A year on from our first TUC Digital Lab workshop exploring how unions might make use of Artificial Intelligence, we got together again to take a deeper dive into the topic.
We were joined by Hannah O’Rourke of progressive digital network Campaign Lab to look at what unions had learned in the last year about whether and how they should develop internal usage of AI.
What do we mean by AI in unions?
When we think about a technology like AI, which has such far-reaching implications, we need to look at how it might impact unions at a number of levels. Our friends at the non-profit digitisation organisation CAST have developed a helpful framework for this (btw, CAST have lots of helpful resources on AI).
You can look at AI’s impact in layers like an onion. At the heart are ways that AI might help add productivity to existing tasks done in the union. This is where much of the consumer hype is, with tools like ChatGPT or Copilot marketed to automate or augment parts of regular workflows. That could include things like partially automating the production of minutes from union meetings, to reduce the time the job takes.
Next we looked at how having AI assistance might help extend existing work into other areas. For example, a rep might be better able to engage with a new group of migrant workers at their workplace if they are able to access machine translation on an ad-hoc basis. Are there potentially new ways of working for the union that might fit within our existing mission?
Then there are questions of how more widespread use of AI might change the union’s mission and demand a response from the union. For example, if a key part of the union’s offer is advice at work, and an AI assistant on a phone can produce convincing-sounding, highly tailored advice for workers (regardless of whether it’s actually the best option), how might that affect how prospective members see the value of joining?
And finally we need to consider responses to AI as it is affecting society. Even though the level of actual job displacement from AI is still unclear, AI is being used as an excuse by employers seeking to cut staff and deregulate. Creative sectors are suffering from the increasing use of AI in imagery and writing, where new tools have been based on workers’ intellectual property, without their agreement or compensation. And AI combined with data surveillance is increasing the information asymmetry between workers and their employers, with employers using this to extract a greater share of value from work. Each union will be facing a different set of challenges for their members and their industries, and will need to develop a policy response. For some this could mean not engaging at all – for others it might mean engaging only within strict guardrails.
Viewed in layers like this, it’s clearer that our how we make internal use of these tools has to fit within the framework of our own external policy. We’ll need internal policies or principles that make sure we’re in alignment. And we can also see that the more we learn about the potential of the technology through experimentation, the better we’ll understand how it works and what the bigger implications could be for the union or our sectors.
Where are unions at the moment?
Another helpful framework from CAST is their stages of AI adoption. At the time of last year’s workshop, most unions were at the Aware stage, where only a few people in the organisation were starting to look at how tools might be helpful in a practical sense.
From this, non-profits that CAST work with typically follow a path of deeper stages. There’s Testing, where more people experiment, but work is in silos and starts to highlight potential risks and ethical concerns. The Aligning stage helps bring experimentation together, developing guardrails for the organisation. At the Integrating stage, the organisation more consistently thinks about supporting use cases which have been proven. And at the Transforming stage, the organisation’s use of AI has moved into deeper areas of changing services offered or the wider mission
Unions in the workshop mostly felt that they were at the Testing stage. Some still felt that they hadn’t moved on from the beginning of the process, but more believed that staff and reps were already experimenting in silos, with or without organisational approval and oversight. A handful of unions were working towards Aligning, by developing pilot projects or principles and guidance to align experimentation better across the union.
What’s been happening in the last year?
We had a recap of work that’s happened since the last workshop, across the TUC, our affiliates and around the international movement.
Within the Digital Lab
At our workshop, we picked up a number of projects in the TUC Digital Lab, based on the requests we received:
- We published a context guide for unions to use in explaining basic aspects of Generative AI around their organisations.
- We supported PCS to experiment with AI in training.
- We worked with tech law experts AWO to develop a resource exploring the legal and regulatory implications of Generative AI, to support unions in developing policies or reps in understanding the implications of new systems being introduced by employers.
Policy work
More broadly, the TUC’s policy team and AI working group drawn from across affiliates have expanded our shared policy work to include:
- A draft Bill for regulating AI, which has formed a central plank of our lobbying in this area.
- A manifesto for creative workers, on how we could address some of the harms in the ways this technology is being developed.
- A toolkit for reps in conducting and using algorithmic impact assessments for AI systems in the workplace.
- Setting out the building blocks for a wider pro-worker AI innovation strategy for the UK.
Union case studies
In the workshop, we looked in more depth at three AI projects, and the learning from them.
- PCS: Using AI chat in union education.
PCS worked with us, Campaign Lab and developer Poteris to pilot an AI conversation practice tool to support their organising training for reps. It simulates a workplace conversation about joining the union. Used by reps as a follow up to training, it offers custom feedback, based on the learning objectives from the course. The aim is to augment rather than replace or dilute the union’s learning offer. It’s been written up as a case study here. Next steps for PCS are to look at other potential scenarios where conversation practice is useful. Poteris have released the work as open source for other unions to use, or are developing a supported service for unions based on it, and expanding the tool to allow quicker scenario creation, member login integrations and data dashboards for union education teams. - NASUWT: AI chat for member advice.
NASUWT have piloted offering tailored information on one the advice pages of their website, using an option for chatbot support. This isn’t designed to offer casework advice, which needs to come from a trained source in the union, but it can help answer informational questions on work conditions. These are often tricky in teaching, due to the fragmented nature of contracts and regulations, but by restricting a chatbot to generate replies based only on the pre-approved page content, the union can be more sure that it can only give accurate responses. As teachers are limited in the times they can contact NASUWT advisors during the day, there are peak loads at lunchtimes and at school closing. Answering more basic questions automatically in this way can free up the union’s limited advisor time for more complicated or urgent member questions. This project involved a lot of time in building a chatbot that could match member intents behind questions, and in working with contact centre staff to build in awareness of some issues that a trained advisor would pick up, such as safeguarding signposting for members who might be feeling bullied. There is a more in-depth case study on developer EBM’s website. - Prospect: Trialing generative AI for productivity
Like several other unions, Prospect have established a formal pilot process to help evaluate whether there is potential for the union in using generative AI assistants, and to do it in a potentially safer and more controlled way. A cross-departmental AI working group of staff has been set up to support each other in trialling Copilot 365. The decision was taken to use Copilot only, to protect Prospect data from being shared outside the union’s own cloud provision, or being used to train suppliers’ models. Use cases being tested include generating Excel or PowerBI formulas to help staff use the tools at a more advanced level, and reducing the time taken to make minutes from recorded meetings. It is not being used in any member-facing activity at least until the union knows more about it. The experiment is running over a defined period and will help the union to devise ongoing internal policies about how AI use might be permitted internally.
We also looked at wider union and social sector innovations from other countries, such as a contract bot that helps reps compare an offer from the employer against the union’s standard recommended agreements, an email prioritisation system that reads and categorises incoming member email to help advisors work with it more effectively, or revising union websites to be more likely to be reflected in generative AI search answers. We also looked at other NGO examples, such as Citizens’ Advice’s tool for advisors to generate case notes quickly from video calls.
Potential transferable useful use cases for AI across unions
After looking at examples and discussing participants’ own experiences, including what had gone well so far, and where we knew AI had particular risks or shortcomings, we did some exercises to ideate and prioritise potential use cases.
We then plotted them on charts as to which offered potentially high impact (whether dealing with difficult tasks or easy but very frequent tasks), and whether it would be easier or harder to implement. The idea was that rather than look at some of the biggest and most ambitious targets, or the most widely hyped but maybe less useful, we could devise a number of short-term and lower risk ideas for unions to test, that might help us understand the tech better, its organisational implications, and its potential for value.
Here are a selection of use case ideas which participants thought would be both impactful and achievable.
- Transcribing meetings and producing minutes.
- Summarising governance conversations to generate reports appropriate to different groups of stakeholders.
- Planning or drafting longer papers and internal reports informed by existing union content.
- Information chatbots to help users answer questions from extensive content sections on union websites.
- Simulating conversations for training activists.
- Providing custom recommendations and learning journey support for activists.
- Data analysis to spot patterns in density, or for organising resource allocation.
- Analysing background information and suggesting questions to prepare better design briefs for union comms teams.
- First drafts of briefings for different internal stakeholders from new policy or campaign documents.
- Internal upskilling and customised IT help across an increasing range of tech tools.
What might stop us?
We looked at some of the factors that could block this kind of experimentation in the union, and whether they could be mitigated satisfactorily.
| Blockers | Mitigations |
| Lack of specialist staff Most unions do not have enough good tech and data specialists in-house, to let them accurately evaluate opportunities or develop interventions. | Make use of the knowledge around the wider movement. Cross-union sharing of experiences. |
| Gaps in data quality and security AI can only be as good as the data it works on. It’s not a panacea to smooth over poorly curated sets of data or content. Using AI tools before being sure of your data environment could lead to answers being given to users revealing insights from data they didn’t originally have permissions to see. | Using models linked to data permission structures in the union (eg Copilot for MS clients). Reviewing data permissions thoroughly before starting. Work on the union’s data strategy. |
| Not everyone is on the same page There is a lot of fear or scepticism of AI amongst staff and activists. And conversely there is a lot of boosterism as people get more used to using AI assistants in their personal life, and seek to bring that into the union without considering the implications. Establishing and communicating a path appropriate for the union is going to be a change management task in itself. | More training, awareness and discussion within the union. More visibility for existing and new materials on the subject (from unions or Digital Lab). |
| Lack of knowledge Unions are still building their understanding in this area, outside of tech-related roles. That can be especially a concern in an environment where there is so much hype around products that are often oversold, and where there is considerable risk in doing it wrong. The risks are real, but there is also a fear of the unknown that could stifle innovation. | Share knowledge between unions on useful technologies and cost-effective/reliable suppliers. Better documentation of case studies. Shared learning internally across the union’s teams. Think about the union’s principles for internal AI and be guided by them when operating in areas of uncertainty. |
| Silo working Union teams often work in isolation, which makes it hard to share knowledge on developments that might apply across the union’s different functions. | Structured conversations across teams. Cross-function learning. |
| Leadership Related to silo working, AI is a change management issue for the whole organisation rather than a function-related one. As such it is hard to identify where leadership should sit, when tech does not have a senior seat in management. Sponsorship needs to come from a senior level, where leaders may not yet be engaging in issues around AI. | Develop leader-level briefings and resources. |
| Money/resources No unions have excess money to spend on licensing, or time to support organisational roll-outs if they aren’t going to be worth it. Pricing is always changing and could become much more expensive if the current AI bubble bursts. We need to better understand what kind of value we can expect and whether it’s worth it. | Running pilots to better understand the value of generative AI use, before spending larger amounts of money and time. Work to a pace and extent appropriate for the union’s tech strategy, rather than feeling pressured by hype to adopt every new tool. |
| Lack of policies Unions will need to develop policies that fit their own understanding of the risks, and which address their own particular industrial concerns. Whether restrictive or enabling, there should be guidance for staff and activists on what is expected of them. It’s hard to know where to start in an area that is changing so rapidly. But not addressing this leaves the union open to huge risk if AI is used inappropriately. | Develop internal policies and manage usage. Shared union internal AI policy development. Review policies regularly for such a rapidly changing area. |
| Campaigns & ethics compliance This is an area where there is greater read-across from union policy to practice than there is with many other areas of digitisation or operational matters. Making mistakes in internal adoption of AI could compromise the integrity of our external policy and damage the union’s reputation. There are also risks around data protection and data security, where unaligned experimentation could cause compliance problems for the union. | Engage in shared policy work between unions through TUC working groups. Involve policy leads in devising principles and internal policies, and on internal working groups. Understand member concerns and engage them in policy work. Consider energy use / sustainability as a metric in evaluating use case pilots. |
What could the TUC Digital Lab do to help?
We asked the group where they would like the TUC to take this work next and got a few helpful suggestions that we could investigate.
- Shared work on supporting unions to develop their own internal AI usage policies.
- Networking together union HR managers to discuss what’s happening in recruitment, as more job candidates make use of AI in applications and tests.
- Supporting unions to make use of the work from PCS’ Repcoach pilot, either as Open Source or a more developed SaaS product.
- Developing shared supplier references for unions in procurement around AI.
- Increasing the Digital Lab’s case studies resource to feature more potentially transferable AI ideas.
- Publishing guidance and case studies around internal Gen AI adoption pilots for unions.
- Looking into union experiences and recommendations around Generative Engine Optimisation for websites.
- Offer exploration sessions for leadership groups within affiliates to develop their own understanding in a similar way to these two cross-union workshops.
If you or your union would be interested in getting involved in projects around any of these ideas, please get in touch.