Why couldn't Alan Rickman be the voice of my iPhone? Photograph: AFP/Getty Images
In an unjustly dark and dusty corner of YouTube is a badly compiled video slideshow of Alan Rickman in various stages of monochrome moodiness. It has the look of something slapped together late one night by a distracted student who should have been revising and YouTube is hardly the most salubrious home for Shakespeare's Sonnet 130.
But Rickman's reading is heavenly – he oozes through it:
"My mistress' eyes are nothing like
the sun,
Coral is far more red than her lips'
red;
If snow be white, why then her
breasts are dun;
If hairs be wires, black wires grow on
her head … "
I found it by following some piece of internet flotsam or other – one of those things you end up doing online when you're almost certainly on deadline. But it came to mind again when the female voice of Siri – the voice control feature on Apple's more recent iPhones – was tracked down recently in the US. It's a woman called Susan Bennett, who has variously voiced a cash machine, phone systems and help messages for Delta Airlines.
In the UK, we are more familiar with Jon Briggs, who started as a technology journalist before ending up doing voiceovers. Both endured lengthy recording sessions of more than 5,000 sentences several years ago, with no idea it would all be deconstructed to populate a voice control system that upwards of 100 million people would use.
Which rather led me to think … couldn't Rickman be the voice of my iPhone? If Han Solo can voice the TomTom satnav, why not? Siri is certainly faster now it has graduated from its two-year beta, and connects to properly useful stuff, including Twitter. (Useful being subjective, I appreciate.)
All the merry punchlines planted by Apple engineers are still there. Imagine Rickman's audible nectar: "I love your voice." "Thank you. It's too soon to try out for X Factor, but some day, Jemima."
"Talk dirty to me, Siri." "I can't. I'm as clean as the driven snow."
"Are you hot?" "We were talking about you, not me."
Alan is more than mildly diverting of course, but Siri is an infant in the world of artificial intelligence. Spend any serious amount of time with it and it clearly doesn't do what a personal assistant needs to do – it doesn't learn. Siri is unlikely to be quicker than Googling the information yourself. Learning and then predicting behaviour is where these technologies really start to make a difference.
The research team that came up with Siri, and sold the product to Apple in 2010, have another project brewing. SRI, based at Stanford Research Institute, has created Bright for stressful, highly computerised environments such as disaster response or cyber security, where fast and accurate response is critical. Various cameras and interfaces analyse touch, gesture and facial movements and learn the user's behaviour – which programmes are used, which emails are read, which mechanical processes are repeated regularly – and then starts to create shortcuts. That might be highlighting an email that it knows multiple other users in the organisation are also grappling with, or making the most frequently used piece of software easier to access.
If you typically pick up a bottle of water and switch to Spotify for a break, it might open Spotify when it sees you pick up a bottle of water. It's not creepy; it's automated. It's a mirror, but one that can pre-empt what you need.
Bright is designed for emotionally pressurised environments more demanding than those most of us work in, but the technology, and certainly the benefits, will filter down into consumer technology just as Siri did. Siri itself started as a digital assistant for soldiers in the field.
The issue is "last meter bandwidth", which describes how we, the user, become the bottleneck between the processing power of the computer and what it is capable of outputting. We spend cumulative minutes, even hours, completing repetitive tasks when machines are capable of taking the strain.
From a computer science perspective, learning the behaviour of a single user is tough. This is the small data problem; unlike big data, where patterns and trends easily emerge, individual human beings can be unpredictable and can change behaviour, which is not helpful for pattern-hunting algorithms.
That inconvenient human unpredictability was what caused the "inaccuracy" in a separate statistical project called Far Out, led by two researchers from Google and Microsoft, which could predict our future location. Even years down the line, and taking house moves, marriages and new jobs into account, our behaviour is usually predictable enough that it can be learnt, and then statistically modelled. If that's one tiny project by two researchers, we can only begin to imagine the kind of behaviour and location modelling being developed and deployed by the military.
The promise has always been that technology can relieve us of the mundane, repetitive chores, yet our experience is usually that those are replaced with new repetitive tasks that ensure the more mundane ones can happen by themselves. And if we do free up some time, we stuff it pointlessly with more "productivity" rather than reclaiming any real life.
Perhaps by the time true artificial intelligence filters down, we will be a little more sophisticated in the limits we set ourselves and let the machine take up the slack. Not only could it pick up the grunt work, but our own domestic Bright will be able to surprise us too. It will build in the delight and relief of serendipity and surface some of the flotsam of the web that we enjoy. A little of Sonnet 130, possibly.
credits to Jemima Kiss
