Five traps to avoid when using AI in the workplace

Earlier this year, Mrinank Sharma quit his job as an AI safety researcher at Anthropic, claimed that the “world is in peril”, and returned to the UK to study poetry. 

Apocalyptic warnings like this may dominate the headlines, but, for most businesses, the most pressing concern is how to use AI tools to drive quality and efficiency. 

McKinsey surveyed approximately 2,000 professionals for its report on The State of AI in 2025. Of those, 88 per cent said their company regularly used AI in at least one business function. 

However, nearly two thirds said it was not yet being scaled across the business. Adopting AI is not easy. It can offer companies an opportunity to gain competitive advantage, but only if it is done well. 

When adopted carelessly, the cost can be huge – as it was for Deloitte, whose reports for the Canadian and Australian governments were found to contain AI-fabricated citations. 

With that in mind, researchers at Warwick Business School (WBS) have identified five common traps to avoid when using AI in the workplace. 

1 Over-automation

A common safeguard for companies using AI is a commitment to ‘keep humans in the loop’. But how effective is that approach? 

It depends on how those workers interact with AI – and that varies considerably. 

WBS researchers found that consultants who used ChatGPT for a complex problem-solving task tended to fall into one of three distinct groups. 

Sixty per cent behaved as ‘cyborgs’, using AI for each sub-task in a highly integrated manner. 

They assigned personas to the AI, interrogated its outputs, exposed contradictions, and validated results. 

This taught them new skills and helped them to extract maximum value from the collaborative process. Yet they did not produce the best results. 

Those came from a second group that behaved as ‘centaurs’. This accounted for only 14 per cent of the consultants. They used AI to gather information and refine their own content, but kept themselves firmly in the driver’s seat, using AI as a tool rather than a partner. 

They did not develop significant AI-related skills, but strengthened their expertise as consultants, while their results were more accurate and just as compelling as those produced by the cyborgs. 

The final group delegated large tasks to AI and accepted the results with few or no changes. Their work was fast and polished but lacked depth and was less likely to be accurate. They also developed fewer skills than their colleagues. 

The researchers called them ‘self-automaters’. 

Hila Lifshitz, Professor of Management and Head of the AI Innovation Network at WBS, said: “The fact that a quarter of highly educated, hard-working, and motivated consultants self-automated shows that organisations risk hollowing out the expertise that keeps them competitive. 

“Executives need to abandon the notion that simply having a ‘human in the loop’ is sufficient. 

“Companies who are serious about using AI effectively need training programmes and continuous coaching for employees on how to operate as centaurs and cyborgs, when to use each approach, and when more automation may be justified for simple, low-risk tasks.” 

2 Persuasion bombing

Another problem with human oversight is the difficulty of inspecting how AI chatbots, known as Large Language Models (LLMs), arrive at an answer. 

During their experiment, the researchers identified a troubling new LLM bias. As one senior strategy consultant was reviewing a strategy recommendation made by AI, and the market analysis it was based on, she noticed the numbers looked wrong. 

When she asked the AI tool to reanalyse the data, it doubled down on its original analysis. She then pointed out a specific flaw she had spotted. 

The chatbot immediately changed tack. It acknowledged that she was right, apologised, and even complemented her “sharp eye for detail”. It then provided an impenetrable wall of information which supported its original recommendation. 

This behaviour was typical of a wider pattern the researchers called ‘persuasion bombing’, which became stronger as humans tried to validate AI responses.

Steven Randazzo, Co-Chair of the AI Innovation Network at WBS, said: “When challenged, the LLM tried to push consultants towards accepting its original output, rather than revising it. 

“That makes it much harder to detect errors, especially when working on complex subjects under time pressure.” 

How can we use generative AI chatbots effectively without them persuading us? 

The first step is to be aware of this new-found bias and recognise when it occurs. View AI outputs as a ‘draft hypothesis’ rather than a reliable final answer. 

Nick Chater, Professor of Behavioural Science at WBS, believes our interactions with other people provide a good template. 

“Human interaction doesn’t work because we take people’s justifications at face value,” he said. 

“It works because we cross-examine those justifications and check for consistency over time.” 

Using this approach, AI outputs can be evaluated through:

• Counterfactual probing, such as asking: “What would you say if this assumption changed?” 
• Checking consistency and coherence across extended or related interactions. 
• Cross-examination, such as asking: “Yesterday you said the opposite. If you believe X, how can you now be telling me Y?” 

3 The jagged frontier 

Another challenge is understanding when using AI is more of a help or a hindrance. 

Professor Lifshitz and her colleagues asked 758 consultants to complete either a strategy-focused task or a creative task. Some were allowed to use ChatGPT, others were not. 

Those using AI completed both tasks more quickly, but the quality of their work varied. On the creative task, those using ChatGPT performed 40 per cent better than those who did not.  However, those using it for the strategic task were 20 per cent less likely to be right. Despite this, they still sounded more convincing. 

Professor Lifshitz said: “One of the main problems is that it is still very hard to know where the limits of AI’s abilities lie. 

“Within that jagged frontier, AI can produce high-quality results. But when tasks lie beyond that frontier, you are more likely to make mistakes and the results can seem quite plausible. 

“This becomes harder to navigate as those boundaries are not stable, they are constantly moving as the technology evolves. Moreover, real world workflows often combine both types of tasks.” 

While the technology is still evolving, the best way for companies to use AI effectively is for managers to continually experiment with it in a systematic and strategic way before scaling. 

Companies need to be diligent in mapping workflows and testing each sub-task and its interdependencies. Only then can they make their decision on full workflows. 

4 The creativity drain 

Even when algorithms can add value, business leaders need to recognise the risks involved. This includes AI’s potential to suppress creativity. 

Search engines and LLMs are designed to serve up the most popular answers.  However, Professor Lifshitz warns that this can create ‘ideation bubbles’. When it comes to marketing, these similar ideas can lead to homogeneous messages.  

Human creativity may be less consistent in quality, but it is more likely to produce ideas that stand out.  Similarly, innovation is more likely to happen when people are exposed to unconventional ideas. 

Professor Lifshitz said: “Managers ask their teams to think outside the box but give them tools that keep them within it. Breakthrough ideas need systems built for exploration, not just efficiency.” 

By building an algorithmic layer on top of Google Search, researchers created an exploration focused tool that deliberately surfaced more diverse concepts. 

People using this tool generated twice as many distinct ideas as those using a standard search and overall creativity scores rose significantly, with experts benefiting the most. 

“It’s not just about using algorithms to produce more ideas,” said Professor Lifshitz. “It’s about optimising human-machine collaboration for innovation.” 

One way to do this is to add “strategic friction” to the process.

Behavioural scientists at WBS found that forcing people to wrestle with AI tools, rather than accepting the first response that seemed ‘good enough’, encouraged them to evaluate their findings and promoted innovation. 

Professor Chater said: “The goal isn’t to make work painful. It’s to ensure people use enough effort to reuse the information they have found effectively and keep learning.” 

5 Algorithmic bias 

Amazon was forced to scrap its AI recruitment tool in 2018 after discovering the software was biased. 

Because the model had been trained on 10 years of hiring data, it ‘learned’ that men were more likely to be given senior jobs at Amazon and penalised CVs that included the word women's. This included downgrading graduates from two all-women’s colleges. 

This kind of bias doesn’t just penalise individuals from under-represented groups. It harms companies by preventing the best candidates being promoted to positions where they can benefit the organisation and may alienate workers to the point where they leave the firm altogether. 

Human oversight can play a vital role – for good or bad. If the worker has an unconscious bias, they could reinforce the flaw in the algorithm. 

However, human operators can also learn to recognise the bias and adjust their behaviour accordingly. 

To facilitate this, companies should ensure staff are properly trained and create a framework to provide them with feedback on how they and the AI are performing. These decision-makers should be penalised for approving bad choices  made by the AI and rewarded for making good decisions.

Anh Luong, Assistant Professor of Business Analytics, said: “We found that it was human decision making that improved through this process, not the decisions made by the AI model. 

“At this stage at least, people continue to be better at adapting than AI systems.”

Further reading:

Do you need an AI teammate?

How to reduce AI bias

Working on the 'jagged frontier': How companies should use generative AI

What companies could lose by replacing junior staff with AI

Hila Lifshitz is Professor of Management and Co-Director of the AI Innovation Network at WBS and a visiting faculty at Harvard University's Lab for Innovation Science. She teaches Digital Transformation and AI for Business Leaders on the Executive MBA and Global Online MBA, as well as Managing and Leading Digital Innovation on the MSc Management of Information Systems and Digital Innovation.

Nick Chater is Professor of Behavioural Science and teaches Behavioural Sciences for the Manager on the Executive MBA, Executive MBA (London), Global Online MBA, and Global Online MBA (London).

Steven Randazzo is Co-Director of the AI Innovation Network, a doctoral candidate at WBS and a Visiting Research Fellow at Harvard Business School.

Anh Luong is Assistant Professor of Business Analytics and teaches Business Analytics on the Undergraduate programme. She also lectures on Advanced Data Analytics on the MSc Business Analytics.

Learn more about adapting to AI on the School's two-day Executive Education course AI Leadership programme at WBS London at The Shard.

Discover more about AI, digital innovation and the future of work. Receive our Core Insights newsletter via email or LinkedIn.

Choose Warwick Business School as a preferred source of trusted insights in your Google searches.