Web dev at the end of the world, from Hveragerði, Iceland

AI summaries and AI healthcare (links & notes)


AI summaries are almost certainly unreliable

It’s like the past few years' discussions on bias in training data and issues with shortcut learning in AI models just didn’t happen at all?

Like, our industry didn’t take in any of it, did it?

Read more…

AI in healthcare

What’s interesting about this case is that when the AI tool for spotting sepsis was deployed as designed, as directed by the vendor, it was essentially unusable. Too many false alarms. Continuing to use it as designed would have killed people. The tool didn’t begin to be useful until they moved it off to the side and turned it into a monitoring assistance for a dedicated team that was responsible for alerting other teams that a patient might be developing sepsis.

Read more…

​Trustworthy AI tools and companies

I’m not against using generative AI tools but, AFAICT there are very few trustworthy actors in the field: OpenAI, Facebook, Google, Stability AI, and Midjourney all have a track record that’s dodgy to say the least

The way they’re rushing into AI doesn’t add to the trust either

Anthropic, which is an AI startup focused on safety, pivoted to be about generative AI in their recent round of funding. And, as it turns out, a keyword in their safety research is “alignment” which is oogedy-boogedy speak for “let’s make sure this hypothetical god-like AI we think we’re fumbling into existence will be on our side.”

“Safety” as used by these people is a pseudo-religious catechism and they aren’t genuinely interested in reducing the harms done by these systems.

I’d like at least one trustworthy actor, who behaves with at least a modicum of decency—at least attempt to wear a fig-leaf of ethics—and is genuinely trying to develop these systems as productive tools for modern society.

And, no, Microsoft isn’t it. Are you kidding me?

I don’t think it’s controversial to say that between the gig economy, Amazon’s labour abuses, the surveillance economy, and crypto, the public in general has plenty of reasons to both distrust tech companies but also to distrust the notion that they know what they’re doing with AI and can do it safely

In fact, I’d say that we have plenty of cause to assume harm until otherwise proven and the precautionary principle should apply.

We’ve had over thirty years of experience in assessing the credibility of online source. Dismissing the fact that you can’t trust LLMs with “how is that different from a search engine” is just arrant nonsense. LLMs don’t have any of the markers we rely on for assessing sources.

Optimistic visions of the future

What depresses me, is that the people most enthusiastically hyping AI don’t get that their “optimistic” vision of the future is worse than a future were these systems either don’t work or remain imperfect.

Perfect art generation will eradicate art as a practice. All of the data sets all of these AIs have been trained on will stop growing. Every creative field will stagnate.

And once you’ve injected an AI into every process, you really think tech cos will resist the temptation to use that to manipulate?

With pervasive unemployment, you really think US companies will allow the taxation required to implement Universal Basic Income? You think that countries will take the destruction of local economies and the theft of entire industries by US tech cos lying down?

Where to you think demand will come from with near-universal unemployment?

You think allies will remain allies when you’ve taken all of their jobs?

The vision of the future where these technologies work is a nightmare, not an utopia.

This is something that AI and blockchain fans have in common. They don’t get that “this is kinda garbage and maybe full of fraud” is the optimistic take

The idea that the worst case scenario is that a bunch of companies will cycle themselves into worse business outcome by buying into the unfulfilled promise of a half-baked idea, that we’ll only have to deal with the abuse and fraud generated by criminals and bad actors, is the optimistic take on this tech

Because the alternative is much worse.

(I’d just like to point out the word “optimistic” in the posts above. If you think “this tech leads to AGI and a new species of slaves and/or the creation of new mechanical gods” is an optimistic vision of the future, then there’s something seriously wrong with you.)

I don’t know if I’ve just been reading too many academic papers but this looks like fairly straightforward guidance. I see that some media outlets are saying that lying about whether you used an AI to generate a work might cause you to lose copyright protection for that work, but that isn’t strictly true.

If you lie and register the copyright for an AI generated work, you will have claimed another’s work (the AI’s) as your own.

All that’s happening when the lie is revealed is the authorship of that work is getting correctly attributed to the AI, whose works aren’t eligible for copyright protection, so the work doesn’t get any. Nobody ‘lost’ anything. You just committed fraud.

Remember when we had the ostensibly reasonable side of the tech influencer sphere saying that web3 was too big to fail?

Good times. Good times.

Not a deranged industry at all.

“The stupidity of AI”

The belief in this kind of AI as actually knowledgeable or meaningful is actively dangerous. It risks poisoning the well of collective thought, and of our ability to think at all.

Honestly, this week’s must-read.

“How to tell if AI threatens YOUR job”

I’ve come to a pretty grim, if obvious, realization: the more excited someone is by the prospect of AI making their job easier, the more they should be worried.

Interesting take.

“This Week in AI Doublespeak - by Gary Marcus”

‘AI “art” is like moving into a house, painting the living room, and declaring that you built the house.'

"‘ChatGPT said I did not exist’: how artists and writers are fighting back against AI - Artificial intelligence (AI) - The Guardian"

“Why open data is critical during review: An example - Steve Haroz’s blog”

“What do tools like ChatGPT mean for Math and CS Education?"

Nothing could be farther from how a calculator works.

“Tooled. — Ethan Marcotte”. “But I’ll just note that labor economics has an old, old term for [gestures around] all this: de-skilling.”

“OpenAI’s GPT-4 Is Closed Source and Shrouded in Secrecy”

“Microsoft just laid off one of its responsible AI teams”. Building an ethics and responsibility team for AI is a productive way of getting all the troublemakers into one room to get rid of them all at once.

“The climate cost of the AI revolution - Wim Vanderbauwhede”

“Don’t ask an AI for plant advice - Tradescantia Hub”. Using AI for specialist problems (which is most of them) is a trap. The AI will lie confidently, and you won’t have the expertise to spot the lie. These are not information systems.

Best of the rest

“All data is health data. – Hi, I’m Heather Burns”

It is merely contextual. So you need to think of all of your data inputs, collections, and sharing, in that contextual way.

“Google won’t honor medical leave during its layoffs, outraging employees”

She was let go by Google from her hospital bed shortly after giving birth. She worked at the company for nine years.

Tech cos are run by scumbags and seeing media outlets frame this act, which would be illegal in most of the civilised world, as something that only “outrages” the employees, and not society in general, is disappointing.

The rest of the best of the rest

You can also find me on Mastodon and Bluesky