Supercomputers to reshape tech landscape
What would you do if you had as much raw computing power at your fingertips as the US government’s most demanding scientists?
These days, that question does not seem so far-fetched. The rise of artificial intelligence has been accompanied by an explosion of processing horsepower, as the companies leading this particular technology revolution have turned to designing their own advanced systems, right down to the level of the silicon.
The results could reshape the landscape in the tech industry, handing power to the companies with the most powerful hardware platforms. And they will challenge every business to think hard about how to apply a plentiful new computing resource that will soon be widely on tap.
The latest evidence of the speed at which this revolution is progressing was Google’s disclosure this week that it had leapt forward into the world of so-called exascale computing.
Building computers that can reach an exaflop, or a quintillion operations a second, has been a dream in the supercomputing world for some time. The world’s fastest supercomputer, Japan’s Fugaku, achieves only half that raw performance. The US Department of Energy is still waiting for delivery of the first full exaflop supercomputer, one of three it plans to use.
Not that Google has just leapfrogged the entire supercomputing industry — at least, not this week. Not all exaflops are created equal. Simply yoking together very large numbers of processors does not amount to a full supercomputer.
Google’s specially designed chips, called TPUs, process the signals inside the most advanced deep learning systems. The goal is not precision of each operation but the overall picture they can assemble, as billions of electronic “neurons” in these electronic brains search for the patterns in mountains of data.
But this still represents a big leap, and Google says it has several such systems already in use. It also plans to make the technology available to customers of its cloud computing division later this year. After years lagging behind Amazon and Microsoft in the cloud market, its best hope lies in fighting back with its most advanced technology for AI.
How this powerful new resource will be used is still unclear. Google once dreamt of taking on some of the world’s hardest problems. But these days, after shedding or closing a number of its ambitious “moonshots”, its research focus is squarely back on the problem loosely defined as search.
After the fist big breakthroughs that made the web easily searchable, it is easy to dismiss advances these days as more routine and incremental.
But the term obscures what are some of the most difficult problems in computer science. Understanding people’s true intent, divining underlying meaning in the world — these are profound questions, with huge implications for any company that can truly solve them. Google’s current business model may involve monetising people’s attention through advertising, but the long-term implications of a computer intelligence that can truly mediate the world are immense.
One potential application shown off by Google this week was a research project called LaMDA, an automatic language generation system designed to sustain a conversation with a person on any topic. Like OpenAI’s GPT-3, which seized the tech world’s attention last year, it makes an impressive party trick, but it is unclear how far these systems will be able to go in producing truly useful interactions.
It may not even be Google itself that comes up with the biggest breakthrough applications. Rather than just selling raw computing power, it plans to give other companies access to foundational AI technologies.
Better ways of automating interactions with customers are an obvious example. LaMDA and systems like it point the way to the next, and more effective, iteration of chatbots. These computer assistants were all the rage a few years ago, but ended up as an early disappointment of AI.
Another potential use shown off by Google this week was 3D video conferencing, using AI to create the impression of being in the same room as another person, separated only by a pane of glass.
The true breakthrough applications of all new technologies, however — the things that simply couldn’t be done before — are impossible to predict. As huge computing power becomes more widely available, the imagination of companies in every sector is about to get stretched.