finger-motion-capturing device (below); to multidisciplinary research looking at making biological cells programmable using computer software.
If there’s a unifying thread connecting all the diverse projects going on under the Microsoft Research umbrella, it’s the sheer variety of research work being undertaken. This is not a model of corporate research tightly tied to product teams and immediate business aims, as is the case with Research at Google – which has a stated goal to “bring signiﬁcant, practical beneﬁts to our users, and to do so rapidly within a few years at most.”
Microsoft Research is more akin to a university research institution, says Blake, a structure that he argues makes for a far healthier and more sustainable entity. ”It’s clear to us that for a healthy research lab you need to have a renewal mechanism,” he says. “If you simply take people who are used to doing research and being free thinkers and you put a yoke on them, like on the oxen, and have them driving the technology wagon, eventually they get tired and where are they going to get their refreshment from? Where are the new ideas going to come from? So that’s why we have this as an integral part of our structure — right in in our DNA is basic research, and publishing, and going to conferences, and free association with the academic community.”
Blake notes that he has recently finished organising an academic conference in his own area of expertise — computer vision — adding that: “We senior people in Microsoft research, we take our turn doing those things and we publish a lot in those conferences and we have researchers visiting us from other universities and we visit other universities. There’s a lot of that stuff going on which is not that different from what you’d see in a university.”
Of course there are important distinctions to a university. For one thing Microsoft Research is privy to vast quantities of business data — which it can use to its advantage as a research aid. Instead of having to build a mini datacenter, say, to test research into improving the efficiency of data centers, Microsoft Research staff can “go and talk to the people who run the Azure business any time they want and try their ideas out and see if they’re scratching the right itch,” as Blake puts it. (And yes, the lab is working on a research project aimed at improving datacenter efficiency.)
So researchers certainly have relationships with product teams at Microsoft — but products being developed by the business do not limit the research work being undertaken, according to Blake. Information and ideas flow both ways.
“We may get a product group saying look we have got to develop this thing in a set time frame, are you going to help us? And mostly people are pretty keen to try and we find out whether we’ve got anything to help. The business goals come from the business; we are not business people here, we are researchers,” he says.
And then from the other direction: ”We go out there quite a lot and sort of sell our ideas [to the business] but it doesn’t bother us if the ideas aren’t taken up immediately because we kind of think maybe it’s not the right moment,” says Blake. “Business has its own cycles and you can’t do everything in business; you have to focus on whatever is the issue of the day. So it doesn’t put us off if we’ve invented something that we think is great and the business is not quite what they need at that moment.”
In the case of Kinect, says Blake, the Cambridge lab responded to commercial pressure from the business to develop the product by drawing on relevant bits of (in some cases years-old) research to see if they could be made to, well, connect — and that research ultimately went on to form the technological foundation for the commercial product.
The Kinect people approached us and because we had ideas at our fingertips we were able to pluck one off the shelf.
“[Prior to the idea for Kinect] we were looking at all kinds of things speculatively, some of the things we never thought they would particularly make products,” says Blake. “But the Kinect people approached us and because we had ideas at our fingertips we were able to pluck one off the shelf – the one that we thought would fit – and it did. And the solution actually surprised us. We had these ideas at our fingertips. We didn’t think those ideas were good for this problem but then we were really under pressure, which we were because there was just a year to work with the Xbox team developing solutions, so we had to place a bet.
“We ended up putting some quite surprising things together but they were things that were in our background and that we had been playing with over years. It would have been no good if somebody had said play with those now. It has to be part of your research experience that you have all these things either at your fingertips or at least in the back of your mind.”
There is one clear influence the business has over the research labs: the type of researchers they choose to hire. “We’re probably not going to hire some analytical chemists because we can’t really see at the moment how that would really impinge on the business – not to say that it’s impossible — but we don’t go out to hire a lot of analytical chemists,” says Blake.
“We hire a lot of people around some of the core disciplines of computing and some of the fringe disciplines of computing and sometimes we go almost outside computing altogether — as with our computational science group, where the primary goal they’re doing is actually the science. But the link to the business is that they’re power users of computation tools, and often their users are stressing our systems so hard that new things get invented. So we have this cluster of areas where we hire expertise that is very broadly related to the business. But then we fire the starting gun and these guys go off and you don’t know what they’re going to come up with.”
Asked which of the current projects going on in the lab he considers most promising, Blake is unwilling to play favourites. “You’re asking me to choose between my favourite children – I cant possibly do that,” he jokes.
“A lot of the ability to do good research is not just deep analytical thinking, which is more how the public probably thinks of research, but with the exercise of good taste — it’s as much about what you choose not to look into, as what you choose to look into,” he says, echoing the Steve Jobs product mantra that ‘deciding what not to do is as important as deciding what to do’. “Opportunity costs, what looks promising, people use their gut instincts to choose things which they think are going to be exciting. That’s why it’s so critical that I hire the very best research staff because it’s that good taste that is one of the things that you’re bringing into the organisation — so I genuinely would find it very very hard to say what’s going to blossom.”
He is willing to touch on promising areas of research — machine learning being a discipline he believes will play an increasingly important role in building new generations of software systems. Machine learning techniques are already being used to build products — such as the Kinect gesture recogniser (which can determine whether you’re raising your elbow or your knee), and to power the Xbox’s recommendation engine for games, TV and movies (which crunches your viewing data to predict what else you might like). But in an age of big data and increasing complexity, machine learning technology is becoming an imperative for more and more applications.
“One of the very early lessons from artificial intelligence is that programming intelligent behaviour is just too hard — you just can’t capture it,” says Blake. “What’s better is for the software to develop in the way that humans learn, the way animals learn: by example. You show them things and those things get generalised and those generalisations become the software – you don’t actually write the software, not entirely. The critical bits get built automatically through these learning programs.
“We have a group here that does machine learning — it’s about one-fifth of the lab — and now those ideas are sort of spreading outside that group.”
In the future maybe what Microsoft will be in is software for generating biological structures, it’s too important for us to ignore.
Specifically, says Blake, machine learning researchers are collaborating with researchers who design programming languages — to explore how software can be developed that can learn and understand uncertainty. ”Now what we’re doing is writing programs which instead of just adding numbers together or dealing with strings actually reasons about probabilities and will estimate how likely things are,” he says. “That’s quite a fundamental capability that we’re pioneers in.”
Asked to look further afield, to consider what Microsoft might be in 10 or 20 years’ time, should it still be around by then, Blake is quick to point out there is no way to know exactly what lies ahead, however farseeing the lab’s eyes or deep and rich its cauldron of ideas. But he does point to the “interface between computing and biology” as a “fascinating area” — and one Microsoft Research is “very involved” with now.
The multidisciplinary nature of this work means researchers with computer science backgrounds are teaming up with biologists. Or, in the case of Microsoft Research principal researcher, Luca Cardelli, have switched their focus from designing programming languages to trying to use computational thinking as a way to unlock biological mechanisms like cell division.
“What Luca and his collaborators have done is they’ve opened up that mechanism a bit further to show a bit more of the detail. But the insight they’ve got has come from computational thinking, if you like, having computational processes and analogy available to express what the cell is doing. And extraordinarily they just published the theoretical paper and at the same time a practical paper. An experiemental paper came out which showed sort of exactly the same thing — but in an experimental setting — so that’s quite a landmark piece of work,” says Blake.
“In the future maybe what Microsoft will be in is software for generating biological structures; it’s too important for us to ignore. We have no idea at the moment whether it makes a business,” he adds. “Some of the things we’re investigating seem way off any kind of business, but who knows whether they might be part of Microsoft’s business in the future.
“I think it’s pretty clear that in 20 years time the intersection of biology and computing will be a big thing… It might be that people are designing drugs by writing programs. Designing them from the ground up and making them out of DNA. They’d just send the programs off to be compiled; the way they’ll do that is they’ll just send them across the web to someone who produces DNA.”
Designing fragments of DNA certainly feels about as far away from churning out the next iteration — or even the next generation — of consumer technology as you can imagine a technology company could be. But Microsoft Corporation is undoubtedly a far stronger, future-proofed business for having such a far-sighted, far-reaching focus.