We asked three leading AI experts at the University of Cambridge to share the biggest risks and essential considerations for leaders who want to drive AI transformation.

The biggest risk with AI adoption isn‘t moving too slowly. It’s moving too fast – investing in costly tools without clarity on organisational needs, and the structures needed to use them responsibly and effectively.

Professionals are quickly learning how it can support them in their roles. But without a clearly define strategy, this can lead to organisational risks.

So how do you lead AI transformation in a way that’s scalable, accountable, and aligned with business value?

University of Cambridge AI and risk experts Dr Alexandru Marcoci, Ray Eitel-Porter and Professor Magda Osman joined a live panel(Opens in a new window) to discuss the key factors that drive successful, long-lasting and risk-aware AI adoption.

1. People will use AI anyway, so you need to plan a way for them to use it safely

Dr Alexandru Marcoci, Assistant Professor of Global Risk and Resilience, Centre for the Study of Existential Risk, University of Cambridge:

If you don't start thinking about the way in which people in your organisations can work with AI, they will use it without telling you.

The idea that we can completely ban AI or not work with AI, isn’t realistic, nor would many organisations want to. So, we need to find ways in which we give access to this technology in a safe way, that respects the principles of organisation, but also the legal environment in which you're operating.

I've seen organisations limit usage or ban AI for all kinds of good reasons, including information security, but employees will still find a way to use it somehow. And one of the reasons why they do it is because they know, and they're right, that it's almost impossible to tell when text was generated by AI.

We all think we have tells. For me, it's the em-dash and the rule of three. If I see that, I have the strong suspicion that the text was written with ChatGPT. Others use different signs, but you can't really be sure. I think that's a very fundamental risk. We need to find ways of giving access to this technology to people, because otherwise they'll use it unsafely.

2. Generic training isn’t enough – you need training that speaks to an employee’s individual needs

Ray Eitel-Porter, Senior Research Associate for Responsible AI, Intellectual Forum, Jesus College, University of Cambridge:

Often, I see organisations rolling out a tool, let's say Copilot or its equivalent, and assuming that half an hour of online training is going to be enough. Everybody's going to adopt it, and it's going to do amazing things for the organisation.

The reality is that even though most of us have played around with AI at home and have used it for private things, once we move into the workplace, people are a lot more cautious. You're there to do a job, you're supposed to deliver accurate outputs and results. You're under a lot of time pressure, so finding the space and the imagination to start thinking about how you could use AI to rethink what you do, is tough.

Plus, you need to consider, ‘what do I need to be careful of, what might go wrong, and what's allowed?’ That typically takes a lot more handholding than just the half-hour training.

You clearly can't sit down with everyone, but you've got to seed that deeper understanding, you need super users dotted around the organisation who can then help and encourage colleagues.

3. Before investing in expensive solutions, you need to fully understand the problem

Professor Magda Osman, Research Fellow, Judge Business School, University of Cambridge:

You can't solve a problem successfully if you don't disentangle it and design a problem statement, so you understand the full breadth of the problem and how AI can contribute to the solution.

What can happen is that organisations get excited about AI, then retrofit, or even create problems for AI to solve that are completely irrelevant to the actual problems that the organisation is trying to solve.

The other thing is, the more complex the problems get, the more people are inclined to think about more sophisticated technological solutions. They're profound problems that might need both technological and non-technological solutions.

The critical point is how you bring both of those different approaches together to address the problem successfully, rather than just cosmetically.

4. The best person to lead AI transformation is the person who’s motivated to do it

Ray Eitel-Porter, Senior Research Associate for Responsible AI, Intellectual Forum, Jesus College, University of Cambridge:

Like any other major change programme in an organisation, the most important thing is having really committed leadership.

I often get asked the question, who should lead AI governance? My view is that it actually needs to be somebody who is really interested and wants to do it. Somebody who has budget and enough political clout to encourage cooperation across the organisation. As when it comes to responsible AI and AI governance, you can't do it just in the technical, legal or risk function. You need all these areas and others to work together.

So it's having a leader who has enough of an understanding. They don't need to be computer programmers, but they need to be able to ask the right questions and direct teams. That's the key.

5. AI isn’t going to replace human workforces, and training junior employees is essential

Dr Alexandru Marcoci, Assistant Professor of Global Risk and Resilience, Centre for the Study of Existential Risk, University of Cambridge:

There is a concern that if you can use AI too effectively, then you might be replaced. That goes back to a very interesting situation we currently have with AI.

Talking to senior people in organisations, mostly in government, but also in the private sector, I often get one of two reactions to any question about AI. There's a group that thinks that AI still can't do anything and that it's far worse than it actually is. When you show them an example of what AI can do, they're always shocked. And then, on the flip side, you have others who think that AI can do everything.

There’s a study from Stanford University that looks at AI deployment in call centres(Opens in a new window). While it shows a statistically significant improvement in productivity, this still only amounts to about 14%. It's not going to revolutionise anything or lead to laying off the entire workforce, but it can improve effectiveness and accuracy.

One of the problems is that in lots of experiments, when we try to compare AI to humans, and when the AI systems are doing a better job than humans, it's usually that they're doing a better job than junior people working in an organisation. It's very rare that you find studies where AI is outperforming people with high-level expertise.

So as long as you still need humans to control AI and design processes using AI systems, but you eliminate the junior workforce, what will you do 10 years from now? Where will the senior people who know how to use AI be coming from if they haven't been trained throughout that time?

6. Over-reliance on AI could lead to a loss of skills, if this isn’t challenged

Ray Eitel-Porter, Senior Research Associate for Responsible AI, Intellectual Forum, Jesus College, University of Cambridge:

There are a lot of risks attached to using AI. One is what we call automation bias, the tendency that people have to over-trust AI answers.

The other risk that goes hand in hand with automation bias is cognitive atrophy – the risk that we're going to outsource too many of our functions to AI, and we will then gradually forget how to do them.

So you can easily see how within workplaces, when people start to rely on AI to do certain things, after a while, either the individual or the organisation loses the capability to do those things independentely, which might be quite dangerous.

7. What does good AI leadership look like? Listening to your workforce

Professor Magda Osman, Research Fellow, Judge Business School, University of Cambridge:

Leaders need to critically examine the motivation and incentives for adopting AI, and the grand claims that are made which are advertising it as a panacea.

If not, it may present problems for the workforce, who might feel that they can't honestly communicate the day-to-day problems that they're dealing with, and the frustrations that AI is introducing.

If there's no forum for people to share these frustrations, it leads to a disconnect between what leadership wants the organisation to do and how the actual organisation is solving problems. Or creating workarounds, if a technology is creating more problems than it's solving.

Organisations are combinations of individuals and teams that are working to solve a collective problem. You need to think, what's the goal that the organisation is trying to solve? It seems obvious, but it's often forgotten.

Watch the full panel discussion (Opens in a new window)

Whether you are an individual looking to lead with greater clarity, or an organisation building AI capability at scale, we can help.

Explore more insights and practical, results-driven courses and programmes(Opens in a new window) from renowned University of Cambridge experts.