## Topic - The Mayor of London's AI and Future of Work Task Force convened stakeholders from across the equality spectrum to discuss how AI is impacting different communities in London's workforce, identify groups most at risk, and develop recommendations for inclusive AI adoption that ensures no communities are left behind ## What was discussed ### Current AI impacts on different communities **Neurodivergent workers** (Carla Cantor, Hillingdon Autistic Care and Support): - AI can adapt pace, language, and format to suit individual needs and improve communication through text-to-speech and AI-assisted writing - Creates safe spaces for exploration and mental health support, enabling creativity for people with co-existing conditions - **Risks include**: dependence reducing independent thinking and problem-solving skills, AI providing incorrect information taken at face value, and social isolation where AI becomes a "best friend" - Neurodivergent candidates can use AI to write excellent applications but fail at neurotypical interviews, never progressing further - Bias and representation issues mean AI can reinforce stereotypes and misunderstand neurodivergent experiences - Inequality of access exists as not all people have equal access to AI tools or digital skills **Older workers** (Tim Whittaker, Wise Age; Maureen Childs; Liz Drury, Age UK London): - Need more analysis of older workers' profile in occupations to identify which jobs are at risk through AI - Two major concerns: ageism in recruitment being institutionalized by AI filtering out older applicants, and older workers not receiving commensurate training in new skills including AI - Older people are "absolutely terrified of AI" in most cases, only using mobile phones and getting biased views from television coverage - Example shared of bank employee telling customer "you're not suitable for online banking" based on age - Little to no training opportunities exist, with only sporadic sessions in community centers creating a widening gap - Many older people sit in community centers "like cabbages" only managing their phones, which is "very sad" given AI's potential advantages - AI could help when hands no longer work through voice prompts and be very creative, but more training is needed despite hallucinations **Women** (Mariam, Fawcett Society): - AI is entering the labor market along gendered structures and existing gendered lines - Women in admin roles within health, legal, and care support are at greater risk, with up to 40% of women in certain industries potentially displaced by 2030 - Women tend to be concentrated in tasks more at risk of automation - AI increasingly used for decision-making systems raises concerns about gender bias and discrimination risk - Up to 46% of London workers will be in roles requiring AI use for a certain percentage of tasks - Job postings now explicitly request AI-related skills - Women make up only 22% of workers within the AI field, creating design and deployment representation issues - **Scale and speed of discrimination** is concerning: past discrimination by one person is now an entire automated system impacting huge numbers simultaneously with embedded feedback loops - **Opacity problem**: equality law burden of proof is on the person believing they've been discriminated against, but without knowing an AI system is being used, there are no channels to flag unfair treatment **Deaf BSL users** (Tom Litchey and Graham Turner, British Deaf Association): - AI sign language services are being sold right now by people who themselves don't sign - Government and councils are buying these AI BSL services and signing contracts without fully knowing the quality - Example: Liverpool Street Station sign language video showed "translation isn't available" on repeat for weeks with staff not noticing - International company approached British government claiming they could provide BSL translation for all government websites in nine months - When demonstrated on 24th floor with AI experts and government observers, hearing people thought it was "amazing" and "fantastic," but as a deaf signer, Tom found it "awful" and completely missed the point - A year later, no progress has been made on that promised solution - **Risk**: if contract had been signed, deaf people across UK would not have access to government websites as promised - There's a real chance hearing people will think they can "plug in" inclusion by bolting on AI service, which would be a backward step given decades of growth in sign language learning in public services - Deaf people are always told they're "very expensive consumers" of Access to Work services, creating anxiety about rush to replace human interpreting with AI - AI services are "nowhere near capable" of handling live in-the-moment interaction enabling people to do jobs effectively **Low-income and gig economy workers** (Manny, Trust for London): - AI is the latest phase in a continuum since smartphones and the gig economy emerged: algorithmic management, robot management, lack of transparency, lack of agency - Workers have no one to talk to, losing the primary way to exercise agency through line manager conversations - Trust for London funds organizations like Organized Platform where people can log on and share how AI works in their workplace for more visibility - Worker Info Exchange has used data rights legislation to challenge employers about data ownership, useful in challenging robotic firing mechanisms **Young people from disadvantaged backgrounds** (Sahil, Streets of Growth): - Work with 15-21 year olds from Mali-Bangladeshi backgrounds who come from significant harm (child/social exploitation) - Biggest challenge is the AI application process and screening when trying to get into employment - Many AI implementations are "not great anyway" in institutions they work with, and layering AI on top of already broken processes with bias creates additional challenges - When organizations aren't organized correctly around data, AI, and bias, young people applying face compounded difficulties - **Positive development**: co-designed an AI training course with young cohorts who developed it themselves, gaining certification (e.g., Claude one-on-one certificate) - One or two individuals got interviews based on work experience plus demonstrating use of these tools and comfort talking about AI **Women with English as second language** (Liz Sewell, Berlina Grove): - Just attended celebration in Leicester supporting community organizations helping women with English as second language access AI - "Can't tell you how excited those women are" because AI has such opportunity to level up ### Training and education needs **Streets of Growth AI training model** (Sahil): - First step was organization understanding and using AI before engaging young people - Used Anthropic's Claude with not-for-profit license at low cost ($48/month for five users) - Claude has strong AI fluency website used for training, all free - Started teaching young people to move from vague prompts like "help me with my CV" to detailed prompts: "I'm 18, applying for retail assistant role, no formal experience, worked on family market store two years, write one-page CV highlighting transferable skills" - Young people use it for catching up on missed school modules, asking for quick information to get back up to speed - Contributed to Alan Milburn Review, explicitly advocating for AI capability - **Key recommendation**: free licenses from multiple providers for young people and organizations to start using tools before getting into jobs or education **BDA deaf workplace AI use** (Tom): - Informal questioning found about half of BDA's deaf staff already using AI in the workplace - Some paying out of their own pocket for AI provision - They use it as a tool for working with English language, reframing the situation - Deaf signers using it out of choice, not imposed on them - Bringing in external AI training company to build on current experiences over next 12 months - Plan to refresh AI policy and develop tips for implementation of AI controlled by deaf people in the workplace for working with English language **Training gaps and needs**: - DWP training programs have generic skills-based training but not AI training designed for novices - Older workers out of job market for some time get support on how to apply for jobs but aren't used to applying online and lack requisite bespoke training about what AI might be important - When asked in interviews about AI skills, difficult for them to answer without proper training - Need to understand needs of different groups and their attitudes toward AI - People economically inactive trying to return to work need AI training hand-in-hand with general workforce reentry support **Training on working with human interpreters** (Graham): - Most appropriate training currently is how to work effectively with human BSL interpreters, not how to "use" them like "penny-in-the-slot machines" - About human-to-human relationship: interpreters work with people first, not language - AI can't read people like humans can and won't be able to anytime soon - Training on working with interpreters as fellow human beings is most powerful way to understand contrast between human and machine services - BDA happy to engage with other organizations to deliver training contrasting what human providers offer vs. what AI is beginning to offer and which might be appropriate when ### Opportunities and benefits **Inclusion opportunities** (Devon, Entitledto): - Huge opportunities for inclusion by allowing people whose English is not first language or through adaptation - Opportunities to accelerate learning across range of areas - Risk of saying some groups can't benefit when it's more about individual appetite to adopt - Example: parents suddenly figured out mobile phones when friends were on WhatsApp because of strong driver **AI capabilities and limitations** (Devon): - As tech company leader, struggled more than imagined to see huge double-digit gains from bringing AI into organization, particularly with LLMs - Helps make engineers "a bit better" but overarching cognitive challenges still require human understanding - However, improvements are coming exponentially month after month - AI sign language solution stronger and more cost-effective than today will come with continuous improvement - What's holding back adoption is things like data access and data governance, much earlier issues than AI itself - Need to unlock earlier data infrastructure to capture AI gains - **Warning**: if focus too much on risks, miss opportunities, particularly for people on edges or struggling to take full advantage **Positive use cases**: - Flexible working opportunities: helping workers plan shifts, see shifts, get visibility over earnings - For people with disabilities, AI can be assistive tech allowing adjustments and equity in workplace, but blanket bans on AI agents prevent use as assistive tech - Can help with time management, tasks, and communications for neurodivergent people - Unseen economy support: helping older carers, childcare providers, volunteers through simple reminders (e.g., taking tablets) enabling them to continue contributing ### Risks and concerns **Discrimination and bias**: - Around 30% of jobs in UK likely to be impacted by AI, and these are jobs done by people least represented in workforce - AI reinforces existing inequalities and concentrates power, profit, and information - Requires instrumental government action to make it rebalancing force rather than concentrating force - Female-dominated occupations face three times the AI exposure compared to male-dominated ones - Disabled and neurodivergent communities fear AI being used by government will further disadvantage them - AI tools often trained by data not representative of disabled and neurodivergent communities - Fear of being denied workplace adjustments because AI informs employers these aren't needed **Quality and transparency issues**: - AI quality doesn't seem to be there yet in some areas, not ready for like-for-like swap in many roles - Lot of uncertainty around future trajectories with very wide spectrum of estimates and lack of robustness in studies - Lot of hype, so doesn't mean there won't be big changes but need to prepare for real spectrum - "AI theater" or perception as reality: people not rehiring in roles might be directly due to AI or because AI noise makes them reluctant - AI is being overvalued currently, need to think about long-term stable trajectory but also likelihood of market correction/crash **Privacy and data concerns** (Mark): - Can appear to work on surface but that's "tip of the iceberg" - Repeating what happened with social media: extraction, privacy, data given away in return for convenience and fun - AI is another vehicle to collect data from users benefiting others rather than themselves - Lack of transparency about data collection and training on user inputs - Usually provided by those not part of usual GDPR and data protection cultures - Lot of hype and over-promise, false marketing claims, big businesses not making money but getting data - **Solutions**: local-first AI running on own machines, open source AI and models that maintain privacy **Procurement and implementation issues**: - Blanket procurement for AI in workplace fails to account for lived experience and co-production in decision-making - Government and councils buying AI services without involving communities in design, delivery, and evaluation - London councils are "very patchy and disjointed," often reinventing wheels and processes on sign language provision - Asking local deaf people about big strategic questions "not really the right way to work" **Job quality and displacement**: - Real question is job quality alongside job replacement - In early experimentation period, frontline workers getting agency to decide how to use AI, but when managers get involved or thinking about future trajectories, this becomes more solidified - Question of "if there are productivity gains, to who do they flow" and "if there are agency gains, where does that go" - Entry-level jobs disappearing, replaced by AI - Job fragmentation and increasing alienation: people increasingly asked to input data into AI algorithms rather than having control over processes or work ### Regulatory and policy recommendations **Need for vision and strategy** (Daniel, Citizens Advice London): - Real lack of vision and strategy across London for how AI will impact work and employment - AI increases fragmentation and alienation of work - Without vision, London risks being in reactive situation rather than controlling environment, similar to pandemic exposing over-reliance on entertainment and hospitality - AI could be used to reduce mundane tasks and make work more pleasurable and interesting, but "we're not grasping that agenda" - Co-production principle needs to transfer into employment: employers should not introduce AI without consultation with workforces and worker representatives - Without strong principles, risk of increasing alienation and unemployment **Deaf leadership and coordination** (Tom and Graham): - Want to establish deaf leadership strategy working group in next 12 months involving councils, PCTs, and key stakeholders to consider how to get best advantage from spend on AI BSL - Received funding to develop deaf leadership in London, happy to build on this in partnership with GLA - Cambridge University partnership produced paper on how government can procure AI sign language solutions in equitable way involving deaf leadership - Partnership with Cabinet Office government team on same issue **Systems change approach** (Jasbinder, GLA): - Using systems-wide approach covering partners, sectors, and organizations together - Way of facilitating that pulls out fear and power dynamics people think is happening with big tech, bringing person back to center - Reorientating system without necessarily spending more money, using resources already available - Private sector approaches like scenario testing: instead of building something and being afraid, test through scenarios to see if suitable for communities - Addresses issues and concerns around bias and exclusion within process **Communities of practice and peer networks**: - Peer-to-peer networking in neurodivergent communities: casual groups sharing how AI helps and hasn't - Respect from someone with similar lived experience who understands what you're planning to use AI for - "Positive deviance": peers who have harnessed AI to their needs can share with those facing similar challenges but haven't been on that journey yet - Most sustainable way of bringing people together to help make sense of changing landscape **Training must develop critical understanding** (Dan, Charity Digital): - Being clear about data use means looking for critical understanding of tools, not just adoption - May well be use but also being aware of trade-offs and choices made as benefiting from them - Training needs to give clear sense of balance between costs and benefits and knowing what we're doing to balance them **Government engagement opportunities**: - DWP and others planning call for evidence across how tech benefits lives properly with opportunities to feed in different perspectives ### Charity sector specific challenges **Adoption patterns** (Dan, Charity Digital): - Supported about 70,000 charity practitioners last year - Really mixed emotional responses to AI: from super excited to actively resistant and fearful - This hasn't changed over last three years monitoring - **Critical gap**: 75% of charity practitioners using AI but only 45% of their organizations have policies or training in place - Only 4% say they have sufficient support and training - Most adoption led by individuals, not organizations, leading to uneven adoption and uneven emotional responses within organizations - Challenge for leaders: how to help organizations make sense of broadening experience and emotional responses **Missing support infrastructure**: - Loads of brilliant resources made by charities for each other - **Gap**: trusted, independent, and stable advice and support - Lots of resources being published that are changing, lots created by tech companies, but something trusted, independent, and stable is missing - Need to build greater resilience within sector by not being overly reliant on one provider - Conversations demanding participation in conversations about features to bring about, not just how to use AI or not ### Cross-cutting themes **Data and infrastructure prerequisites**: - Data access and data governance are earlier issues holding back AI adoption more than AI itself - Need to do earlier work around data and infrastructure to capture AI gains **Individual vs. organizational readiness**: - More about individual appetite to adopt than group characteristics - Organizations need to be ready alongside individuals having skills **AI as concentrating vs. leveling force**: - AI broadly concentrating inequality, power, profit, and information - But can be leveling up tool with right investment and market shaping - Women returning to work after care or young people being upskilled as digital experts are ways to counteract concerns **Big tech concerns**: - Lot of tech promise being sold by big tech with their own agenda to maximize profit - Not necessarily interested in what they think of as marginalized markets and segments - Essential for lived experience to be involved at producer side to ensure things are properly harnessed to needs of everyone - Can't just leave it to typical big tech because it won't reach everybody - Need to balance social impact against profit and enable more people with different ways of looking at this to be involved in designing tech ## Takeaway / Understanding The consultation revealed AI is already having significant impacts across London's workforce, with effects varying dramatically by community. While AI offers genuine opportunities for inclusion and leveling up—particularly for people with disabilities, those with English as second language, and young people gaining new skills—serious risks exist around discrimination, job quality, privacy, and exclusion. The most vulnerable are those already underrepresented in the workforce: older workers facing ageism, women in female-dominated occupations, deaf BSL users being sold inadequate AI services, low-income workers experiencing algorithmic management, and neurodivergent individuals facing both opportunities and risks. Critical gaps include lack of coordinated vision and strategy, insufficient training infrastructure, procurement processes that exclude community involvement, and adoption happening at individual rather than organizational level. Success requires co-production, deaf and disabled leadership, peer support networks, critical AI literacy (not just adoption), and ensuring AI serves social impact rather than only profit maximization. ## Action items - Share meeting recording and summary note with AI and Future of Work Task Force to ensure equality perspectives are included in all planning - Share information about broader government call for evidence on how tech benefits lives when details become available - Follow up with Tom Litchey (BDA) about deaf leadership work and potential partnership with GLA on developing strategy working group over next 12 months - Disseminate information about Anthropic's Claude AI fluency training resources and not-for-profit licensing options to organizations working with disadvantaged groups - Consider how to support development of trusted, independent, and stable AI advice infrastructure for charity sector - Explore how to integrate AI training with economic inactivity programs to support people returning to workforce - Investigate establishing communities of practice and peer networks for different equality groups to share AI experiences - Consider assessment framework for determining when human services vs. AI are appropriate, particularly for access services like BSL interpreting