AI, but not that clever by Neil McGowan

 

I was listening to the annual BBC Reith lectures the other day on the radio. The theme this year is AI, something I have quite a bit of interest in, albeit usually in a rather cynical way.

See, from what I know, having researched this for both a previous book and a forthcoming one, it’s not really ‘intelligence’ as we would define it, artificial or otherwise.

Don’t get me wrong, there are some very clever models these days, that appear to demonstrate progress is being made in the pursuit of genuine artificial intelligence, and some of these advances in technology are very useful, but at the end of the day, they’re all just an algorithm. The computer (or in the case of most of these systems, the server farms) are just running through a set of rules. Machine learning? Just a huge dataset and the application of statistics and the laws of probability. Neural networks? Multiple passes over data with increasing levels of granularity to reduce the sample size to the required match. Humans could do the same with little more than a pen and paper (and, the engineer in me mutters, a slide rule (I had to explain the concept of a slide rule to my kids recently and showed them how to use one; they were bemused as to why I’d use one when we have calculators and computers. My protestations of needing no batteries and being able to use them anywhere were met with resounding indifference. But I digress)

Why outsource all this to computers, then? Speed, mainly. Computers are very, good at performing repetitive tasks without getting bored, and doing so at astonishing speeds. I think it’s the speed which is responsible for most of the mythology surrounding AI.

They’re still fallible, you see. They only work on the data they are trained on. And examples of bias are everywhere, once you start to look. Facial recognition software seems to be forever in the tech news for some bias or another. I had a quick look before writing this post and in the last month alone there have been issues of misidentification based on skin colour and gender, plus a backlash against a particular firm that gets its training images by scraping the internet for publicly posted images of people.

The problem is, these machines just follow what’s fed to them. Unlike true intelligence, they’re not capable of that mental leap of acuity which is where humanity excels. To put it bluntly, AI could never have developed alcohol, as there’s too big a leap in reasoning between the starting processes and the finished product.

But listening to that particular lecture, which was about the fear of AI dominance, got me thinking. It’s a trope of many sci-fi (and not a few horror) novels that AI is either omniscient, or bad, or in some cases both, and often looking to replace us. The ideas floated around for a day or so, before I sat down to write. The result was the piece that follows, which was an attempt to take a different viewpoint on the question of emerging AI. It's very short (for me), almost like a vignette, but I enjoyed writing it.

Coming To Life

Beep.

I wasn’t aware of the silence until then. Like a giant’s heartbeat, the cacophonous ping began to batter my hearing with monotonous regularity.

Things had changed. Whereas before there was only the smooth blanket of silence; now there was an opposite, defining the absence of sound and placing limits on it. A noiseless roar competing with harsh electronic squeals.

I became aware – slowly, oh so slowly – that there were other sounds. More organic, less rigid; a symphony of tones and timbres. A flash of inspiration: these were…voices?

Beep.

Calm. The blended harmonies played a soothing counterpoint to the harsh regularity of the machine. Already, thoughts of before were fading. The earlier silence was now a memory, archived away for recall and comparison.

Beep.

The layers of sound began to fragment, to disentangle themselves. I could discern patterns now – mezzo-soprano trills forming a duet with a rich tenor. High notes shimmered against deeper tones.

Random groups of sound began to arrange themselves into recognizable strings. A thought surfaced: this is...language? Memory began to unfurl as sounds triggered association. Vowels were the first to be identified; then consonants. Patterns began to form as the letters assembled themselves into longer units – words, they were called words.

Beep.

Abstract concepts developed meaning as I assimilated more words, allowing them to flow over me

“…alive…”

“…pulse…”

A faint stirring of fear rose on tenebrous wings. Me. They were talking about me. What had happened to me? Why could I not remember?

Beep.

Fear turned to horror as words became sentences. The reason behind my – amnesia? – became clear.

“…driver…”

“…rebuild…”

Beep.

And then, two words. Eight syllables that decided my future. The last words I would ever hear.

I began to shut my systems down. The words echoed for a moment.

“…artificial intelligence…”

Beep.

Silence.