Public money into an AI startup? Was it seen as a way of bypassing data regulations to get hold of patient data, with the hospitals owning shares? What exactly were Sensyne promising?

Is this the start of the 'AI' hype bubble starting to burst?

It appears it wasn't public money invested, they were receiving equity shares in exchange for the data.

On that basis they haven't actually lost the money, just failed to gain it at some point in the future when they sold the shares. It also appears that most didn't ever share the data either.

Hopefully though it is the AI bubble bursting - it's about time it did.

"It appears it wasn't public money invested, they were receiving equity shares in exchange for the data."

Well if you're giving something over, it doesn't have to be cash to be public money - data has a value. Setting up this deal would cost the public purse a fair bit - by the time all the legals, responses, use of publically funded staff has been accounted for I'd be surprised if there was much change from a million quid.

So yes they bought shares which had a value with public money / assets which had a value and you'd expect that at least the initial value of the shares could be redeemed, else its a loss. And they lost it.

I can't find anything about whether data was shared, whether it was used or is still held and could end-up in a fire sale one day.

or is it perhaps the bursting of the patient-data-for-unenforcable-promises bubble?

No, silly me,. that would be too much to hope. :(

just keep the patient data in a poorly or unsecured server? Just asking.

Don't worry, I'm sure they are already on to that one too.

Everyone knows the stock line that investment companies always use

The value of stocks and shares may go up or down. Past performance is not an indicator of future performance

I wish they had stuck to doing their primary job and use actual intelligence, rather than the fabled artificial intelligence, which generally is fairly dumb.

The thing is, there's a lot that automated pattern matching over health data can help with, if only it is developed correctly.

I have a condition that requires regular platelet counts. As I understand it, this particular test is still largely counted by eye under a microscope by the path. lab at my local hospital, something that really ought to be able to be automated, which would speed up the process.

The hope for AI is/was that if you let a self-training AI look at patient data or test results, it might me able to spot trends or correlations for symptoms leading to diagnosis which would be useful. But it seems that there are problems, as IBM has recently got out of the business, and other AI projects appear to be struggling.

I don't know for certain, but I read somewhere that one of the problems is that it's not really possible to get these self-training AI systems to explain how they came to a particular diagnosis, and this causes concern in the very conservative world of medicine, as they like to understand how decisions are made.

"it's not really possible to get these self-training AI systems to explain how they came to a particular diagnosis"

So the answer is 42...

It would be interesting to understand how consent was obtained for data sharing, given the well documented problems with obtaining this at national level. Particularly for GOSH.

From my thin understanding, as soon as you anonymise data then it is no longer personally identifiable, so it's no longer "you". GDPR isn't specific about anonymisation but a working group has pointed out that the act of anonymising data is "legitimate interest" because it's for the benefit of the data controller (GOSH) and the subject (patient).

Whether the patients knew what they were signing up for is another matter, all this legalese isn't in plain English and T&C and Privacy Policies tend to be huge. I doubt they knew and for many, I doubt they care.

Caldicott Guardians and NHS data protection impact assessments are pretty robust and in this case AI doesn't need identifiable data to function and you can't retrieve individual patient records from a model. From a personal information protection standpoint, that's as good as it gets.

Anon because I know enough but not expert.

It's been shown several times that given enough different data, some 'anonymized', and some public domain, it's easy enough to de-anonymize the data to the point where you can re-construct information that is useful to the large data consumers, and possibly damaging to the data subjects.

This is something that does not seem to get through to the admins. and politicians.

That's true; in the same way that you're never really anonymous and that your information is already out there. I wish that the powers that be (NCSC rather than the ICO [1]) set out specific guidelines as to what point a record is anonymous. Is clearing name, DoB, address, email, contact numbers etc enough or do you need to anonymise the fact that they visited their GP called X on 1st of March 2022?

Regarding AI however, no amount of fusing different sources can retrieve individual records from a trained neural network. Training is a one way process and you lose the individuality of training data.

1. https://ico.org.uk/media/1061/anonymisation-code.pdf

When our elected representatives have an interest in acquiring the data they usually make it easier for us to get the benefits of the great program by tossing out the usual rules and make participation opt out.

Only mean-spirited people would think it might not be our best interest they have in mind ....

Where do "our elected representatives" come into this?

Well they make the laws, normally requiring informed consent but seem to make laws that allow organisations like NHS to switch to opt out.

At least where i live they made special laws for electronic medical records resulting in everyone automatically being enrolled in the system and making it complicated to opt out.

I'd assumed the [1]NHS data grab resulted from similar special rules created by your representatives. I might of course be wrong and they just use a creative argument to collect the data for "legitimate interest" reasons.

[1] https://www.theregister.com/2021/05/13/nhs_data_grab/

I am not a fan of cancel culture or firing for the sake of it but there are serious ethical issues at these hospitals and arguably compounded by financial incompetence.

You could argue that it was an exchange and not purchase - but I'd argue that value was purchased with patient information, highly unethical and is in financial reports = it's been bought somehow, exchange of goods.

ICO should have it's big boots on and the boards need clearing out.

Regrettably, we've seen similar problems in local authorities that have invested (and lost) real money in a whole variety of speculative commercial ventures (from property developments to electricity retailers).

While it's easy to criticise, what's fundamentally behind this is a chronic shortage of funding (which is why 43 hospitals have roofs that could collapse at any minute and social care is increasingly unavailable to those who need it) that is leading managers to seek other sources of income, sometimes unwisely and sometimes with tacit encouragement from central government.

Public services need more money and if we're not prepared to give them it while still insisting they operate they will increasingly be driven to find it in other ways.

This is just sensationalism and widely distorting the truth. The various trusts didn't actually pay any money for the shares. They provided data and in return were given some shares which are now pretty much worthless. THEY HAVE NOT LOST ANY MONEY. You could summarize it as they were tricked into giving away data for free but even that's a distortion of the facts.

Do you although consider it a distortion of facts to say they gave away (allegedly anonymized) data for something that has no real value to them?

You missed "tricked" from your quote. There is no evidence that the company's management intended to collapse the company. If they had then "tricked" might well apply. As far as can be seen the trusts were participating in a development that could be of use to them in their everyday operations. It's entirely possible that they might have participated in that without the offer of shares.

The complicating factor here seems to have been that the shares were listed and hence received a value that had to be entered into the accounts. If that were not the case then they could have been entered with a nominal value reflecting the cost of the effort needed to prepare the data for sharing. That is the actual cash loss which has been incurred here.

Technically they have not lost money as they exchanged data for shares. But that data had a monetary value as it was exchanged for shares that had a monetary value. If the trusts failed to sell the shares which are now worthless as they cannot be traded, then they have lost money.

If you exchange something of value for something else of value that subsequently turns out to be worthless then you have lost money. The money you would have got if you had sold your something of value.

The trusts that got shares for data showed those shares as assets with a value £x in their financial reports last year. This year’s financials won’t show assets with a value of £x as they now have a value of £0. So they have lost money.

But not to worry everybody as it's been "Annonymized"

I think you'll find that is worth considerably more (to the right players) than whatever share promises they issued to the hospitals to get it.

I smell bait, followed by switch.