Columns

Computers are still remarkably useless at performing some simple, everyday tasks. Unless I've assigned verbal tags - metadata - to my photographs, if I ask Aperture to find me the photo of an arc of daffodils, it has not a clue what to look for. I can hum it a melody, but iTunes is stumped as to which track might contain my riff. Whether or not I've stumped up for QuickTime Pro, my Mac is equally hopeless at finding a scene of breaking waves among a library of video clips.

Small inroads into pattern recognition by machine have been made, but they remain disappointingly small. Your Mac might recognise your fingerprint, but MacPickPartnerOutFromShoppingCrowdX hasn't reached alpha. At least some of this is due to the way our brains recognise meaningful structures and familiar objects within non-verbal media: getting an image analysis application to distinguish faces from breasts is amazingly tricky, yet newborn babies get that straight quite quickly.

A lot of bright minds, and lots of funding, have been brought to bear on these problems without much progress. Automatic image analysis can chop up simple pictures and identify regular features - for example, to read car number plates. Some approaches are based on the way the brain is perceived to work, by modelling nerve cells and their interactions. Although academically fascinating, there's a reality gap, as we're similarly ignorant of the way the brain works. 'Neural networks' may sound impressive in a software

ADVERTISEMENT

specification, but could prove illusory.

Long and sometimes bitter experience tells us the ideal answer is to utilise the best properties of humans and computers in symbiosis. Expecting humans to perform repetitive, error-prone tasks while computers wrestle with complex pattern recognition, is a sure way to find how easily both can err. Yet that's what a lot of security software does when it throws an alert asking, for the hundredth time, whether we want to authorise something it recognises as potentially hazardous. Human factors experts have long recognised this as a disastrous model for human-machine interaction, with tragic evidence from air accidents as evidence.

Apple's delightful parodies of Windows Vista security measures aren't just good advertising. They're an object lesson in interface design; indeed, the very philosophy of computing. It's mere blamestorming to assail us with alerts asking questions we can't answer. 'So when the bogus blogsite invited you to click on the Trojan, and the warning came up, you clicked on Allow?' So it's your fault that the network got hit! If the operating system has warned you, everyone from the system administrator up can pile the blame on you.

But what's most frustrating about so many security warnings and other trappings of shallow security is that they give no information. I'm being asked to make a decision that the computer recognises is potentially important. Yet it isn't prepared to trust me with the merest morsel of reasoning behind its concerns. My human skills are being ignored by a system that struggles to distinguish between friend and foe, and couldn't tell who typed in a password, as long as its characters were correct.

I hope Apple staff watch its adverts and discuss what lies at their root. Macs should empower users in every task. At no stage should my computer take on tasks it can't do at least as well as I can, and at no time should it foist responsibility and ultimately guilt onto me if it isn't prepared to give me key information to help me in making a choice.