Friday 30 June 2023

Collision 2023 whizzed by with plenty on AI

Collision 2023 talk with Hinton
Godfather of AI: Geoffrey Hinton... he's the tiny figure in the centre. Squint and you can see him.

I dropped in on Collision 2023 during the week. It's billed as the biggest tech industry event in Canada. 

Think of your usual office meeting. Now set it in the middle of a Madonna concert: that's Collision. 

I don't think I'm as used to so much cacophony since COVID.

The day's top billing was with the Godfather of AI, Geoffrey Hinton. The rest was filled with big sounding talks like The Future of Quantum Computing, but which really were just company sales pitches. I bet investors get the same deal. Rather than higher level or theoretical discussion, it was the limited perspective and plans of a single company. 

If I made a social media app and then billed a talk as 'The Future of Social Media', that'd be disingenuous. 

The talk itself was fine for what it was; I just felt the title advertised something the talk was not. 

Another presenter was still with Google. Now Google is in so many ways awesome, putting out products we are hopelessly addicted to. However, this mouthpiece was so peppy, positive and reassuring about everything AI it verged on soporific. 

And honestly, I've listened to car salesmen I'd trust more. I don't buy the corporate line there will be no job loss thanks to AI. 

Creative Destruction has been bringing diminishing job creation returns, while productivity boosts drive money to the owners, not the workers. GM, Ford, etcetera employed millions back in the day. The new drivers of the stock market (Meta, Google, etc) employ a small fraction. 

Corporate Mouthpiece?

The Google Mouthpiece claimed AI cannot think, it can only thunk, which while cute was directly contradicted by the Godfather of AI, who's left Google and can now say whatever the heck he likes. He's run tests with large language models and says they show reasoning abilities. Generative AI is a lot more than just advanced autocomplete. 

I'm more inclined to believe him than the happy happy corporate shill. Funny enough, it was the journalists earlier who were noting that Generative AI doesn't do well with nuance. Neither do some agendas. 

Hinton expressed concern about autonomous AI battlebots, wealth polarization, mass disinformation and other points of concern that I really should have written down because now I can't remember. Old age sucks, man.

99% of researchers are working on making AI better, and only 1% is working on making it safe and ethical. 

I don't doubt that avaricious, insatiable, manipulative and machiavellian AI being built for the stock markets and don't have the faintest clue what an ethic is, or if you can buy it. 

Hinton's talk only lasted 20 minutes or so, which is a selling point for the event; the talks are short enough there's no time to get bored. Here though, it was much too short. Honestly, I could have listened to Hinton all day. 

A lot of money was poured into the super slick Collision glitz fest, and I'm sure many a useful connection was made; hopefully productive deals, too, especially for all the startups and smaller companies. One can hope.

For me, though, the talks were disappointing, especially the corporate ones. I wanted more candor, less pablum. The most interesting talks were by gadflies who'd quit their prestigious positions to warn the world about AI, and the nuanced take of the journalists, who will one day be writing about 'our new robot overlords.'

And I for one just hope to be a part of it.

No comments:

Post a Comment