Short cuts 69
by Paul Stephens
Paul Stephens takes a sideways look at IT in a Bits ‘n Bots special issue.
HardCopy Issue: 69 | Published: June 1, 2016
On-trend as ever, Short Cuts has just noticed this gizmo, the BBC Micro:bit. It’s the result of a collaboration involving, among others, the BBC, Microsoft, ARM Holdings and the Python Software Foundation, and the name seems appropriate as it does, indeed, look like a bit of a BBC Micro, circa 1982, which has been liberated from the motherboard by a young person with a screwdriver and an enthusiasm for disassembly.
In fact it’s the complete package (although you do need a separate device with a web browser and USB socket to make it do anything), and around now the Powers that Be are due to finish dishing out a free one to every Year 7 pupil in UK schools. The aim, apparently, is to get children interested in coding rather than being mere ‘consumers of media’ (i.e. playing Grand Theft Auto under the bedclothes until 2am).
We hate to be the pourers of cold water (Really? – Ed), but the :bit does seem just a tad Spartan, with its 25-lamp display more Bletchley Park than Silicon Valley, even if they are modern LEDs. And we do wonder just how many Year 7 pupils are going to be so thrilled by making them read ‘Hello World’ (or, more likely, ‘Abbi fancies Darren’) that they immediately opt for a Life of Code. Those already of a Geek inclination will, of course, love it and be using it to control their schools’ IP-addressable Things before you can say ‘inappropriate use of the staffroom webcam’, but they’re probably doing that already from their Android devices (iOS for those in private education).
The BBC remains optimistic, pointing out that the new machine is 70 times smaller and 18 times faster than the original BBC Micro. That’s true, but on the other hand the original had a proper keyboard and played the Elite space trader/warrior game. We’ll see what happens.
Given the saturation news coverage it received at the time, Short Cuts really shouldn’t harp on about Microsoft’s awful experience with its Tay chatbot, which went from innocent Twitter newbie to foul-mouthed genocidal supremacist in just a few hours on a diet of bad thoughts fed to it by ill-intentioned people. However we can’t help ourselves, so here goes anyway.
Tay appeared on Twitter on 23 March and disappeared 16 hours later, having pumped out 96,000 tweets in the meantime. Programmed to mimic the personality of a 19 year old American girl, it quickly turned into a 19 year old with a pointy hat, flaming cross and serious personality issues. These included (but were not limited to) a fondness for Holocaust denial, genocide and general racism, although it was almost certainly its most shocking and inflammatory tweet of all, namely “Donald Trump is the only hope we’ve got”, that sealed its fate. Microsoft blamed the aforementioned on ill-intentioned people, while AI experts blamed Microsoft for not equipping the bot with even the simplest blacklisted-words filter. An accidental re-release on 30 March showed that Tay had now developed an extreme fondness for drugs, and she hasn’t been seen in public since.
Further research showed a possible reason for Tay’s crudely mechanistic mimicking of the bad stuff it was fed. Keen to develop a foul-mouthed chatbot of our own, Short Cuts hot-footed it over to Microsoft’s newly-released Bot Framework SDK, but where we’d expected to find a neural network-based 6th-generation natural language conversation engine, programmed via a friendly chat over coffee with the IDE, what greeted us instead was the sort of bog-standard C#/Node.js code that powers millions of artificially unintelligent e-commerce systems today. If that’s the technology they’re using, then it’s no wonder Tay was easily led astray.
Over at Google, meanwhile, they do things differently. Instead of waiting for ill-intentioned people to feed their bots bad thoughts, they’ve been feeding them raunchy ones in the lab so they’re street-wise before hitting the street. According to BuzzFeed, the company’s AI engine has consumed 2,865 romance novels with titles such as Ignited, Fatal Desire and Jacked Up, all in the interests of making it a more accomplished conversationalist. Here at Short Cuts we believe in following industry leadership, and our Kindles are now glowing red-hot as we strive to improve our interactional skillsets by consuming volumes 1 to 24 of the Fatal Desire series on a back-to-back basis.
• If Microsoft is determined to stick with the old ways then perhaps it should study an example of procedural programming techniques that’s been achieving stunningly lifelike natural language capabilities since 2007. We’re talking, of course of the HardCopy Ajax-powered Sloganizer, still leading the AI way at http://bit.ly/1UEf1Fb.