Today
@AINowInstitute
published my essay on the UK’s industrial approach to AI over the last 10+ years.
I argue that successive governments have failed to present a coherent vision for the role that the AI industry should play within the UK economy... 🧵
The new Tony Blair Institute report finds huge time savings for the public sector from the use of AI tools... by asking GPT-4 what tasks can or cannot be automated 😵💫
Mika is a superstar who has contributed more to the labour movement — and the Labour party — than a small-minded bully like Rosie Duffield ever will. What else will have to happen for the leadership to withdraw the whip?
Today a member of TUC staff, Mika Minio-Paluello, is being subjected to anti-trans abuse online, because she is a mother.
This is unacceptable.
Everyone should be safe and respected at work.
Thrilled that our new
@AdaLovelaceInst
report ‘Regulating AI in the UK’ is now out in the world! 🤩
In it, we summarise the UK’s plans for AI regulation — & suggest how they can be improved to ensure AI is safe & accountable.
🧵 on our findings...
@youngvulgarian
We faced this problem and our only viable option was buying an adult dog from a family who didn't want her anymore. We love her but it's not ideal that anyone who wants a pet is basically caught between absurdly restrictive shelters and a wild west of often unscrupulous sellers
This — from
@mer__edith
— is excellent not only on the TikTok ban, but on the unspoken assumptions that make working in "public policy" so uncomfortable at the present moment:
A pic + some items from
@Civil_War_Spain
's fantastic tour / 'walking museum' exploring Barcelona's civil war history — can't recommend it highly enough! ✊🏻🔴🟡🟣
Link to the report here for anyone interested. Many of the other assumptions they make are defensible, but overall I'd say this type of analysis should be taken with a huge helping of salt
🤯🤯🤯
New LLM-powered research from
@AdaLovelaceInst
forecasts UNPRECEDENTED public sector productivity gains from the AI revolution 📈🚀
Buckle up, because you're not going to believe these findings... 🧵 (1/7)
Come work at Ada!
I'm transitioning to a new role within the organisation from October, so we're looking for someone new to drive forward our work on data and AI governance here in the UK.
It's an amazing team and a brilliantly rewarding job. Please share widely!
📢We're hiring!
We are looking for a UK Public Policy Lead to deliver our work influencing law and policy, as part of Ada’s mission to ensure data and AI work for people and society.
Apply by 4 September 2024:
"AI is an inherently conservative force when it is considered without any reference to the socio-economic context within which public services operate."
Sharp, well-taken critique of the latest AI-in-public-services hype wave:
@hkatewilliams
can't really explain why but this reads like they're in school and have been given the task of writing about your article for homework 😭
Great to speak to
@NewStatesman
about how we shouldn't take the existing AI paradigm "off-the-shelf" from Silicon Valley — and the need for Labour to instead formulate a positive industrial and regulatory vision for AI
Gratuitously nasty how ministers have repeatedly shoehorned trans people into their speeches this week. Should (although doubtless won't) be a moment for so-called 'gender critical' voices in Labour to reflect on the hateful politics they have chosen to ally with and enable
"Six cabinet ministers used their moment at the podium to draw attention to trans and gender issues, with some frontbenchers promising fresh moves, reviews or guidance."
November's AI Safety Summit is a chance to envisage alternative AI futures in which these powerful systems are made to work for people and society.
New
@AdaLovelaceInst
blog with
@wonderlikeours
on how to make the most of it — and short 🧵of key points
"Starmer will inherit a broken state, a shattered economy, a despairing populace and an unstable world. He will long for the kind of boom that cushioned Blair’s arrival."
Just kidding!
It's great that the new Government wants to make AI work for the public sector, but much of the debate is driven by hype, bluster and industry pitch decks.
@ImogenParker
and I have some thoughts on how to develop a credible agenda...
2/7
Packed committee room for
@The_TUC
@ConnectedByData
event in Parliament on workers' experiences of AI — powerful testimonies echoing recent
@AdaLovelaceInst
research on the need for greater transparency & human oversight when AI is deployed
The law hasn't kept pace with AI uptake (as our
@AdaLovelaceInst
work with
@A__W______O
shows: ), leaving people vulnerable to intensified exploitation and discrimination.
Great to see proposals published that would put workers back in control:
"Power in AI accrues disproportionately to a handful of private firms. A publicly-funded international organisation might be better able to avoid economic pressures overriding meaningful research."
Superb read on 'CERN for AI' proposals by
@Elliot_M_Jones
Mixed feelings about the UK Government's decision to cut compute investment.
If we want to build a vibrant AI sector, greater public involvement is essential. I hope this doesn't mean it's off the table.
On the other hand, the previous approach didn't exactly stack up...
🧵
I've pledged to cycle 280km (🙀) over the month of February to support
@elop_lgbt
—a fantastic charity that provides low-cost 'pay what you can' counselling & mental health services to LGBTQ+ people 🏳️🌈🏳️⚧️
If you want to sponsor me you can do so below 🥰👇
"Tech is embedded within institutions and power structures. Regulation is an important tool for reshaping these structures, strengthening the hand of communities in the face of power that is all too often unaccountable."
🔥 from
@fhr
in
@NewStatesman
:
"Without public digital infrastructure ... any discussion of European companies challenging big tech’s position and European digital sovereignty is almost meaningless."
Good take on the Mistral/Microsoft tie-up from
@ZWarso
:
Some lovely words and aspirations here, but – unless backed up with market-approval and enforcement powers – that's all they'll be: words and aspirations.
Read our blog on why 'safety commitments' without statutory underpinning are unlikely to work:
In an historic first, tech companies from across the globe have committed to developing AI safely.
From
@OpenAI
to
@Meta
, 16 companies have signed up to the fresh ‘Frontier AI Safety Commitments' 👉
#AISeoulSummit
New from me,
@agstrait
and
@wonderlikeours
on the state of "AI safety" ahead of the Seoul AI Summit. We argue:
- 'evals' in their current form aren't effective
- voluntary agreements don't work
-
@AISafetyInst
needs to pivot, and fast
Link below 👇
Foundation model evaluations are used by firms, governments and regulators across the world to assess whether AI systems are fit for use.
Our new report by
@Elliot_M_Jones
, Mahi Hardalupas and
@willie_agnew
finds current approaches lacking, and suggests some ways forward
📢Evaluations are a useful method for identifying and mitigating the risks posed by foundation models.
However, they should be used alongside other tools, such as audit, incident reporting and post-market monitoring.
Read ‘Under the radar’:
(1/6)
I *loved* this piece from
@sunhahong
in the new
@logic_magazine
— a sort of ethnography of Silicon Valley techno-mysticism as experienced by (and refracted through the lens of) contemporary Korean culture:
Has a gorgeous day out cycling near Grantham with
@edwardwatson
... but most importantly completed my
@elop_lgbt
cycling challenge with a few days of February to spare! 🎉🚴🏳️🌈
New briefing on foundation models in the public sector — we cut through the hype to ask how practitioners are thinking about these systems, their opportunities and risks.
Full report (by the brilliant
@Elliot_M_Jones
) out soon but for now check out the briefing and thread below:
📢As AI technologies advance rapidly, Government must consider how to use foundation models in the public sector responsibly and beneficially.
Our briefing for policymakers and public-sector leaders provides context to support decision-making.
@piercepenniless
I'm continually baffled that they actually allow these rather than just rolling out more (electric) docked bikes — I guess it's much cheaper (or even a moneyspinner?) for TfL/councils to just license Lime etc
This is incredibly disappointing. The LGB Alliance does nothing to support lesbians, gay men or bisexuals: it's solely dedicated to attacking trans people.
Granting this hate group the veneer of respectability undermines the vital, life-saving work of legitimate LGBTQ+ charities
#Breaking
Transgender rights group Mermaids has lost its case to have gay rights organisation LGB Alliance stripped of its charitable status in what is believed to have been the first case of its kind in the UK
Gorgeous three days cycling from the Irish Sea at Workington to the North Sea at Tynemouth. Endorphin levels (and levels of muscle soreness) couldn't be higher 🥰🚴🏼🏞️
Late posting, but wanted to mark a gorgeous long weekend cycling 170 miles from Morecambe to Bridlington on the grandiosely-named "Way of the Roses" 🌹🥰
(Thank you
@Sustrans
for a fab route, which I'd recommend to anyone curious about long distance cycling!)
Future generations will marvel that we* ever took seriously figures as self-evidently ridiculous as Birbalsingh (or indeed Rowling, Stock et al)
*a cynical, click-hungry subset of the British media class
It was amazing to be part of this brilliant, diverse cohort over the past couple of weeks — and inspiring to learn about so many new perspectives on AI governance.
Thanks
@NathalieSmuha
@Hendrickx_Vic
and everyone else at
@KU_Leuven
for hosting such a wonderful Summer School!
It was such a privilege to get to know this talented, curious & kind set of participants to the
#AISummerSchool
's 3rd edition.
Massive thanks to all for the rich exchanges, and to the stellar lecturers & team (esp.
@Hendrickx_Vic
@GAVClaw
) without whom this could not take place.
"Tradition is the ground on which everyone stands, Marxist and monk alike. The concomitant risk of complicity in tradition’s tyrannies is therefore everyone’s problem."
Wonderful piece on Gillian Rose, the thinker I revisit most often:
Simon is right that deepfakes pose a profound structural challenge to democracy — but the agenda of the government's AI Safety Summit explicitly *won't* cover 'wider societal risks' such as misinformation
(source: )
There is a deep fake audio circulating this morning of Keir Starmer - ignore it.
It’s a reminder why the upcoming AI summit organised by the Prime Minister is so important.
From the Slovakian elections a few days ago to today’s incident, this is a new threat to democracy.
Exciting new research from
@AdaLovelaceInst
colleagues looking at the British public's attitudes to AI — people want to see independent regulation to ensure that AI systems are transparent, explainable and accountable
📢How do people feel about AI?
Today, Ada and
@turinginst
launch our new report on the findings of our nationally representative survey exploring public awareness of, experience with and attitudes towards different uses of
#AI
.
Brilliant new
@AdaLovelaceInst
report from
@annastudman
@ThinkSpeakMavis
on socioeconomic inequalities in digital health services.
Required reading for anyone involved in designing, procuring or delivering healthcare services that use data or AI 🩺🚑👇
There's a lot wrong with today's AI sector, and greater public control could be part of the solution.
At
@AdaLovelaceInst
we'll be publishing later this year on the possible options for countries looking to build out public compute. Watch this space...
The current media furore around AI centres on future existential risks from sentient AI systems. There's a real danger that governments scope regulation in response to these narratives — leaving existing failures & abuses of AI ungoverned.
Excellent thread on why this matters 👇🏻
Week two of a Labour government and the Health Secretary is defending an inherited policy founded on conspiracist innuendo and flawed evidence — and one which is likely to do immense harm to trans and gender-questioning children in the UK. So much for an 'end to the culture war'.
Puberty Blockers. A 🧵
Children’s healthcare must always be led by evidence.
Medicine given to children must always be proven safe and effective first.
I know there’s lots of fear and anxiety.
Let me explain why this decision was taken.
1/9
Incredibly useful and timely resource from my colleagues
@Elliot_M_Jones
@cp_dunlop
— should be required reading for anyone trying to navigate the confusing (and contested) world of AI policy 👇
Foundation models, large language models, general-purpose AI, generative AI - what do these different terms mean?
Our new explainer aims to cut through the confusion and discusses why language is important – but tricky – in this fast-moving area. (1/2)
So the optimistic spin is that this week's developments present an opportunity for a reset.
Government needs to figure out what it wants from AI, and use public compute as a lever to reshape the sector and ensure it serves the public interest. It now has a chance to do that
We
@AdaLovelaceInst
are looking closely at the government's new data protection proposals.
Short 🧵 on some of the longer term questions we're asking around data rights, the UK's approach to AI regulation, & how to rebalance digital power 👇🏻
Yesterday
@SciTechgovuk
published a new version of the Data Protection and Digital Information Bill – a set of reforms to the UK’s data protection regime.
Here are 3 questions we’ll be asking as we analyse the Bill informed by our independent, interdisciplinary research. (1/4)
"Singer rarely talks about experimenting on ‘normal’ philosophy professors like himself: it seems there is always someone ahead in the queue – primates, babies, disabled people."
Brilliant from
@LJFinlayson
on the culs-de-sac of abstract moral theories:
"How is Prozac like a large language model?" 💊
Superb critical analysis from
@AINowInstitute
on the role of the US FDA in reshaping the pharmaceutical sector — and what lessons it can teach us about AI.
Worth a read for anyone working on AI regulation and industrial policy 👇
[NEW] AI is often compared to snake oil: the FDA played a key role in making medicines safe, effective and ensuring we know how they work. We can learn from where it’s worked - and where it’s fallen short.
Read more here:
New blog with
@ellieshearer
and
@DantonsHead
on "public compute" — what it is, why it's important, and how we can leverage it to reshape the AI sector 👇🏻🚀
Public compute investments should be used to reshape AI development, challenge concentrated power and promote the creation of public value.
In their new blog,
@halcyene
,
@ellieshearer
&
@DantonsHead
identify 4 challenges for this market-shaping approach.
We are announcing new grants for research into systemic AI safety.
Initially backed by up to £8.5 million, this program will fund researchers to advance the science underpinning AI safety.
Read more:
Matt from London on the tour this morning with his
@claptoncfcfan
football top in Spanish Republican colours and International Brigade symbols. Antifascist shirts always welcome, especially with the gathering storm
Loving this initiative from
@collect_intel
and
@audreyt
— a live, open conversation on generative AI and democracy, using
@UsePolis
to surface points of disagreement and consensus.
You can still contribute until the end of the day 👇🏻
@odavis_
Yes, to be clear, TBI isn't alone in this: there have been a few studies using similar methods (some better, some worse)
As I understand it the jury is out on whether LLMs are fit for these sorts of enquiries (seems others are looking into this eg )
Cautiously optimistic about this. It's right that the new Government thinks critically about how AI could support the UK's longer term aspirations — not only 'growth' but non-economic aims such as decarbonisation and rebuilding the public realm
A new plan led by
@matthewclifford
will use AI to boost growth and improve our public services.
The AI Opportunities Action Plan will identify how to boost uptake of AI across the public and private sectors to increase productivity.
(1/2)
a lot of resonance with our own work at
@AdaLovelaceInst
on the need to open up big tech data with access mandates, and create new public institutions such as a "BBC for data"
My colleague
@wonderlikeours
is on the money here.
AI is being integrated at pace into our lives, with potential harms to people and society.
Ministers admit we need binding legal requirements to make AI safe — so why risk catastrophe by delaying?
Neat overview of some of the (many) things we've been up to at Ada this year — from unpacking the term "foundation models" to investigating the impact of data-centric technologies on healthcare and climate change
In our last blog post of the year, Ada's outgoing Director
@carlykind_
and Interim Director Fran Bennett reflect on the year's biggest developments, milestones and achievements, and look ahead to 2024.
Government is set to announce
#Budget2024
investments in public sector AI. But an agenda led by politicians and tech companies won't deliver public value — frontline workers and service users should be the ones driving this.
@ImogenParker
nails it here:
The Data Protection and Digital Information Bill was a bad piece of law which would have supercharged exploitative data practices while undermining legitimate efforts to pool data for research.
I'm glad it's gone — let's replace it with something better in the next Parliament.
We understand that the controversial Data Protection and Digital Information Bill will not be taken forward before the close of Parliament – meaning that it ‘falls’ and will not become law. 🧵
Been revisiting Wendy Brown lately and continually struck by the relevance of her insights to the recent "participatory turn" in AI policy (and policymaking more generally). In particular:
@odavis_
Yes, to be clear, TBI isn't alone in this: there have been a few studies using similar methods (some better, some worse)
As I understand it the jury is out on whether LLMs are fit for these sorts of enquiries (seems others are looking into this eg )
@herbiebradley
@halcyene
Im actually working on this at the moment: taking GPT4’s assessment of whether it can automate a task, and for tasks it says it can, prompting it to perform a real-world example of that workplace task, providing whatever context it needs, and evaluating the output. Kinda tricky
"We’ve missed a key piece of the puzzle: demanding that firms articulate, clearly and with evidence to back it, what the benefits of AI are to the public."
Typically incisive from
@ambaonadventure
@sarahbmyers
on the NAIRR and 'public AI':
"Our facial images are gathered en masse via CCTV cameras, the passport database and the internet. At no point were we asked about this, nor is there any statutory legal basis for it."
🔥 from Shami Chakrabarti
It was exciting to take part in this conversation last year on what AI policy can learn from the US Food and Drug Administration.
Some critical insights here for anyone thinking about how to regulate AI in the public interest, in the US and elsewhere ⬇️
Today we published 10 key insights from that conversation to set the stage for deeper engagement, and welcome questions and insights from others on the issue. Look forward to sharing more soon!
Read the interim memo here⤵️
The UK has ambitions to be a world-leader in AI regulation, but will the Government's proposals deliver on this?
Register for the webinar launch of our forthcoming policy briefing alongside independent legal analysis from
@A__W______O
One highlight of
@KU_Leuven
AI Summer School so far: learning about Lau Mazirel, an early proponent of data privacy and LGBT+ rights... who arranged the bombing of a registry office in occupied Amsterdam to prevent the Nazis from targeting their victims 🔥
📢Job alert! We are looking for a Public Participation & Research Practice Lead to develop, deliver and oversee our public participation and research projects, team and strategy.
Apply by 22 April⬇️
But there's a snag.
The outgoing Government's vision was effectively 'trickle-down AI': more compute means more AI which means we all benefit... somehow.
In reality the public benefit from AI — and the value proposition underlying public compute spend — isn't so obvious
There appear to be several serious flaws with the "independent"
#CassReview
(not least selective use of evidence and one-sided briefing). Politicians and journalists should be ashamed for feeding this circus, which risks compromising care for trans youth
@Tom_Westgarth15
It would be more interesting to see if people with genuinely disparate views could make the "disagree, agreeably" thing work. A podcast where John McDonnell and Nadine Dories debate fiscal policy over cheese and crackers, for example
Worth clarifying that the LGB Alliance only 'won' on a technicality. It's hugely damaging that they will still enjoy charitable status, but this shouldn't be seen as substantive justification of the claim that they are a genuine 'pro-LGB' organisation:
There's so much about today's industrialised AI — its environmental impacts, who can build it, its place in the wider economy — that we can't understand unless we consider the role of compute. Fantastic new report from
@JaaiVipra
@sarahbmyers
delving into all this and more:
🚨NEW REPORT surveys the state of compute, a core dependency in building large-scale AI.
We review the hardware, software, and infrastructure needed for AI––and how the government can regulate for competition rather than concentration.
Thanks to
@AdaLovelaceInst
colleagues for helping to inform this piece, and
@AINowInstitute
for publishing it!
Read the other essays in the collection for critical insights into how countries across the world are approaching industrial policy for AI 👇
"Data-driven systems now underpin our lives... we need to decide together how this data about all of us should be used."
Powerful new video from
@ConnectedByData
:
New! We’ve had enough of blue-light glowing humanoids when it comes to
#data
and
#AI
. So we’ve made a video that puts people front and centre on an issue that is overly dominated by sci-fi visions and corporate interests.
Comprehensive thread explaining why the reforms to the Equality Act being considered by the Government — and proposed by our so-called "independent" equalities watchdog — are both unworkable and deeply regressive 👇🏻
"To be sure, horror at technology is stuffy and reactionary, but that is not all it is. It is also a shudder at the violence that an irrational society inflicts on its forced members."
(Adorno by way of Hatherley)
@martin_oneill
Worth reading some of the critical literature emphasising the limits of LLMs & actually existing social/environmental harms (rather than hypothetical existential risks). This paper got two of its authors fired from Google & has been hugely influential
@aveek18
there is clearly a distinction between products for which there is a market, and products for which there isn't, right? I don't think we should reduce 'what people want' to short term consumer need, but the gulf between AI firm rhetoric and evidence of public demand is striking
Super interesting explainer from
@1Br0wn
looking at AI supply chains. Required reading for anyone thinking about how AI systems (and those developing, deploying or using them) can be made accountable for their impacts on people and society
📢AI systems rely on technology companies with various roles from development to training.
This explainer explores AI supply chains and discusses how developers and deployers of AI systems can be made accountable for identifying and addressing risks.
I really enjoyed recording this podcast with
@atarkowski
and
@EmmaThwaites
exploring how public policy needs to change in the age of data and AI.
A blog post summarizing our discussion is now live on the
@ODIHQ
— check it out, and listen to the podcast, below 👇
Our
#DataDecade
public policy blog is available to read now 📖
Matt Davies(
@halcyene
) and Alek Tarkowski(
@atarkowski
) explore how data policy and policymaking need to progress in the next decade
You can also listen to the podcast🎧
On one hand, the escalating absurdity of these stories is funny. We're witnessing some of the worst, most stupid people in Britain humiliate themselves in real time in a desperate attempt to prolong & deepen a moral panic with no organic constituency outside the media class
Now that the Prime Minister of the UK has — ludicrously! — waded into the debate about whether or not Kathleen Stock should speak at the Oxford Union, this discussion of her track record is well worth revisiting
Kathleen Stock wildly misrepresents her activism to portray legitimate, democratic and substantive political opposition as persecution. She doesn’t believe in open debate, she wants a universal ceding to her own position. I documented the reality here.
This makes sense, but also carries risks as these 'twin goals' will often be in tension. What is good for UK industry will not always be good for digital public services, and vice versa — and at times under the previous govt DSIT seemed to see its role as a PR dept for tech firms
Well well (/ht
@GavinFreeguard
). GDS, CDDO and are all moving into DSIT.
I wonder if this will lead to better integration between the twin goals of advancing digital/data/AI in govt and championing them in the wider economy.
As we accelerate towards a more opaque, technologically-mediated world, data protection law will be fundamental to keeping human beings in charge.
Our blog on why the Government's reforms to GDPR risk leaving us unprepared for an era of widespread AI use:
@dmk1793
Honestly baffled that people (including some "progressives") take this guy seriously... the idea that British stagnation has been driven or caused by a single-minded focus on social policy seems so obviously at odds with reality I can only assume it's an attempt at trolling!
There's a growing 'compute divide' in AI, which means that smaller firms and academic centres are squeezed out of frontier research.
This is particularly acute for UK researchers, because we've failed to invest in public provision of compute resource relative to other countries
These are only first steps, but we think they're important foundations for the longer-term project of ensuring AI delivers sustained value for the public.
The full blog is here—we'd love to hear what others think.
6/7
Later this year, we'll publish an evidence review drawing on research by
@ThinkSpeakMavis
,
@annastudman
,
@Elliot_M_Jones
and others into how AI has been experienced 'on the ground' by service users and public sector professionals. Watch this space!
7/7
#Eurovision
and information rights? You might wonder what the link is (you may read this thread and still question the link) but read on for our Eurovision x Data Protection scorecard. 👇
If you're at a loose end in London this Thursday, come along to 🌈✨DRAG BINGO✨🌈
@stowtradeshall
!
Fab performers, amazing raffle prizes & guaranteed fun—with all proceeds to
@elop_lgbt
's life-saving queer mental health services 🏳️🌈🏳️⚧️
Tickets here 👇🏻