UST: A Guide to Building Responsible AI Systems

AI, digital transformation and economic resilience continue to dominate the global technology agenda.

At the centre of this conversation is Heather Dawe, Chief Data Scientist UK and Head of Responsible AI at UST, who is focused on how technology and responsible AI can create sustainable value for both organisations and wider society.

UST is an AI and digital transformation company employing more than 30,000 people worldwide. Its global teams partner with many of the world’s largest enterprises to design, deploy and scale AI and wider digital transformation initiatives.

With more than 25 years’ experience across government, the public sector and industry, Heather is a recognised authority on applied data science and AI ethics. A regular contributor to national and international media, she champions human-centred innovation that delivers progress while maintaining accountability.

Here, Heather explores UST’s approach to ethical AI, the impact of emerging technologies and how organisations can prepare for an increasingly AI-driven future.

How can organisations build AI systems that are both innovative and responsible?

It’s completely possible for AI systems to be both innovative and responsible. 

Key to this is to be aware of and recognise where AI systems can fail – get jail-broken, hallucinate and exhibit bias, for example, and to ensure the guardrails that will control and protect against these failures are built into these systems. 

We find that a fundamental part of responsible AI innovation is ensuring responsible AI practices are baked in at the earliest opportunity, not left to be an afterthought just prior to the planned go-live. 

This is one of the main reasons so many AI projects fail – and it is entirely avoidable.

What practical steps can leaders take to embed fairness and transparency in AI governance?

The reasons why AI can exhibit bias – such as AI unfairly favouring males over females in automated resume review– as well as hallucinate and have security weaknesses that get very technical quite quickly. 

I don’t think leaders necessarily need to understand the technical detail, but I do think they need to have an awareness of the impact of unfair and unethical AI systems and what they can do to guide their business to avoid such events.

Transparency in AI governance plays a role here, as does training, consistency of approach and recognising that this is a fast-moving space where new AI risks are frequently emerging. 

How do you see collaboration between government, industry and academia shaping responsible AI?

I think this tripartite collaboration is important but ,if I’m honest, right now I do not see it working as effectively as it should on a global stage.

The recent ability granted to Grok users by xAI to create pornography based upon images – the majority of which have been images of women and children – supplied to Grok by the user is an appallingly relevant example. 

This is something many people had seen coming and I think governments need to be more pre-emptive in their approaches to controlling such AI services rather than reactive. 

I also believe the majority of industry and academia would welcome the opportunity to contribute to greater collaborative efforts here. 

What role do skills and training play in ensuring ethical AI adoption across organisations?

Skills and training are very important. For example, I’ve been aware of the ways in which data bias can lead to unfair and unethical AI for years. 

While this awareness is increasingly spreading across industry, I do not think enough of us yet understand the potential implications on business and wider society of such data bias and so I’m not able to confidently say that AI will always be fair and ethical.

In the coming years, we will see AI governance functions growing across industry. Aong with assistive technologies, these skilled and trained professionals will form a key part of ensuring AI used across industry is consistently responsible.

What global principles or frameworks do you believe are most effective in guiding responsible AI development?

I think some of the highest profile and most widely used frameworks – EU AI Act, NIST AI Risk Management Framework and ISO/IEC 42001 – are effective and clear guides. 

One of the gaps we see at the moment, however, are technologies driven by these frameworks that ensure AI risk management using one or more of these frameworks is straightforward and accessible.

We are actively filling this gap with our clients through the development of data-driven solutions that assist them in this space. 

How do you see responsible AI evolving as generative and autonomous systems become more deeply integrated into global industries?

I’m writing a book about this with my colleague, Adnan Masood, at the moment.

As AI continues to evolve at a fast pace and agentic AI leads to increasingly autonomous systems, along with this comes the new and changing risks of these systems making mistakes and unfair decisions. 

The potential for this to happen at scale and with significant adverse impact with agentic AI is considerable. 

However, the opportunity to gain business benefit with agentic AI is also considerable. 

Key to achieving the latter while avoiding the former are robust evaluation methods and wider guardrails – including, of course, humans-in-the-loop – which ensure agentic AI remains in control at all times or is automatically taken out of service.

This is an evolving space that has huge potential and it’s an exciting area to be working in right now.