Here's a technology development that could erode a modern cultural intuition regarding the difference between brains and computers. Computers, the assumption goes, are precise and fast and wizard-like in going through data lickity split and sorting it out precisely. People are sloppy and erratic thinkers, slug-like when doing most math or anything else tediously detailed, but nonetheless through sheer volume of interconnection among dumb-as-mud neurons we're wiser and more able at processing info on the fly and seeing patterns and meaning in ambiguous circumstances than any computer no matter what software one puts in it. A typical human brain deals with the fog of life by telling jokes and knowing when and how to laugh. Computers, I think, not so much.
The news is an int'l team (Rice and UC Berkeley in the US plus at Singapore's Nanying Tech. U. and a computing science institute in Switzerland) that told at a conference in Italy it has made prototypes and plans commercial use soon of a kind of computer chip that by design makes a lot of mistakes but, as with brains, that's okay. It is 15 or more times as efficient (ie, it runs cooler) and faster that conventional ones. For some uses, such as image or sound processing, the output is just fine. We can look at or listen to things that are pretty fuzzy and make them out. Battery powered e-slates and hearing aids are among planned apps. One imagines there are many others to come.
Stories:
- Houston Chronicle - Eric Berger: Hot trend in computing: Chips that sometimes get it wrong ; Good lede start, "At first blush it seems a daft notion..." ; Berger gets the lowdown on the local Rice computer scientist who has been developing - and gaining notice among colleagues - this idea for nearly a decade. We also get a sense of the close ties between Rice and Nanying Tech. University. We also learn the name of the strategy: "probabilistic computing."
- PC Pro - Stewart Mitchell: "Inexact" chips save power by fudging the maths ; Other than noting that such chips might best not be used for calculating financial trading sums, one sees nothing here beyond what the press release provided. Surely a site such as PC Pro has access to other chip designers who may have something to say?
- Gizmag - Darren Quick: "Inexact" computer chip makes mistakes, but is 15X more efficient ; Gizmag too may not have added much value to the press release except for one thing. Quick or somebody there did check the pub's own previous reports - and links to its report a bit more than a year ago on the same research's status.
- The Verge - Andrew Webster: "Inexact' computer chip is 15 times more efficient than current models ; A short press release rewrite.
Years ago a kind of software employing "fuzzy logic" made some news because it is so counter to the expectation that computers must above all be reliably precise and accurate. This week's media accounts stories follow the lead of the Rice U. press release, describing the surprise but not elaborating on the mold- and metaphor-breaking aspect of the work. Nonetheless this does appear to have cultural impacts too, making computer brains seem significantly able to be more like our own. The researchers have, I gather, mainly lopped off the circuits that checked the last few decimal places in numerical calculations. That seems a bit like not taking the time to count and weigh the fish in the bucket one by one but merely declaring "looks like a dozen to me, about ten pounds." That's quicker, easier, and may serve the purpose. Nanotechnology can make devices that are as complex right down to the molecular level as are living cells and organelles. And computer science has fast chips that make mistakes but stay largely on course. Life and technology continue the great convergence.
Such vague but high-falutin' interpretations are absent from the first burst of news coverage. It tends to follow the lead of the Rice U. Press release. Which is wow, these so-called pruned chips are sloppy, they violate dogma, but are a significant improvement anyway. The presentation, it says here, won a prize for best paper at the conference.
Grist for the Mill: Rice U. Press Release;
- Charlie Petit
