Every so often I hear someone blathering on about The Singularity. If you’re not aware of it, it’s basically heaven for atheist geeks, where the computers become so powerful they take over the world because they can design new computers more powerful than us mere mortals and then we ourselves become immortal as our digital selves exist in this digital nirvana for ever more. I have a similar dreamy future view of Watford winning the premiership, and quite frankly that’s probably more likely to happen.
Why do I hold up such a strong view in the face of people as smart as Ray Kurzweil, the poster boy of The Singularity movement? Because of every time I use my sodding mobile phone the way in which is tries to be “helpful” with its sodding autocorrect functions.
Autocorrect has two gene paths: predictive text on early mobile phones, and autotext and the like from applications like Microsoft Word. With touchscreen keyboards and all their limitations, the need to finish words before you have typed them becomes increasingly important to overcome the madness of touchy-slidey-QWERTY.
Some of the time it’s useful, but too much of the time it just ends up appearing to input random words or, as I found to my peril yesterday, substituting a perfectly servicable new email address with one that already existed in my contacts list for somebody completely unrelated.
But such a simple thing, surely, can’t compare with techno-heaven? Why am I dismissing humanities chance of eternal salvation on a memory chip because of the limitations of our current technology?
Well, because autocorrect, machine translation and a lot of the efforts of the realm known as artificial intelligence have given up the pursuit of giving machines the ability to understand, and instead have moved onto them using what they are good at, raw processing of numbers, to make best guesses on the basis of probability.
Google were one of the first to get this, with their Google Translate products. In the past, translation software tried to infer meaning to be able to accurately translate languages. Language is is remarkably contextual, and without meaning it seemed very difficult to make meaning in another tongue. Take, for example, the words “What is this thing called love?” and then think about all of the permutations of meaning (with different punctuation maybe to help you) that those six words can possess.
What Google did was to say hang all of that, and instead mine vast amounts of know translations (UN documents and so forth) and then spot sentences and phrases and try to place them contextually amongst other known texts to come up with a best guess of a translation. And this approach works a lot better than anything before – it’s very servicable. It’s not, however, in any way intelligent. Put it to task to translate the works of Shakespeare and it’ll do a good job; ask it whether Shakespeare is beautiful art and it will throw back a translation of your question.
Autocorrect doesn’t even have that pseudo intelligence. And whilst it might at some point in the future as the power of the devices we have in our pockets continues to grow exponentially (until that exponential growth, like all exponential growth, eventually flattens out) it’s never going to be able to do anything but infer from analysis rather than guess from context and meaning.