AI EVERYWHERE?

Dick Pountain /Idealog 359/ 11 Jun 2024 09:48

A few weeks ago I  'attended' an interesting webinar organised by IT security firm Sophos in which one of their researchers, Ben Gelman, demonstrated how someone with no knowledge of programming or web design can construct a convincing e-commerce site complete with product pictures,  audio, even video for around £4.23 in 8 minutes. He did this live on screen using GPT4, Stable Diffusion and a couple of other free tools.Then he dropped the bombshell: by  adding just a tiny knowledge of HTML, a couple of lines,  he created from this an agent-based framework capable of churning out hundreds or thousands of similar sites from simple text templates.

The web will soon  be swamped by such sites since  it takes a lot longer than 8 minutes to take one down, or they may even disappear themselves once a quota of suckers have been hooked. It’s happening already, we’ve recently been ‘fleeced’ by items of clothing and pottery that bear little resemblance to AI-augmented pictures on certain sites.

Last week we went to a restaurant we’ve enjoyed for years and suffered a revolting, shoddily prepared meal. Went to Tripadvisor and found hundreds of still gushing five-star reviews, so on a hunch I filtered for one- and two-star reviews and found a few dozen accurate and mostly temperate complaints about drastic decline (‘What have they done to….’). Do all those five-stars mean most customers have no taste buds? Or were they paid for? Or are they AI generated? The fact I even ask that question speaks volumes…. Once it becomes known that you can use AI to prop-up dodgy businesses  the technology has arrived in the mainstream. 

On a completely different note, Apple just announced its deal with OpenAI to integrate ChatGPT into its operating system on all  devices including Siri. Add to this similar projects from Meta and Google and it’s clear that it won’t be long before ‘AI-query’ becomes a commodity service on the same level as internet access and telecommunications, with transactions in the billions and trillions. Ideally they’d all merge into a single ‘information utility’ but of course that ideal is quite impossible to realise for several very serious reasons. 

Reason one is intellectual property. All those streams of content are owned by different, competing corporations whose only rationale is profit rather than public education. That hare has started running already with Scarlett Johansson’s suit against OpenAI for illicit use of her voice.Reason two is that even were the  AI vendors to get all necessary permissions to use other people’s content to train their GPT models, that content is going to become polluted at an exponentially-increasing rate by the gibberings of billions of dodgy websites created by their own customers.    

Reason three is the killer though. We’re all now aware of the colossal amounts of compute power needed to train and deploy GPT systems. Given present technology it’s quite impossible to train or fully query such a system on your local computer/phone/tablet (at the so-called ‘edge’) so these services will remain mostly cloud-based for the foreseeable future. Advances in analog-based AI processors and similar technologies can reduce telecommunication bandwidth requirements to some extent by more ‘edge processing’, but cloud AI servers will still consume as much of that old-fashioned utility, electricity, as a medium-sized African nation.

This is all happening during a world-threatening climate crisis which most sane commentators agree requires us to find cleaner ways to generate electricity and equally importantly to use far less of it. AI companies are already starting to worry about  where all that electricity is going to come from and the Wall Street Journal recently reported that OpenAI is in talks to buy “vast quantities” of energy from the startup nuclear fusion company Helion, in which CEO Sam Altman has invested $375 million. Fusion power occupies an ontological niche rather like that of quantum computing, somewhere between hope and reality, real-soon-now-perhaps… 

And then there’s the question of how to pay for all this juice, which leads into the realm of blogger Cory Doctorow’s concept of ‘enshittification’. He summarises his caustic take on tech evolution thus:

“Here is how platforms die: First, they are good to their users; then they abuse their users to make things better for their business customers; finally, they abuse those business customers to claw back all the value for themselves. Then, they die.”

Read his argument at The ‘Enshittification’ of TikTok. Corporations like Apple and Amazon didn’t spend the original AI research money and hence have to pay for it now, by buying the AI companies or by paying some kind of rent. The cost is so substantial they must get it back from their customers. The days of freebie services are numbered. 

[ Dick Pountain thinks it will all end in tears]