FAKING FRIENDSHIP

Dick Pountain /Idealog 351/ 07 Oct 2023 02:08

“Man told to kill the Queen by his AI girlfriend!” History may some day recognise that as the most perfectly-formed tabloid headline ever, but right now it’s merely a symptom.To put it another way, it’s an answer to the question ”what do you get if you cross a moral panic with a feeding frenzy?” I’ll have to admit that I hesitated before embarking on yet another column on AI, which I’ve been covering for over 40 years, right from the days of Lisp, Eliza and ‘expert systems’ up to creating pictures of yucky seafood dishes with Stable Diffusion. A couple of events forced my hand though, convincing me that evading the topic isn’t an option this month.   

The first of those events was the CogX Festival 2023. I first covered this event back in 2019 when it was held near where I live, at the newly-opened King’s Cross Granary Square site. Around 10,000 people attended some 500+ sessions given by leading AI techies from the UK and USA. The audience was overwhelmingly young, 20-somethings wearing backpacks and plaid shirts and either already working as AI developers, or wanting to. I was given a press pass and heard genuinely revelatory talks about deep-learning convolutions and IPU (Intelligence Processing Unit) architectures. 

During the Covid years CogX continued via Zoom and was inevitably far less exciting, but I still ‘attended’ perhaps a dozen sessions. CogX 2023 has just been held in the 02 stadium and attended by around 90,000, hearing over 1000 speakers on 10 stages, including celebrities like Stephen Fry and the Queen of Jordan. I didn’t even bother applying for a press pass and the ticketing structure was quite byzantine - you could get a free pass to get in, but then had to purchase ‘add-ons’ to actually hear anything, which could amount to anywhere from £500 to over £1000. I’d guess that’s simply because the emphasis is now on attracting investors rather than recruiting plaid shirts: AI has arrived, applications and commercialisation take precedence over research and development. 

The other significant event took place at home where, among several old friends we had to dinner were a director of an important science picture library and a professional photographer for a large media corporation. Both were apprehensive about the effects of AI on their businesses. The main AI large-language models have already ‘scraped’ the entire contents of R’s picture library, and there’s currently nothing they can do to prevent this. J travels the world photographing hard-to-capture events: a picture of his that once went viral was of a huge wave hitting a lighthouse in which the face of Poseidon appeared among the foam. He pointed out that nowadays anyone can replicate such a picture in Midjourney in five minutes without leaving their sofa. 

CogX to its credit still records the festival sessions and releases the videos for free on YouTube after it’s finished, so I went through the list of videos looking for in-depth technical ones that might spark a  future column. I found very few of those, but did find one that was highly relevant to that question I posed above – Gillian Tett of the Financial Times interviewed Yuval Noah Harari, author of the best-selling popular philosophy/anthropology book ‘Sapiens’.

Harari’s take on our current situation is that the first wave of social media sought to capture our attention by using clicks and likes, and sold these on to advertisers at colossal profit. But this will soon be displaced by a coming second wave of AI-driven media that instead solicit our affection by mimicking human feelings. AI agents will in effect give back attention to the alienated, isolated millions who believe that no-one is listening to them. This is a far more dangerous matter – instead of encouraging passive scrolling and shopping, it may recruit people to perform actions (like killing the Queen). Harari argues that AI needs to be regulated, sensibly but avoiding a total ban that would throw away all the potential benefits, and to this end he suggested to Tett three concrete policy proposals:

 1) To gain public trust, AI regulation will require international institutions that will have to be financed by a 20% levy on tech companies profits.         

 2) There should be an absolute ban on passing off AI agents as human beings (with jail penalties like those for counterfeiting or fraud).

 3) Bots must not be granted freedom of speech, and their owners should remain legally responsible for all their utterances.

I’d probably add to these that royalties be paid for content scraped from existing sources, similar to those from music streaming services (which themselves urgently need reforming). And perhaps some kind of vetting of AI girlfriends for homicidal impulses might be a good idea too.   

[Dick Pountain was disturbed to find that ChatGPT has better musical taste than many of his friends]