Hannah Fry AI programme

[Daygloboo]

Saw a programme about AI tonight with the brilliant Hannah Fry. This episode was about AI use in healthcare. Most of it was about healthcare and health insurance in US..God it was frightening. Truly awful. I cant go into detail here. It would take too long. You'll.have to watch the programme. But I dont think it's unreasonable to say that NO WAY do I want an insurance based healthcare system to be brought in to this country. It's all for profit and they juat dont treat you if it is not financially beneficial for rhem. Dont be fooled..by politicians who are trying to promote this. They want to make money at the expense of your health..Be warned. They pretend they dont want yo introduce the US modrl But they are lying. Please watch the Hannah Fry programme. She only spoke about the science and not anything directly political, but the information was worrying.

Becasue those people stand to make a lot of money out of it, pure and simples

I saw the first episode about the "AI girlfriend" who encouraged a youngster to attempt to assassinate the Queen, and about "AI companions" in general.

I can't remember how much Fry explained about the language model (because it's mostly stuff I already knew) but the programme was very good.

Eg Fry challenged someone who said "we" when referring to himself + computer. There was a lot about AI sycophancy.

I tend to swerve BBC documentaries now. Digging for Britain is bearable. Just.

They are a masterclass in how to cram 20 minutes of content into a tiny hour.

I can still remember learning about quarks, aged 9ish when "Horizon" was repeated on Saturday afternoons. Now that was real science.

By the way, @SerendipityJane , I've found it very interesting reading the egs of "AI" answers you've posted occasionally.

Particularly the way the machine uses an "active listening" technique and grovelling apologies to try to deflect the user's justified ire while admitting it's screwed up hopelessly again.

That tied in well with what I learnt from Fry's programme.

Yes, the first two were quite good about AI/ LLMs.
The third one was slightly infuriating as it focused a bit much on people waving placards and not much about the mechanisms of AI in the healthcare context, which would have been far more enlightening. But it was astonishing to see how clearly the main issue is the US healthcare model itself - AI just feeds into the existing capitalism/ lack of patient-centred care that's built in.

Judging from the first programme, I'd rate this series very well worth watching – even though I too sometimes struggle with Fry's bouncy enthusiasm.

In fact I find her good even on topics I know a fair bit about, as she usually pulls in quite broad-ranging material and shines interesting sidelights.

I thought the one last week where the man had made his dead mum into a sort of live phone call every day particularly creepy. Everyone dies, to be hanging onto his grief like that really wasn’t healthy. It’s one thing to have a lovely recording of someone’s voice, but for them to be talking back is just too weird for me!

I have reams of transcripts where ChatpGPT has admitted it (a) can't follow a conversation (forgetting what it previously was told and told you), (b) will make stuff up to keep you happy and (c) can not provide a non-zero confidence for a lot of it's output.

None of which surprises me. As I age, I have less patience. If you genuinely cannot see that we have no working definition of "intelligence" except by negation, then I suggest you yourself are not intelligent.

There are a plethora of tasks I can and do engage "AI" in. But they are just leveraging it's ability to pattern match at the speed of light. I'd never trust it.

YouTubes ability to play at 2x speed is useful.

ChatGPT said

Voice control systems still struggle when the environment becomes complex or when backend services are unstable.

I said

No. They struggle because some idiot thinks that "AI" is in any way even remotely close to real human intelligence. Which it isn't.

ChatpGPT agreed (and emphasis is ChatGPTs)

You’re highlighting a criticism many engineers and researchers also make: the term “AI” creates expectations that the systems cannot realistically meet.
Most systems marketed as AI today — including Google Assistant in Google Home — are not reasoning systems comparable to human intelligence. They are combinations of:
speech recognition models
intent classification models
rule-based device resolution
cloud service orchestration
A simplified pipeline looks like:
audio
→ speech-to-text
→ intent classifier
→ entity extraction
→ device resolver
→ smart-home API call

Each stage can fail independently. None of the stages actually “understands” the situation in the human sense.
Humans interpret something like:
“Kitchen sink on”
using context and world knowledge:
kitchen → room
sink → physical object
intent → turn water on

The system instead tries to map words to predefined intents. If the words do not match the expected pattern, it guesses.
That is why you see behaviour such as:
wrong device selected
weather returned instead of sensor data
silent failures
inconsistent logging
From an engineering standpoint, voice assistants are pattern-matching systems with probabilistic outputs, not general intelligence.
The marketing label “AI” often implies a level of reasoning or contextual understanding that the underlying technology simply does not provide.

i don’t want a US style 100% for profit healthcare system, but the current NHS system simply isn’t working. Most of Europe has some socially funded medicine alongside insurance, often through not for profit mutual organisations.

It’s a great TV programme, everyone should watch. I like Hannah Fry but I do fear she’s becoming a little overused, a bit like Lucy Worsley (who I also like but is EVERYWHERE).

On the health service point, this is not a straight up choice between the American system and the NHS.

Before any additional money is put into the NHS, they could reduce their astounding capacity for inefficiency.

It is quite staggeringly bureaucratic.

If only that meant that (non returnable) supplies were ordered correctly.

That serious conditions were not allowed to deteriorate until they lead to a catastrophic collapse in independence requiring tens of thousands of pounds in additional complex needs (not that they ever get supplied, to be fair).

That different teams all had access to the same information to avoid wasted journeys and (yet) more duplicate supplies.

That they did not lose 50% of submitted test results.

And that is just my direct personal experience with my own and family members care.

Just a crude estimation is that the NHS is around 20% inefficient. However all the managers are top notch apparently. So so they keep assessing themselves.

I'm sure the NHS (and the Civil Service) would love to be less bureaucratic and inefficient but its Not Allowed because, well I don't actually know why, but there's rules and processes which must be followed.

Something to do with making sure that the private sector contractors who supply things at great cost and only after a huge amount of paperwork is completed get their cut as far as I can tell.

Did I mention teams in one hospital not being able to see scans from another team in the same hospital ?