AI has moved quickly from something organisations were asking whether they should use to something employees are already experimenting with. But as adoption grows, many organisations are wondering how to use AI responsibly while understanding its impact on people, inclusion and decision-making.
This article explores key insights from our webinar, AI in the Workplace: How to Adopt It Responsibly and Reduce Hidden Bias, hosted by FAIRER Consulting’s Global DEI Consultant Alice MacDonald, with AI experts Claire Roberts, Co-Founder / AI Transformation Strategy & Training at Full Fathom Five UK, and Stu Dorman, Chief AI Officer at Sabio.
For many organisations, AI adoption is somewhere between experimentation and widespread individual use. People are trying different tools, managers are finding ways to use AI under pressure, and employees are experimenting with everything from drafting and summarising to analysis and decision support. At the same time, guidance is not always keeping pace with how quickly these tools are being adopted.
This situation creates an interesting tension. Organisations may have AI policies or guidance in place, but that does not necessarily mean people will use them at the point when they need to make a decision. If someone is working to a tight deadline, they are unlikely to stop and search for a policy document before putting information into an AI tool. Responsible AI therefore needs to become part of everyday behaviour, rather than something that exists only in a document.
It is also important to look beyond generative AI. AI is an umbrella term covering technologies that can recognise patterns, make predictions, understand language and generate content. Machine learning (e.g. fraud detection), predictive AI (e.g. sales forecasting), conversational AI (e.g. virtual assistants) and generative AI (e.g. ChatGPT) are all being used in different ways, meaning the risks and opportunities will vary depending on how the technology is applied.
AI adoption is not necessarily equal across the workforce. Lean In data suggests that men are 22% more likely to be regular AI users at work than women, while women were also more likely to worry that using AI could be perceived as cheating. These differences raise questions about who feels confident using AI, who feels able to experiment and what might be driving those differences.
It is important not to assume that an adoption gap is simply a skills issue. Confidence, workplace culture, psychological safety and previous experiences at work can all influence how comfortable someone feels using a new technology. As Claire Roberts discussed, people from different groups can experience different levels of risk in the workplace, and those experiences can influence how willing they are to experiment with AI.
For organisations, this means listening to people who are less confident or more hesitant about AI rather than assuming they are simply less interested in technology. Their concerns may highlight issues around governance, ethics, privacy or fairness that the organisation needs to understand. Those concerns may also reflect questions that customers and other stakeholders are asking.
This is where FAIRER’s Conscious Inclusion programme can support organisations. The programme focuses on the behaviours and assumptions that shape people’s experiences at work, helping individuals recognise bias, build curiosity and develop more inclusive ways of working.
AI did not invent workplace bias. Many AI systems learn from human-generated data, meaning patterns that already exist within organisations and wider society can find their way into the technology. If an organisation has historically promoted one type of person more quickly, for example, those patterns may be reflected in the data used by an AI system.
The risk is that AI can make those patterns more consistent and less visible. Bias can enter through the data being used, the way a system has been designed, the prompt someone gives it or the way people interpret its output. It can potentially affect recruitment, performance, promotion, learning, communication and customer-facing decisions.
A useful example discussed during the webinar was a manager putting six months of written feedback about an employee into an AI tool and asking it to summarise their strengths, risks and readiness for promotion. The output may look clear and objective, but it could simply be repackaging biased feedback, missing important context or turning uncertainty into a neat story.
There is also the risk of automation bias. Because AI produces fluent, structured answers, people may give its output more weight than it deserves. AI can support a decision, but it does not take responsibility for that decision. The person using it still needs to apply judgement, question the output and understand what sits behind it.
Before introducing AI into a process, organisations need to be clear about what they are actually asking the technology to do. Is AI the right tool for the problem? Who is affected by the decision? Whose voice is being included? Who remains accountable for the outcome?
It is also important to consider the quality and fairness of the inputs and outputs. Organisations should ask whose history or data is represented, what patterns the system may be prioritising and how people will interpret and act on the results. Testing AI outputs for quality, fairness and impact should be part of the process rather than something considered after a problem occurs.
FAIRER has seen the value of taking this kind of systemic approach through our client work. For a global manufacturing organisation, we supported the development of a global DE&I roadmap through a review of policies and processes covering areas including talent acquisition, talent development, performance management and work-life integration. Data mapping was also used to establish a diversity baseline and examine areas including recruitment, retention, performance, reward and training.
The same principle applies when organisations introduce AI: understand the system and the people affected by it before scaling it. If existing processes contain barriers or bias, automating them does not make those problems disappear.
The conversation around AI and bias is not entirely negative. AI can also provide opportunities to challenge some of the biases that humans bring to decision-making.
During the webinar, Stu Dorman discussed examples where AI has approached problems differently from humans because it has not been constrained by the same assumptions or established ways of thinking. AI can potentially help organisations identify patterns in areas such as customer service quality, where human reviewers may bring their own conscious or unconscious biases to assessments. Used carefully, technology can provide another perspective and help organisations identify inconsistencies that might otherwise be missed.
The important point is that AI should not automatically be treated as either biased or objective. Its impact depends on how it is designed, what data it uses, what problem it is being asked to solve and how people use its outputs.
One of the strongest themes from the discussion was that sending employees on an AI training course is unlikely to be enough. People need the skills to use AI, but they also need the time and space to rethink how their work could be done differently.
As Claire explained during the webinar, organisations can give people a new set of skills without giving them the capacity to apply those skills. Someone might complete Copilot training, for example, but six months later simply be writing emails a little faster. The bigger opportunity is to rethink how work is actually designed, and that requires leadership. Employees need to know that experimentation is encouraged, while also understanding where boundaries exist. Leaders also need to create an environment where people can raise concerns about AI without being dismissed as resistant to technology.
FAIRER has seen the importance of this combination of leadership, behaviour and organisational systems through our work with clients. For a major transport business, we delivered Conscious Inclusion workshops to managers and technicians, creating opportunities for employees at different levels to share experiences, build trust and reflect on everyday behaviours.
If organisations want responsible AI use to become normal behaviour, guidance needs to be available when people are actually using the technology. Relying on employees to remember a policy they read months earlier is unlikely to be enough.
This is where behavioural nudges can play an important role. Instead of simply telling people what they should do, organisations can build prompts, checks and reminders into the workflow itself. Responsible AI then becomes part of the way work is done rather than another policy sitting separately from everyday activity.
Training also needs to reflect people's different roles. A recruiter using AI to support shortlisting faces very different risks from someone using it to summarise meeting notes or draft marketing copy. Generic AI literacy is useful, but role-specific practice can help people understand what responsible use looks like in their own work.
There are also two sides to AI and learning: learning for AI and AI for learning. People need to understand the tools, risks, prompts and accountability, but AI can also provide opportunities to practise workplace scenarios, challenge thinking and explore different approaches.
1. AI adoption is already happening. Organisations need to understand how employees are using AI now, rather than waiting for adoption to happen through formal strategies alone.
2. AI can amplify existing bias. Technology can reproduce patterns that already exist in workplace data, processes and decision-making.
3. Adoption is also an inclusion issue. Differences in confidence, risk and psychological safety can influence who uses AI and who does not, risking creating a two-tiered workforce.
4. Responsible AI requires human judgement. AI can support decisions, but accountability remains with the people using it.
5. Don't automate broken processes. Before scaling a process with AI, understand whether the existing process is fair and effective.
6. Training alone isn't enough. People need the time and space to rethink how their work could change.
7. Measure more than efficiency. AI should be assessed by its impact on people, inclusion and customers as well as productivity and cost.
Explore how bias, behaviour and decision-making affect your organisation. FAIRER's Conscious Inclusion programme helps organisations move beyond awareness towards practical behavioural change, embedding inclusion into everyday interactions and decisions. Alternatively, watch the webinar recording to hear the full discussion with Alice MacDonald, Claire Roberts and Stu Dorman.