AI will accelerate the ‘kippleisation’ of the Internet
You’d think ‘journalist’ was a fairly easy job to understand. Unlike, say, ‘innovation evangelist’ or ‘scrum master’, the role is pretty clear: find things out, then write them down. You’d be wrong, though. A relative recently phoned me to tell me his great idea: I should take every piece of low paid writing work going, and then set the artificial intelligence (AI) ChatGPT to work doing it.
I’d be quids in, he thought. AI was going to destroy jobs like mine, he said, but if I got in early I’d have the advantage, for some inexplicable reason (the fungible nature of the commodity ‘content’ he was proposing seemingly lost on him). In any case, leaving aside both the implied insult (I do not write ‘content’) and my lack of interest in joining a race to the bottom, there is one other slight snag with this cunning plan: despite the acres of breathless coverage in recent weeks, AI is crap.
Here is the truth: even ChatGPT is not as amazing as people seem to think. Its writing is turgid, it is prone to serious errors, and its knowledge is a kilometre wide and a nanometre deep.
However, it is just good enough to be used to produce more of the useless dross in which we are drowning the world: such as spam e-mails and SEO-led digital marketing nonsense designed to be read not by humans but machines. Indeed, news that Buzzfeed was planning to have its quizzes written by machines tells you more about the quality of Buzzfeed than of AI. Investors still loved the news, naturally.
Large language model AI probably won’t stay bad forever, of course. That is the way of technological development. It is possible that AI could turn out to be a dead end that doesn’t scale, but the more likely outcome is incremental, logarithmic improvement. Think of it this way: the Intel 4004 would seem pretty laughable today, but it started us on the road to today’s virtually unlimited processing power. Who knows? It may even turn out to be exponential.
Let’s hope that by then Silicon Valley has cracked the problem of getting AIs to buy and pay for things. After all, the robotic factories cranking out goods to be delivered by robotic trucks still need to be bought by someone – and if everything is done by AIs then that won’t be us.
For the moment, however, the level of both the writing ability and the knowledge demonstrated by AI is low. The fact that academics have reacted with horror, featuring that students will use it to ‘write’ their essays, speaks volumes about how low our expectations of university students are.
Unfortunately, low standards have never been much of a barrier in many areas of life. E-mails from overpaid and underemployed managers, for instance, could easily be written by an AI. Indeed, an AI’s horrible prose and propensity for earnestly stating falsehoods and inaccuracies as facts would likely be an advantage, or would at least represent normal operating procedure. In effect, bullshit jobs that not only don’t need to be done but actively, albeit in a small way, make the world a worse place are ideal for automation, especially those that fill it up with nonsense.
And, never in short supply, you can be sure that a tidal wave of nonsense is coming.
In his novel Do Androids Dream of Electric Sheep, Philip K. Dick mused on ‘kipple’ or rubbish, which acts as a metaphor for civilisation’s decline: the character JF Isodore observes that kipple not only exists but somehow reproduces itself, expanding to fill any and every space:
Kipple drives out nonkipple […] No one can win against kipple, except temporarily and maybe in one spot, like in my apartment I’ve sort of created a stasis between the pressure of kipple and nonkipple, for the time being. But eventually I’ll die or go away, and then the kipple will take over. It’s a universal principle operating throughout the universe; the entire universe is moving towards a final state of total, absolute kippleisation.
The Internet is already drowning in this nonsense. In kipple. You have certainly noticed it: how, increasingly, headlines over-promise and under-deliver, how dubious websites game search engines, how e-commerce sites push sponsored results rather than what you actually want to buy, and how ‘how-to’ videos take 10 minutes to get to the point, possibly including ones about how to escape flaming, overturned supercars (in case you’re interested: many come with a combined knife and hammer in the dashboard).
Driven by algorithmic feedback loops, the kipple exapands, multiplying at the cost of all else. For its part, Google, which once delivered us from kipple, has so succumbed to filling up its results with rubbish such that it is beginning to resemble Altavista circa 1996: usable, but only just.
Alas, as anyone who has ever turned on a television set knows, the horrible truth is that ‘bad’ is usually good enough. Sometimes it’s significantly more popular. Most pertinently, it is almost always more lucrative. As such, even in its current primitive state, AI has a bright future.