Jeni Tennison
LightNews LightNews
banner
jenitennison.com
Jeni Tennison
@jenitennison.com
9.1K followers 890 following 510 posts
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
Posts Replies Media Videos
jenitennison.com
Listening to the recent Ezra Klein podcast on the AI apocalypse, I found this segment astonishing. I had no idea AI psychosis was so common that journalists are receiving *daily* emails from people demonstrating it.

www.nytimes.com/2025/10/15/o...
Snippet of podcast transcript: I get emails like the call you got, now most days of the week. Yep. And they have a very, very particular structure to them, where it’s somebody emailing me and saying: Listen, I am in a heretofore unknown collaboration with a sentient A.I. We have breached the programming. We have come into some new place of human knowledge. We’ve solved quantum mechanics — or theorized it or synthesized it or unified it — and you need to look at these chat transcripts. You need to understand that we’re looking at a new kind of human-computer collaboration. This is an important moment in history. You need to cover this. Every person I know who does reporting on A.I. and is public about it now gets these emails. Don’t we all?
October 18, 2025 at 6:03 PM
jenitennison.com
The irony of the clickbait article about the pope's condemnation of clickbait being the most viewed article...
The Guardian Most viewed list of articles is topped by one reading "You won't believe what degrading practice the pope just condemned"
October 10, 2025 at 6:50 AM
jenitennison.com
Important read on digital ID based on experience in India.
Our investigations into Aadhar revealed that once such a system is implemented, it is almost impossible to prevent its proliferation into every interaction between the state and the resident, which turns each failed transaction into a potential criminal breach. Politicians know that national IDs are hugely unpopular, which is why they invariably start with a scape-goat to sell the system to a sceptical public. These scapegoats offer a glimpse into the insecurities hidden in the national psyche. In India, Aadhar was sold to middle-class India as a technological balm to the widespread perception of corruption.
October 3, 2025 at 5:28 PM
jenitennison.com
21yo got some excellent shots of the dog today.
The dog is asleep and his jowls are squished against a pillow. Photo of a side mirror in which you can see the dog sticking his head out of the back window.
September 16, 2025 at 8:40 PM
jenitennison.com
@timbernerslee.bsky.social obviously forgot to wear his t-shirt again
On the left, Tim Berners-Lee is wearing a t-shirt that says "I didn't invent the internet". On the right, Vint Cerf is wearing a t-shirt that says "I did not invent the Web". On the left, Tim Berners-Lee is wearing a t-shirt that (on the back) says "I invented the web". On the right, Vint Cerf is wearing a t-shirt that (on the back) says "I invented the internet".
September 15, 2025 at 7:31 AM
jenitennison.com
I'm sure Palantir creating AI summaries of our complaints will make us all feel so much more heard.

(Compare and contrast the community-based efforts of organisations like Maternity Engagement Action: maternityengagement.uk/campaign/see...)
Screenshot: The problem is that the NHS doesn’t listen well enough. The most common reasons for complaints is poor communication.195 We will update complaints regulations and increase the use of AI tools to ensure complaints data is collected - and responded to - far more quickly. One AI tool is already in development and being tested on the Federated Data Platform (FDP) and we will resource further expansion of this approach in 2025 to 2026. Screenshot: AI-led warning system building on the capabilities in the Federated Data Platform (FDP). This will mean we are able to analyse data to identify where quality issues are emerging in parts of the NHS. Where the AI system flags a concern, it could trigger a CQC inspection.
July 3, 2025 at 3:04 PM
jenitennison.com
Outside of the data and technology parts, the 10 year plan reflects a consumerist version of public voice. The power we're granted is marketised power, to choose the services we use and log our dissatisfaction with them.
Screenshot: Pushing power out to patients and the public The steps outlined in this chapter will push power from the centre to local systems and providers. This is a good thing in its own right: it is providers, not Whitehall, who know most about local need. However, it is only an intermediary in this plan’s defining goal: a patient-controlled NHS, that provides real choice, real control and real convenience for patients. To achieve that, we will introduce a new Choice Charter. It will be rolled out progressively across England, starting in the areas of highest health need. It will have 5 new mechanisms. First, it will make NHS funding flows increasingly sensitive to patient voice, choice and feedback. Patient reported outcomes and experience data - as well as direct feedback - will be important determinants in how we appraise performance and award performance-related pay. As we outline in chapter 9, we will also trial new Patient Power Payments, where patients will be contacted after care and given a say on whether the full payment for the costs of their care should be released to the provider. This will be the start of the Screenshot: trajectory of money flowing with patients, and being responsive to patient voice, over the course of this Plan. Second, patients will have new powers over how they use NHS resources from our significant expansion of personal health budgets (PHBs). As we outlined in chapter 2, we will at least double the number offered one by 2028 to 2029, and make PHBs a universal offer for all who would benefit by 2035. Third, patients will be able to exercise greater control over their health and care through the NHS App. As we outline in chapter 3, the App will revolutionise patients’ interactions with the NHS. They will be able to get rapid advice, book appointments instantly, rate their experience, choose their provider and schedule appointments - all with maximum choice. Screenshot: Fourth, over time and where safe and clinically appropriate, more patients will be able to directly refer themselves to more diagnostic services, removing often time consuming and inconvenient processes where patients are gatekept unnecessarily. In the future, we expect patients to be able to access lab testing via their phone (e.g. blood tests); to use AI to triage (e.g. via the NHS App) or to directly proceed to tests (also via the NHS App and where clinically appropriate) with tests available in the community via community diagnostic centres and neighbourhood health centres. A more pre-emptive NHS App and advances in wearable technology will, over time, support patients to self-refer confidently, including directly through the NHS App (see chapter 3). Fifth, where patients need elective treatment, they will have a choice of different providers, allowing them to choose the provider that best suits their needs, with information available on journey times, waiting times, quality, outcomes and patient experience. We will facilitate this through the NHS App.
July 3, 2025 at 3:04 PM
jenitennison.com
It's great to do user research when developing apps, but at a recent workshop we @connectedbydata.org held with @nationalvoices.bsky.social, participants said patients (and staff) need to be involved much earlier and deeper to shape the kinds of technologies that are being created in the first place
Screenshot: As the Patients Association105 and other groups have long argued, one of the most important ways we can maximise the inclusive potential of digital technology is by cocreating it with patients, especially those previously excluded by other NHS services most of all. We will involve patient groups and patient organisations in the development and testing of new App functionality from the outset, harnessing their expertise to ensure excellent, inclusive user experience.
July 3, 2025 at 3:04 PM
jenitennison.com
Practically every public dialogue I've seen also includes some strong reservations, particularly about AI adoption in places that involve sensitive data, care, and the potential of job losses. It's unusual (and convenient?!) that the 10 year plan dialogue found the public focused on the positives.
Screenshot: they readily accept the use of their data for applications beyond direct care, as long as strict privacy and security conditions are in place and met. They want to see it support the innovations that could, in future, save lives. We also heard that they would expect the NHS to derive value from its use. If NHS data supports a profitable product, the NHS should share in the proceeds. Screenshot: We have spoken to the public around the country about how they want their data to be used. We heard their desire for a rigorous approach to privacy and security, which will inform our redesign of the opt-out system. We also heard their surprise at the barriers we put up to using NHS data that does not contain any identifying information (e.g. anonymised aggregate hospital information). We will reform the legal framework to allow for health data to be used to improve the NHS and for research that benefits its patients - including, through the Health Data Research Service announced by the Prime Minister in April 2025.
July 3, 2025 at 3:04 PM
jenitennison.com
Yes, the conclusion of Chapter 8 - Powering transformation (p. 128) talks about being a "world leader" in clinical trials and public participation, but the kind of 'participation' being talked about is opting in to their data being used in clinical trials, not shaping what research gets done.
Screenshot of the text: We will support the scale of this aspiration by getting the basics right. We will be a world leader in clinical trials and patient involvement in research. The UK will have a streamlined regulatory environment, fit to support entrepreneurs and scientists. The phrase "and patient involvement in research" is highlighted. Screenshot: From 2025, we will make it far easier for patients to volunteer to participate in clinical trials by promoting ‘Be Part of Research’ Screenshot: and integrating it with the NHS App. Initially, this will allow patients to search the clinical research database and request to be contacted. We will enhance the service for commercial partners to link eligible patients who have signed up with available studies in target areas. In time, we will link clinical trial recruitment to the NHS App and SPR, so that patients can be proactively notified of clinical trials that might benefit them.
July 3, 2025 at 3:04 PM
jenitennison.com
The Devereux Review includes a lot of criticism of ONS's Integrated Data Service. Sounds like that's now going to be shelved, perhaps making way for a National Data Library with similar goals?
IDS has proved the most difficult programme for ONS. It has now delivered some ground-breaking data linkages, while protecting personal information. Building on earlier work, ONS can now, for example, assess the impact of various forms of surgery on patients’ propensity to return to work. However, at present the benefit cost ratio is very low, and IDS has low take-up by Government analysts outside ONS. It is also the case that substantial sums initially ringfenced for IDS have – with the HMT approval – been used to fund more general technology and data costs. And despite the IDS business case originally being in part predicated on the need urgently to replace the Secure Research Service, the two services are now running in parallel. where programmes are required, it is better for them to co-ordinate related activity across ONS (as ARIES), rather than gather resources into the programme and act largely independently of the rest of ONS (as many people perceive IDS); and to keep governance arrangements and project management proportionate to the task. Given that IDS also relied heavily on securing data from other Departments, several people – in and outside ONS – thought the programme initially adopted a “high handed” approach, essentially demanding data, and downplaying others’ concerns for example around security.
June 27, 2025 at 7:16 AM
jenitennison.com
Clearing out the attic and came across this from 13 years ago @johnlsheridan.bsky.social
Photo of the Jeni Tennison (Leaving) Order 2012, created by John Sheridan and colleagues at the National Archives when I left to join ODI
May 24, 2025 at 1:08 PM
jenitennison.com
UK creative industry flexing today!
Newspaper rack at a supermarket. Every paper has the same image covering the whole front page, which says "MAKE IT FAIR". The AI in FAIR is highlighted. Under it, the text says: The government is looking to change the law to favour big tech platforms so they can use British creative content to power their AI models without our permission or payment. Let's protect the creative industries – it's only fair. Creative rights in AI coalition - an initiative from the News Media Association
February 25, 2025 at 9:27 AM
jenitennison.com
At the launch of the blueprint for a modern digital government!
A paper copy of the blueprint for modern digital government, subtitled A long-term vision for digital public services, a six-point plan for reform, and the role of the new digital centre of government
January 21, 2025 at 1:55 PM
jenitennison.com
On Meta's shift to using Community Notes...

Twitter's implementation of CNs was based on @audreyt.org's work, with an algorithm designed to surface "bridge making statements" that were acceptable to people with differing views.

moda.gov.tw/en/press/bac...
Interview snippet from Audrey Tang, saying "A very similar idea, after we adopted this in our petition platform and our poll list platform, Twitter, now called X.com, adopted this idea into community notes. So, if you want to look at plurality technology currently reaching the most people, it's probably community notes. So, if you have a tweet, and it's just part of the story, or it's horribly wrong, you sometimes see a note attached to it. But it is contributed by a fellow Twitter user that managed to convince people on both aisles. So, there is an open algorithm that measures the bipartisanship, the bridge making potential of a note, and they only feature those notes as community notes."
January 9, 2025 at 10:16 PM
jenitennison.com
Xmas cracker present. Apparently it's a magic trick.
There is a spring with a ring looped onto it. Above is a piece of paper labelled "SOLUTION" which has a drawing of the ring on the spring, with an arrow appearing to rotate the ring and another arrow showing the spring moving to the right.
December 25, 2024 at 6:47 PM
jenitennison.com
Finally, while a lot of the report is focused on legal frameworks, it also recognises that having collective data rights is neither altogether necessary nor sufficient.
Even without collective data rights, it is still possible to take action against harms caused by data and AI. Data protection law, public law, equality law, consumer protection law and online safety regulation, among others, all provide mechanisms to address harms arising from algorithms, albeit imperfectly and inconsistently. Community organizing and collective pressure on companies and governments can also lead to changes. The argument is not that collective data rights are necessary for groups and communities to address the collective harms they experience, but rather that having them would make it easier to do so. Equally, having collective data rights in place would not solve all problems. Acting on collective data rights requires community-level organizing and civil society organizations that are resourced and equipped to take advantage of those rights. Public and private sector organizations using data and AI need to develop skills to conduct meaningful consultations and facilitate community deliberations. Standards are needed to support required levels of transparency. Again, these are things that can be developed alongside collective data rights frameworks.
December 19, 2024 at 12:27 PM
jenitennison.com
Based on that analysis, the report then discusses a set of collective data rights that could be introduced, the kinds of design decisions that would need to be resolved if they were, and the feasibility of introducing them given where we are today.
Ex Ante Rights Who should be able to represent a community in the exercise of ex ante data rights? How should minoritized groups be represented in decision making? Right of community control over data: What kinds of data should communities have control over (if any)? When should community consent be required to use data? When should it be unnecessary? Low feasibility. Giving communities control over data would be a substantial change to the way data and AI systems currently operate, with challenges around the definition of “community.” They are likely to be resisted. Possible starting points that could shift norms in this space would be local community control over sensor data and Indigenous data sovereignty. Right of groups to provide data for decision making: What requirement should there be to incorporate community-generated data? Medium feasibility. It would likely be impossible to require that data supplied by a community be included within data-based decision making, but a requirement for such data to be considered could be feasible. Possible starting points are around environmental data and the creation of official statistics. Right to be consulted during impact assessment: What kinds of impacts should be in scope for impact assessments? Who should be required to be involved in the impact assessment process? High feasibility. Consultation with affected stakeholders during the impact assessment process is already viewed as good practice, although data protection legislation limits this to data subjects. Possible starting points would be uses of data and AI by public sector organizations, which already have established consultation practices. Ex Post Rights Who should be able to represent a community in the exercise of ex post data rights? Right to require independent review of data-based decisions: How should reviews of automated decisions affecting communities be triggered and enacted? Medium feasibility. Data protection and other regulators already have powers of investigation, but both regulators and regulated organizations may resist the additional work that independent reviews would entail if they became widespread. Possible starting points would be industries where regulators are keen on having more investigatory powers, such as over the largest digital platforms. Right to redress for group harms: What actions should be required when collective harm is identified? High feasibility. There are already some mechanisms in place for representative action and patterns from consumer law that can be followed. Online safety law provides a starting point that could be expanded into other areas where data and AI systems are being deployed. Enabling Transparency Right to access information about data and AI systems: What information about data and AI systems should organizations be required to publish? What information about data and AI systems should organizations be required to provide on request? High feasibility. The importance of transparency is universally recognized as a driver for more trustworthy and responsible data and AI systems. Areas of contention are likely to fall in the details about the depth of transparency required. Possible starting points would be uses of data and AI by public sector bodies, which have established access to information practices. Right to explanation for automated decisions about communities: What information about particular decisions made by data and AI systems should organizations be required to provide on request? Medium feasibility. Legislation already requires explanations for automated decision making that affects individuals, which means the technology for generating explanations is being developed. However, organizations may claim that explanations are too revealing of inner system workings or too costly to provide. Possible starting points would be uses of data and AI by public sector bodies, which have established access to information practices. Right to access data used by data and AI systems: What data used by a system should an organization be required to share with the community affected by it? Low feasibility. There may be multiple reasons for not sharing nonpersonal data used in decision making about communities, including commercial confidentiality, lack of relicensing rights or concerns about re-identification. Possible starting points would be sharing information about what data sets were used, rather than the data they contain.
December 19, 2024 at 12:27 PM
jenitennison.com
The report summarises this analysis against three categories of powers that groups and communities might want to have:

- ex ante controls to shape AI systems
- ex post redress for when things go wrong
- transparency to understand what's happening and support action
Ex Ante Controls provide communities and groups an opportunity to shape the design and operation of a data or AI system prior to and during deployment. Are there legal constraints on the functioning of the system that would protect the community? Laws put in place at a national, regional or international level to protect communities against potential harms. Will the impact on the community, and groups within it, be considered during the system development? Requirements for impact assessments both before implementation and after deployment. Can the community prevent the development of the system or its use on them? Controls over which systems get created, arising from requirements for consent from the community and/or a right to object to its use. Can the community, or groups within it, determine which data is used by the system? A collective equivalent of the rights to rectification and erasure. Can the community, or groups within it, influence which mechanisms are put in place to avoid harms or realize benefits? Opportunities for consultation with groups and communities during the impact assessment process. Ex Post Redress mechanisms that provide communities and groups to gain redress for harms caused by the use of the system. Can the community require human review of the system’s decisions? Rights similar to those granted around automated decision making about individuals. Does the community have legal rights to redress for harms caused by the system? Legal rights necessary for legal action to be taken. Does the community, or do groups within the community, have standing to take legal action? Not everyone can bring legal action on every kind of issue. Often this needs to be an individual who has been directly harmed, but sometimes representative organizations have legal standing. Is it practical for the community, or groups within the community, to take legal action? Practical considerations include any time limits on when legal action needs to be taken and the costs and risks of legal action. Enabling Transparency mechanisms that ensure communities are properly informed about automated systems; it is impossible to make informed decisions, or to exercise legal rights, without transparency. Will the community be informed about the existence of the system? Requirements to inform communities about the existence of automated systems. Will the community be informed about how the system functions (including what data it uses)? Requirements to provide detail about the system, to make it easier to identify harms that might arise, or uses of data to object to. Will the community be informed about the rationale behind particular decisions? Requirements to either proactively or on demand provide details about the rationale behind a particular decision in which the system was involved. Will the community be informed about the impacts of the system? Ongoing information about the impacts of the system, to inform legal action and debate about any changes to the system. Will the community be able to access the data that is used by the system? Data access and portability rights to enable understanding and objection.
December 19, 2024 at 12:27 PM
jenitennison.com
But we already have a bunch of laws that protect us from various kinds of collective harms. Aren't they enough?

We @connectedbydata.org commissioned @awo.agency to look at three scenarios where classic data protection doesn't apply to find out.

connectedbydata.org/resources/aw...
- a police force’s use of historical crime data to determine patrol allocations in ways that increase stop-and-search use in over-policed neighbourhoods; - a train-operating company using algorithms based on historical data to determine surge prices in ways that disadvantage consumers; and - a social media company removing legitimate content in ways that undermine the free expression rights of those interested in LGBTQ+ or non-English content.
December 19, 2024 at 12:27 PM
jenitennison.com
Current data protection laws focus on the protection of personal data, and individual-level harms such as privacy breaches or faulty automated decision making.

This stems from scoping decisions made in 1980 by an OECD expert working group.

www.oecd.org/en/publicati...
Legal persons, groups and similar entities 31. Some countries consider that the protection required for data relating to individuals may be similar in nature to the protection required for data relating to business enterprises, associations and groups which may or may not possess legal personality. The experience of a number of countries also shows that it is difficult to define clearly the dividing line between personal and non-personal data. For example, data relating to a small company may also concern its owner or owners and provide personal information of a more or less sensitive nature. In such instances it may be advisable to extend to corporate entities the protection offered by rules relating primarily to personal data. 32. Similarly, it is debatable to what extent people belonging to a particular group (i.e. mentally disabled persons, immigrants, ethnic minorities) need additional protection against the dissemination of information relating to that group. 33. On the other hand, the Guidelines reflect the view that the notions of individual integrity and privacy are in many respects particular and should not be treated the same way as the integrity of a group of persons, or corporate security and confidentiality. The needs for protection are different and so are the policy frameworks within which solutions have to be formulated and interests balanced against one another. Some members of the Expert Group suggested that the possibility of extending the Guidelines to legal persons (corporations, associations) should be provided for. This suggestion has not secured a sufficient consensus. The scope of the Guidelines is therefore confined to data relating to individuals and it is left to Member countries to draw dividing lines and decide policies with regard to corporations, groups and similar bodies (cf. paragraph 49 below).
December 19, 2024 at 12:27 PM
jenitennison.com
Our poor cat Ash severely injured his back legs on Friday. One had to be amputated and the other is in plaster. It was a real relief seeing him tuck into a meal this morning.
A black cat eating a bowl of chicken pieces.
November 17, 2024 at 1:18 PM
jenitennison.com
Beech woodland on a misty autumn day
November 16, 2024 at 6:38 PM
jenitennison.com
They are gorgeous! Here's our Gordon Setter Bramble helping my eldest work from home.
My eldest is sitting on the sofa trying to do some work on her laptop. The dog is sitting on top of her looking at the laptop screen and patting the keyboard with his paws.
November 15, 2024 at 11:00 PM
jenitennison.com
Couple of references to the Federated Data Platform (Palantir) in this speech.

It would be nice if there were more details about these use cases on the FDP site, especially if they want other providers to start using it.

fdp.england.nhs.uk/home
Take those Trusts using the Federated Data Platform. Several, using it to better organise theatre lists, delivering an average of 100 more procedures every month. Chelsea and Westminster are using it to trial AI generated discharge reports. North Cumbria, using it to double the number of people on the waiting list being validated each week. And Bath, using it to help one person do the same work as it used to take multiple. All showing real benefits and freeing up staff time as well as speeding up care. Take colleagues I met recently at The Harbour mental health unit in Blackpool, using AI and automation to speed up routine tasks and release time for care, helping to drive great improvement in a trust that has struggled in the past. And take colleagues at Great Ormond Street and GPs in South West London, trialling AI voice assistance to automate notetaking. Patients and doctors both reporting benefits. So my fifth and final challenge – for now, is we need you to make full use of the opportunities we’re given, and that flow from our ability to act as one NHS. Really exciting ones, like I have already talked about the FDP. A central resource, that will underpin our ability to benefit from AI in the future, and which is already delivering real benefits now. We can’t afford to miss this opportunity to drive more co-ordinated care, and provide more productive services, at scale. Nor can we afford single organisation solutions, when this asset is there to be used. We’re ahead of plan, with 40 trusts now live, and 45 more committed. And now is the time for those remaining organisations to get on board, and to start learning from the front-running trusts, on how to make the most of it.
November 14, 2024 at 10:53 PM