The Watson Disease: Google and TPUs
June 9, 2017
I think IBM Watson has performed a useful service. Big Blue demonstrates that writing about future technology and its applications can send useful signals to investors, competitors, and analysts.
The Watson Disease is, to my way of thinking, a weird combination of marketing hyperbole and fanciful thinking. The outputs in the form of articles, interviews, and marketing collateral are entertaining and sometimes fun.
One example I spotted appears in “Cloud TPUs: A Chip to Make Every Business as Smart as Google.” The headline assumes that “every business” is less smart than Google. It follows that the less smart will want to be as smart or smarter than Google. It seems to me that Facebook has been zipping along quite nicely. So has Amazon. Too bad that these firms are one which the real journalists at PC Magazine include in the category of firms which want to be “as smart as Google.”
The guts of the story focus on Google’s response to the strong market uptake of Nvidia technology for assorted smart applications. AMD is working to catch up, but it seems to be cornering the bitcoin mining niche while Nvidia is capturing hearts and minds across applications.
Google wants to be the big dog with its TPU or once secret Tensor Processing Units. These confections perform magic when it comes to one trick pony machine learning functions. TPUs are specifically engineered to do artificial intelligence stuff.
The write up reports that:
Cloud Tensor Processing Units (TPUs) are part of a trend toward AI-specific processors, and for Google in particular these cloud-based TPUs are the underlying computer element driving a top-to-bottom AI rewrite fundamentally redefining how Google’s apps, infrastructure, software, and services function by building intelligence in from the ground up.
That’s quite a statement enriched with some amazing words like “in”, “from” and “up”.
The Google wizard explains that Google’s approach has been to build “a kind of drag racing car.”
When will I be able to drive this race car?
Well, there’s the promise of “soon.” Just like IBM Watson’s takeover of the smart software sector. “Soon.”
The car analogies provide a metaphorical anchor for the TPU revolution.
The exemplary use case explains that a Japanese used car dealer creates Web pages automatically.
Yep, used cars.
But there are more application opportunities. I learned:
“There are limits, it’s not magic, but it’s really exciting how many places it’s applicable and in how many businesses it makes sense,” said Hölzle [a Google wizard]. “We’re aiming to be the cloud platform for machine learning and analytics. We’re making it much more accessible to average companies because it works across so many circumstances, from AlphaGo and data center cooling optimization to image and speech recognition trained on the same neural network.”
An interesting nugget finds its way into the sci-fi vision. I highlighted this statement:
Once you have a machine learning project, 10 percent of time is spent on ML and 90 percent is on data preparation, cleaning, interpreting results, and iterating to find better models,” said Hölzle. “We have something in Cloud ML to automatically try out multiple models to find what works best. You may have to pay more because the training step is hundreds of thousands of times more compute power, but you get the optimal model and higher accuracy just by pushing a button and waiting four hours.”
I think this means that human grunt work is required. Where’s the smart software? Well, not yet. I like the “pay more” phrase too.
I wish I could just ask Watson. In the meantime, I will check out the Nvidia technology. I fear there is no antibiotic for the Watson Disease. Scary.
Stephen E Arnold, June 12, 2017
Comments
-
- Subscribe to Beyond Search
Feature archive
News archive
-
