AI Ethics and Bias: What Artificial Intelligence Is Doing to Society, and What Nobody Is Telling You
Technology · Guide
AI Ethics and Bias: What Artificial Intelligence Is Doing to Society, and What Nobody Is Telling You
AI systems now make decisions about your credit score, your job application, your benefits entitlement and the news you read. Most of the time, you will not know this is happening. This plain-English guide explains how AI bias works, who it harms, what digital exclusion means for people without digital skills, and what the law currently does and does not protect you against.
Quick read: key points
- AI bias is not science fiction. It is the documented tendency of automated systems to produce outcomes that are systematically unfair to particular groups, usually because the data they were trained on reflects existing human discrimination. The consequences range from unfair loan refusals to wrongful arrests to benefit denials.
- Approximately 19 million people in the UK are digitally excluded. Around 9 million adults lack the basic digital skills considered essential for employment. These people are not simply missing out on convenience. They are being actively disadvantaged in systems that assume digital access as a baseline.
- A February 2026 Amnesty International report found that the UK government’s unchecked use of AI and automated systems in the social security system is leading to the exclusion of people with disabilities and other marginalised groups, with serious consequences for claimants including erroneous benefit decisions.
- Filter bubbles and echo chambers created by AI recommendation algorithms are measurably contributing to political polarisation, the spread of misinformation and declining public trust in democratic institutions. A 2025 parliamentary inquiry described this as a threat to UK democracy.
- As of June 2026, the UK has no dedicated AI Act. The EU passed one in 2024. The UK’s approach is deliberately light-touch: five non-statutory principles applied by existing sector regulators. 85% of the UK public believe the UK needs strong laws to force companies to make AI safe. The government has signalled a comprehensive AI Bill may come in 2026 or 2027.
- The groups most harmed by AI bias and digital exclusion are the same groups already disadvantaged by other forms of inequality: older people, disabled people, low-income households, ethnic minorities and people in rural areas. AI does not create these inequalities. It amplifies them.
Key terms
| Term | What it means |
|---|---|
| Artificial intelligence (AI) | Computer systems that perform tasks that would previously have required human intelligence: recognising images, processing language, making predictions, screening applications and recommending content. AI systems learn from data. What they learn depends entirely on what data they are given. |
| Algorithmic bias | The systematic tendency of an automated decision-making system to produce outcomes that are unfair to particular groups. Bias in AI systems usually comes from biased training data, biased design choices or biased assumptions baked into the model. It is often invisible to the people affected by it. |
| Training data | The dataset an AI system learns from. If the training data reflects historical human discrimination, the AI system will learn those discriminatory patterns and reproduce them. Amazon’s recruitment AI, for example, trained on ten years of CVs that were predominantly from men, and learned to penalise CVs that included the word women or came from all-female colleges. |
| Automated decision-making (ADM) | A process in which a decision affecting a person is made entirely or substantially by a computer system without meaningful human review. ADM systems are used in credit scoring, benefits assessment, recruitment, insurance pricing and criminal justice risk assessment. |
| Digital exclusion | The condition of being unable to participate fully in digital society because of a lack of internet access, a lack of devices, a lack of digital skills or some combination of these. Digitally excluded people are two to three times more likely to be unemployed. |
| Filter bubble | A state in which a person is exposed primarily to information that confirms their existing views, because recommendation algorithms show them more of what they already engage with. Filter bubbles are not created by users choosing to avoid challenging content. They are produced automatically by AI optimising for engagement. |
| Echo chamber | A social environment in which views are amplified and reinforced by repetition, with little exposure to contrary perspectives. Related to filter bubbles but involves social reinforcement rather than algorithmic curation alone. |
| Deepfake | An AI-generated video, audio or image that realistically depicts a real person saying or doing something they did not say or do. Deepfakes crossed a quality threshold in 2025 and 2026 that made them difficult to distinguish from genuine footage without specialist tools. |
| Large language model (LLM) | A type of AI system trained on large volumes of text to understand and generate human language. Systems such as ChatGPT, Claude and Gemini are built on LLMs. They can produce fluent, confident-sounding text that is factually incorrect, a problem known as hallucination. |
How AI bias works: the basic mechanism
AI systems do not think. They find patterns in data and apply those patterns to new situations. If the data reflects historical human behaviour, and if historical human behaviour included discrimination, the AI system will find the discriminatory patterns and apply them. This is not a flaw that can simply be fixed by making better AI. It is a consequence of using historical data to predict future outcomes in a world where the past was unfair.
The process has three failure points. First, the training data can be biased: it does not represent all groups equally, or it contains historical outcomes shaped by discrimination. Second, the features used to make predictions can be proxies for protected characteristics: using postcode as a predictor of creditworthiness, for example, effectively uses race and class as inputs even when they are not explicitly included. Third, the people who design the systems can introduce their own assumptions, blind spots and priorities into the model without being aware they have done so.
A documented real-world example: Amazon’s hiring AI
Amazon developed an AI recruitment tool to screen CVs and rank candidates. It was trained on ten years of CVs submitted to Amazon, most of which came from men, because the technology sector is predominantly male. The system learned that male CVs were correlated with successful hires and began to penalise CVs that included the word women or applications from all-female colleges. Amazon scrapped the tool when the bias was discovered. The episode illustrates how AI can embed and amplify existing discrimination without any intention to discriminate.
Where AI decisions affect ordinary people in the UK right now
Credit scoring: lenders use AI to decide whether to offer credit and at what rate. Benefits and welfare: automated systems flag claims for review or denial. Job applications: CV-screening tools sort applications before a human sees them. Insurance pricing: AI calculates premiums based on data about your behaviour and postcode. Healthcare triage: NHS digital tools help prioritise referrals and diagnoses. Criminal justice: risk assessment tools inform sentencing and bail decisions. Content moderation: automated systems remove or promote online content. In each of these domains, an algorithmic decision can have life-changing consequences, and most people affected will never know the decision was made by a machine.
Digital exclusion: the people AI leaves behind
The debate about AI ethics tends to focus on the people who use AI and are harmed by its bias. But there is a second, less visible group: the people who are not using AI at all, and who are being harmed by a world that increasingly assumes they do.
| Group | Scale of digital exclusion | Practical consequence |
|---|---|---|
| People without internet access | Approximately 1.52 million people in the UK are completely offline (Ofcom, 2024) | Cannot access online-only government services, benefit applications, NHS appointment booking or job application portals |
| People without basic digital skills | Approximately 9 million adults in the UK lack basic digital skills essential for employment; approximately 8.5 million lack foundation-level digital skills for everyday life | Two to three times more likely to be unemployed. Cannot use digital tools that are increasingly required for participation in society |
| Benefit claimants | 36% of Pension Credit customers were offline in 2024 (DWP); 46% of Employment and Support Allowance customers would need help from others to apply for or manage benefits online | Dependent on telephone or in-person services that are being reduced. At higher risk of errors in automated benefit assessments |
| Older people | Significantly higher rates of digital exclusion than the general population; 21% of Attendance Allowance customers reported declining internet use in 2024 | Excluded from digital services while being among the highest users of public services |
| People with disabilities | Higher rates of digital exclusion due to accessibility barriers as well as income and skills gaps | Amnesty International’s February 2026 report found this group disproportionately harmed by automated social security systems |
| Rural residents | Only 45% of rural premises have gigabit-capable connectivity versus 78% of all residential premises | Structural connectivity disadvantage that compounds skills and income barriers |
The Digital Poverty Alliance estimates that more than 19 million people across the UK remain digitally excluded in some form. The government’s Digital Inclusion Action Plan, published in February 2025, acknowledged digital inclusion as a cross-government priority and identified low-income households and disabled people as two of the priority groups. However, the plan’s commitments are primarily voluntary industry pledges rather than statutory requirements.
AI in public services: the specific harm to vulnerable people
Amnesty International published a significant report in February 2026, based on research conducted between October 2024 and January 2025 with 782 people. Its conclusion was direct: the UK government’s unchecked use of AI and automated systems is leading to the exclusion of people with disabilities and other marginalised groups, with serious consequences for claimants in the social security system.
What the research found
Automated systems and AI in the assessment and provision of social security introduce a significant risk of errors in decision-making due to biased or discriminatory algorithms. Digital exclusion can result from a person’s living conditions, educational attainment, health status and income levels, and these are complex factors not fully captured by automated social security systems. One woman interviewed described how gender and socioeconomic status created barriers to accessing services online. The report described the situation as a perfect storm, in which pre-existing flaws in the social security system are being exacerbated by technology, and new problems linked to these technologies are being created.
Information bias, echo chambers and what AI does to your view of the world
Beyond decisions made about individual people, AI has a second, broader effect on society: it shapes what information people see, what they come to believe is true, and how they understand people who are different from them.
How recommendation algorithms create filter bubbles
Every major social media platform, news aggregator and video streaming service uses AI to decide what content to show you. These systems are optimised for engagement, which means they show you more of what you already clicked on and less of what you scrolled past. Over time, this creates a personalised information environment in which most of what you see confirms your existing views. Research published in AI and Society in 2025 confirmed that algorithmic personalisation reinforces echo chambers and filter bubbles, creating spaces where misinformation persists and thrives. You did not choose this environment. The algorithm created it to keep you on the platform longer.
Deepfakes and synthetic media
In Ireland’s 2025 presidential election, a deepfake video falsely depicted the eventual winner withdrawing his candidature, including fake footage of national broadcasters confirming the news. This was released days before polling day. The World Economic Forum reported in March 2026 that deepfakes crossed a critical quality threshold in 2025 and 2026, with earlier tell-tale glitches eliminated. They are now accessible to anyone with a smartphone. During elections in 2024 and 2025, cloned voices and visual persona deepfakes became a live feature of democratic politics across multiple countries, including the UK. A 2025 parliamentary inquiry described disinformation, including AI-generated disinformation, as a threat to UK democracy.
AI-generated misinformation
Large language models produce fluent, confident-sounding text that is frequently factually incorrect. A 2025 survey by the European Broadcasting Union evaluated over 3,000 AI responses assessed by professional journalists. Nearly half had at least one significant problem and a fifth contained major errors on the accuracy of information. The problem is not that AI lies deliberately. It is that these systems have no understanding of truth. They generate statistically likely sequences of words, not accurate accounts of reality. When people cannot distinguish AI-generated content from genuine reporting, misinformation produced at scale becomes a structural threat to the public’s ability to make informed decisions.
The UK’s approach to AI regulation: where things stand
As of June 2026, the UK has no dedicated AI Act. The EU passed a comprehensive, risk-based AI Act in August 2024, with high-risk AI systems in areas including employment, credit, education and benefits facing strict obligations. The UK’s approach has been deliberately different.
| Element | Current status |
|---|---|
| UK AI Act | Does not exist as of June 2026. The government has signalled a comprehensive AI Bill may be introduced in 2026 or 2027 but no Bill is currently before Parliament. |
| Five AI principles | Safety and robustness, transparency, fairness, accountability, and contestability and redress. Set out in the March 2023 White Paper. Non-statutory: they are not backed by law and cannot be directly enforced. |
| Existing sector regulators | The ICO (data protection), Ofcom (online safety and telecoms), FCA (financial services), CMA (competition and consumer), MHRA (healthcare). Each applies the AI principles within their existing remit. |
| Data (Use and Access) Act 2025 | Received Royal Assent June 2025, in force from 5 February 2026. Eases constraints on automated decision-making while preserving some safeguards. Requires the ICO to produce a statutory Code of Practice on AI and automated decision-making by May 2026. |
| Online Safety Act 2023 | Applies to AI chatbots and AI-generated content on regulated platforms. Requires platforms to assess and mitigate risks. Protections for illegal content came into force spring 2025; protections for children in summer 2025. |
| AI Security Institute | Formerly the AI Safety Institute. Renamed in February 2025 to focus on national security risks. Critics warn this narrows attention away from ethics, bias and rights. |
| Public opinion | 85% of UK adults agreed in a June 2026 survey that the UK needs strong laws to force companies to make AI safe and secure, and that voluntary guidelines are not enough. |
The case for the UK’s light-touch approach
The government argues that a principles-based, sector-led approach allows regulation to adapt to a rapidly changing technology without locking in rules that quickly become obsolete. It allows the UK to attract AI investment from global technology companies that might be deterred by prescriptive legislation. The pro-innovation stance is intended to position the UK as a global leader in AI development. The AI Opportunities Action Plan, published in January 2025, set out 50 recommendations for boosting the UK AI sector, including dedicated AI growth zones and a National Data Library.
The case for stronger statutory protection
Critics, including 85% of the UK public in the June 2026 AI Compass survey, argue that voluntary principles are insufficient to protect people from the documented harms of AI bias and automated decision-making. The EU AI Act bans specific high-risk uses outright, including social scoring and unconsented biometric surveillance in public spaces. The UK has no equivalent ban. Amnesty International’s February 2026 report documented real harm to vulnerable people from automated government systems that currently face no specific statutory constraint. Renaming the AI Safety Institute as the AI Security Institute, critics argue, signals a narrowing of focus away from the ethics and rights concerns that affect the most vulnerable people. The gap between the 85% who want strong laws and the government’s position is one of the widest in any policy area.
Your rights when AI makes decisions about you
| Situation | Your rights under current UK law |
|---|---|
| An automated system made a significant decision about you | Under UK GDPR (as amended by the Data Use and Access Act 2025), you have the right to know when a significant decision has been made about you using automated processing. You have the right to request human review and the right to contest the decision. The ICO’s draft guidance on automated decision-making under Articles 22A to 22D sets out what organisations must do. |
| You think an algorithm was used to discriminate against you | The Equality Act 2010 applies to AI-mediated discrimination in employment, credit, housing and services. If an algorithm produces a discriminatory outcome, the outcome is unlawful regardless of whether the algorithm intended to discriminate. Contact ACAS (employment) or Citizens Advice (other areas). |
| You were harmed by misinformation online | The Online Safety Act 2023 requires regulated platforms to take measures against illegal content including false communications. The Act is enforced by Ofcom. You can report content to the platform and, if the platform fails to act, to Ofcom. |
| An AI system was used in a benefit decision | You have the right to request mandatory reconsideration of any DWP decision. You can request that the decision be reviewed by a human. If you are refused, you can appeal to an independent tribunal. Citizens Advice and your local MP’s office can help. |
Practical steps: protecting yourself in an AI world
| Step | What it involves |
|---|---|
| Ask whether AI is being used | When any significant decision is made about you, you have the right to ask whether automated processing was involved. This applies to benefits, credit, insurance, recruitment and healthcare. Ask the question. You have a legal right to the answer. |
| Check your information sources | Actively seek news from sources with different political perspectives. Use fact-checking resources including Full Fact (fullfact.org), the BBC Reality Check and Reuters Fact Check. Be sceptical of content that produces a strong emotional reaction before you have verified it. |
| Check your credit report | Your credit report is produced by an algorithm. You can access it free of charge through Experian, Equifax or TransUnion. If there are errors, you have the right to have them corrected and they may have caused you to be unfairly refused credit. |
| Build or strengthen your digital skills | The Good Things Foundation (goodthingsfoundation.org) runs a national network of community digital skills support. Learn My Way (learnmyway.com) offers free online courses. Your local library, community centre or Citizens Advice may offer in-person digital support. |
| Report content that misleads | On social media, report content you believe to be false. On news platforms, check sources before sharing. Every piece of misinformation you do not share reduces the scale of its reach. |
Free support organisations
| Organisation | What they offer | Contact |
|---|---|---|
| Good Things Foundation | Runs the UK’s largest network of digital skills support through community organisations. Operates the Learn My Way platform for free online digital skills courses. | goodthingsfoundation.org. learnmyway.com |
| Digital Poverty Alliance | Campaigning and policy organisation working to end digital exclusion. Publishes evidence on the scale and impact of digital poverty in the UK. | digitalpovertyalliance.org |
| Information Commissioner’s Office (ICO) | Enforces data protection law including rights around automated decision-making. Free guidance on your rights when AI affects you. Can investigate complaints. | ico.org.uk |
| Citizens Advice | Free advice on benefit decisions, discrimination and consumer rights including those involving automated systems. | 0800 144 8848. citizensadvice.org.uk |
| Full Fact | UK’s leading independent fact-checking organisation. Verifies claims made in news, politics and social media. Free to access. | fullfact.org |
| Age UK | Runs digital skills programmes for older people. Campaigns for digital inclusion through the Offline and Overlooked programme. | 0800 678 1602. ageuk.org.uk |
Sources used in this guide
Amnesty International, February 2026. UK: Government’s unchecked use of tech and AI systems leading to exclusion of people with disabilities and other marginalised groups. Primary source for findings on automated decision-making in the UK social security system, based on research with 782 people from October 2024 to January 2025.
Digital Poverty Alliance, 2024. Source for 19 million people remaining digitally excluded in the UK, and the finding that digitally excluded people are two to three times more likely to be unemployed.
DWP, Digital Skills and Preferences of DWP Customers, 2024. Source for statistics on digital access among Pension Credit, Attendance Allowance and Employment and Support Allowance customers.
Ofcom, Adults’ Media Literacy Tracker, January 2025. Source for digital exclusion statistics including the 1.52 million people completely offline and connectivity figures by geography.
Age UK, Facts and Figures about Digital Inclusion and Older People, July 2025. Source for digital exclusion statistics specific to older people.
HM Government, Digital Inclusion Action Plan: First Steps, February 2025. GOV.UK. Source for government policy commitments on digital inclusion and identified priority groups.
HM Government, AI regulation: a pro-innovation approach (White Paper), March 2023. Primary source for the five non-statutory AI principles and the sector-led regulatory approach.
House of Commons Library, AI regulation in the UK, 2026. Research briefing CBP-10003. Source for the June 2026 AI Compass finding that 85% of UK adults want strong laws, and for the legislative history and current status of UK AI regulation.
Data (Use and Access) Act 2025. Primary legislation. Source for changes to automated decision-making rules from 5 February 2026 and the ICO’s statutory Code of Practice duty.
Online Safety Act 2023. Primary legislation. Source for duties on regulated platforms including AI chatbots and content featuring deepfakes.
World Economic Forum, How cognitive manipulation and AI will shape disinformation in 2026, March 2026. Source for deepfake quality threshold, Ireland 2025 deepfake election incident, and AI-generated content in the 2024 to 2025 electoral cycle.
Full Fact, Full Fact Report 2026. Source for the description of disinformation, including AI-generated content, as a threat to UK democracy following the 2025 parliamentary inquiry.
European Broadcasting Union survey, 2025. Evaluated 3,000-plus AI responses assessed by professional journalists. Source for the finding that nearly half had at least one significant problem and a fifth contained major errors.
Ahmad et al., Bias in AI systems: integrating formal and socio-technical approaches, PMC, 2026. Source for the systematic review of how AI systems reproduce and amplify structural inequities.
Published by The Knowledge Loop Company | www.theknowledgeloop.com
Disclaimer:
The content published on this site is for general information purposes only. It does not constitute legal, financial, medical or professional advice of any kind. Nothing on this site should be relied upon as a substitute for advice from a qualified professional. If you need advice specific to your situation, please contact a solicitor, financial adviser, or other relevant professional. All free support organisations referenced in our guides are independent of The Knowledge Loop Company.