Thread by @EinsteinsAttic

So let's unpack this.

We hear a lot about "#dashboards"; we were even promised sight of one, back in March* - though we've not seen it yet.

But what are they, in reality?

__
*healthtech.blog.gov.uk/2020/03/28/the…

https://twitter.com/Paul0Evans1/status/1307323299966353413

Well, for starters, we know quite a bit about @NHSEngland's 'dashboard(s)': back in March, @MattHancock gave it powers to DEMAND confidential patient information from care providers*, for (broadly defined) "#COVID19 purposes"...

__
Under the #COPInotices:
gov.uk/government/pub…

...and we know @NHSEngland's hoovering up *at least* 70 different datasets into its #DataStore (the list at data.england.nhs.uk/covid-19/ is no longer comprehensive due to the controversial contents of some of the datasets).

As its "#PrivacyNotice" points out,
england.nhs.uk/contact-us/pri…

...@NHSEngland is using @Microsoft @Azure to process all the patient data it holds, which it #pseudonymises (i.e. replaces obvious #identifiers with a pseudonym; what in law and in practice is still #identifiable #PersonalData) before @PalantirTech sucks copies across into its...

...#Foundry 'data integration platform', which runs on @awscloud.

Everyone's a bit vague about what happens next, but we do know that it's not just @PalantirTech but Dom Cummings' favourite firm, @faculty_ai, plus @Deloitte & @McKinsey 👇 who are making/linking the #models...

...which '#interpret', '#predict' or otherwise #process the data that's been fed into the Foundry platform; the *results* of which processing are presented to decision-makers via the '#dashboards'.

And remember, at this point, we're still only just talking about @NHSEngland's...

..set-up.

God only knows whose #economic models they're using (@hmtreasury's? @bankofengland's? Something Dom got @faculty_ai to hand-roll*?) OR what data (one assumes @hmrc RTI & tax data, @DWP's plus @ONS business microdata at least). And one has...
__
*theguardian.com/politics/2020/…

..to hope they're keeping any #electoral, and definitely any #PartyPolitical data and processing ENTIRELY separate!

But back to the '#dashboards'.

Ask yourself - what do they *actually* show? At BEST this is #partial data, of varying #quality, some with significant #timelags...

...being fed to #models, #algorithms & combinations thereof ('daisy-chained' together) each #parameter of which is a '#knob' that can be twiddled to affect the output.

And what is that output? The idea they're working on/from a "single source of truth" is patently ridiculous...

...it's an #OfficialAbstraction, comprising only of that which they've CHOSEN to #capture - notably lots in #health, but virtually nothing in #SocialCare - or can even be bothered to collect.

Is what they're looking at "real"? Is it even close to #GroundTruth?

And whattabout...

..the #models, which @faculty_ai, @Deloitte, @McKinsey and others are building: how #reliable or even #realistic are they?

Have they chosen or trained any #MutantAlgorithms?

And if they have, HOW WOULD WE KNOW?

And as whoever-it-is fiddles with the #knobs to try to #predict...

...the effect of a new #policy or #rule (of 6), to what extent are they "following the #science" rather than juggling a bunch of #DataDriven delusions?

Things built out of #logic may appear to be #logical; much as things using #bits are #digital. Neither NECESSARILY makes 'em...

...any good. And it certainly doesn't make them any more #true.

N.B. I'm not saying we shouldn't TRY; we should! But what we do should be done openly, being honest about the limitations of our #knowledge and of our #capacity. Not shrouded in secrecy, lies and half-truths.

/ends