@jenitennison.com
Data nerd/wonk. Founder of Connected by Data, campaigning for communities to have a powerful say over data and AI. Into trans rights, neurodiversity, board games, lego, dogs, spreadsheets. www.jenitennison.com
They're also both worse at reaching everyone than road infrastructure.
Which do we think DPI is actually like and what does that imply about its design? Is what I'm wondering :)
But very likely getting too hung up on analogies.
Though revealing, right, as both those infrastructures have lower redundancy / duplication but that means when they go wrong it's a lot more disruptive than a traffic jam.
Water mains are simple and pretty robust, so we manage. Rail frequently falls back onto road infrastructure.
Anyway, I probably read way too much into that first sentence. It did say "wouldn't build a dozen roads", not "two roads".
There's probably some Goldilocks zone of just the right amount of duplication in infrastructure, dependent perhaps on reliability of the pieces and diversity of actors? Hmm.
(It's useful to have single points of truth, but more diversity in that truth than you might imagine – for example, multiple potential "lists of countries".
Government's job is to provide key datasets *and* support eg standards that enable different sources to link together and be exchangeable.)
(I feel on firmer ground with data infrastructure than DPI, where I'll bow to others' expertise.
With data infrastructure, multiple data sources, even about the same thing, _do_ provide resilience, cater for different needs and provide different perspectives...
Like, shouldn't our digital public infrastructure be small pieces, loosely joined, and easily swappable, rather than big overarching systems that suck in all functionality? Wouldn't that provide resilience and choice, and facilitate evolution over time?
It even allows us to have different financing models: toll road vs open road.
Of course there are always limits with analogies and exploring what doesn't apply is sometimes as revealing as exploring what does.
But I wonder if these features of real road infrastructure could be instructive for DPI?
And for good reason!
Having multiple roads between two places builds resilience into the system – when one is taken down, others can take the load.
It enables different types of traffic to travel at different places: stops cyclists from being in danger and tractors from holding us up.
As a big road analogy fan, I frowned at the first line of this piece by @dianecoyle1859.bsky.social & @eaves.ca:
"A government wouldn’t build a dozen roads connecting the same two places."
Of course they would! How many roads are there between London and Manchester??
www.imf.org/en/publicati...
. @jenitennison.com will feature in our #CamFest event exploring how AI and digital research in Cambridge is shaping global impact.
📅 Monday 16 March 2026
🕠 17:30—18:45
📍 The Glasshouse, Innovate Cambridge
🎟️https://www.bennettschool.cam.ac.uk/events/from-cambridge-to-the-world/
More pondering on AI as I wrangle with what it means for public servants... Here's where I've got to so far: There is a useful path to take between hype and doom. It's time to keep experimenting and learning as nobody knows what the future will bring. jasonkitcat.com/2026/03/02/s...
The Responsible AI Advisory Panel was established by the Government Digital Service (GDS) under the Department for Science, Innovation and Technology (DSIT) to bring together views from across the public sector, industry, academia and civil society on the responsible use of AI. The Panel is one of the key initiatives supporting the government’s AI for the public good agenda, as set out in the Roadmap for Modern Digital Government. The goals of the panel are to: ● provide expert strategic advice to senior officials and ministers in DSIT ● inform GDS on best practice for developing responsible AI ● provide constructive challenge and scrutiny on government's approach to developing responsible AI products The panel has an independent chair and 14 external panel members. Ministers also attend panel meetings, alongside senior officials.
The UK Government has published further information about the GDS Responsible AI Advisory Panel, including membership www.gov.uk/government/g...
@jenitennison.com is Chair
#AIpolicy #techpolicy #genAI #govtech etc.
I'm a bit late sharing this, but this from @jenitennison.com is a really useful review of the trajectory of recent AI Summits www.deccanherald.com/opinion/empt... - and the need to support the 'necessary friction of democracy' in AI governance
Or a Bluesky thread
Is it just me? Am I being snobby or showing my age? Is it just a feature of unsophisticated prompts and generic AI, untrained on writer style, something we'll find a way to get better at over time? Or is friction sharing ideas systemically useful?
I guess that slop did trigger some ideas after all.
In archiving, the restricted size of the archive forces useful decisions about what to retain (but laden with the values of the chooser).
Does capacity to write act as a useful constraint? Should sharing ideas be frictionless? I think it should be generative (not in an AI way), and slop isn't.
But as a reader, I find I just can't engage with genAI content. It's like the writing slips down well travelled paths in my brain without leaving much impression, whereas good writing surprises me with the thoughts it triggers.
I wish I could read it – I'm sure the ideas behind the slop are good!
I completely get the reasons for using AI to write. I really struggle to have capacity to turn the (I think interesting!) thoughts in my head into prose. I can absolutely see the utility of genAI to expand and make intelligible ideas that would, without time I lack, be a rough sketch or unshared.
Earlier I started to read a piece on a subject I'm really interested in, but it was written by/with AI and I just got bored by it.
There's the repetitive linguistic rhythm, but also it's that it's strangely context free. It lacks links to related thought, personal anecdotes, cultural references.
Palantir must be wracking their brains for a partner who won't have prissy little "ethical" concerns about massive scale domestic surveillance and legal autonomous weapon systems.
And who has a safeguard-free, gas-guzzling AI ready to mecha-Hitler.
www.theguardian.com/us-news/2026...
Not necessarily, though basic cash redistribution is clearly an option. It could go on infrastructure (community art studios) or training, or offset tax breaks.
We need diversity in creative works, especially given the homogenising effects of AI.
And the creators most likely to disappear are those who are already struggling to make ends meet, because they're not mainstream.
So instead redistribute to support diverse creators, local journalists etc.
The model described requires a bunch of pretty tricky transparency around exactly what's used in training, and redistribution to the most valuable works / voluble creators.
That's one version of fairness, but I don't think it's as useful as an approach that targets struggling / unusual creators.
This is interesting. What's described aiui is essentially a levy on AI companies using publicly accessible works for training purposes, which is then redistributed to creators.
The big sticking point for me is how that money is redistributed...
In the UK addresses – for example, "29 Acacia Road, Beanotown" – are maintained by local authorities. The list of address changes as flats, houses and offices get developed and demolished. These local lists are collected into a national dataset and made accessible to other public and private sector organisations. While this happens address data gets tangled up in a complex web of other organisations who end up holding some intellectual property rights in the data, particularly the Royal Mail and the Ordnance Survey. The Royal Mail is a business and the Ordnance Survey is a business owned by the government. Because they are businesses their primary goal has become to generate revenue for themselves by selling the data, rather than maximising the public good that could be created from using the data. As a result the UK's address data has become expensive, hard to access, not always accurate, and hard to correct.
An opportunity to put address data higher on the UK agenda peterkwells.com/2026/02/21/a... post from @peterkwells.com
Also:
How to respond to the DSIT survey on increasing access to public sector data peterkwells.com/2026/02/21/h...
#openaddresses #opendata #datapolicy
Amazing you're doing this!
Is the consultation going to cover what happens when services go wrong or people can't find the information they need?
Are you doing any evaluation of this consultation process? I'm fascinated to understand the details of it, as well as the outcomes.
It’s on Google, even though it is making search worse. It’s on WhatsApp, even though I absolutely do not want “homework help” or “relationship advice” from my messaging app. In the year of our lord 2026, it is everywhere.
And it *doesn’t work*.
![Stuart Russell, a professor of artificial intelligence at the University of California, Berkeley, who closely follows India’s progress, said: “If we get to AGI [artificial general intelligence], AI is going to be producing 80% of the global economy. All manufacturing, most agriculture, all services will be just done; managed by AI, produced by AI.”](https://cdn.bsky.app/img/feed_thumbnail/plain/did:plc:mo7wgmn53nbs64qzdzdpupde/bafkreibylcofnqqtd2ovhasy3ujnyvngtmf22epx2jy3oyzstglsmnipta@jpeg)
Stuart Russell, a professor of artificial intelligence at the University of California, Berkeley, who closely follows India’s progress, said: “If we get to AGI [artificial general intelligence], AI is going to be producing 80% of the global economy. All manufacturing, most agriculture, all services will be just done; managed by AI, produced by AI.”
Imagine, he said, an Indian village priced out of having a health centre. In the future, AI could design the hospital and “along comes a bunch of giant quad copters carrying the materials, and a bunch of robots come and assemble everything. Two weeks later, you’ve got a hospital.”
I'm astonished this statement could be reported so uncritically.
Let's be real: *IF* AGI is ever developed, all its efforts will be spent making OpenAI profitable and sending Musk to Mars.
From www.theguardian.com/world/2026/f...
@tomski.com vv
If you’re interested in open government data, please read (and repost) www.nationalarchives.gov.uk/blogs/digita...