AI UK 2025

View from QEII conference centre, Westminster.

A few weeks ago I attended AI UK 2025 – a showcase of the latest applications of AI.

There were a range of talks, many focusing on the bigger picture of AI ranging from it’s impact on the future of research in the UK and how AI is set to influence our every day lives.

I couldn’t attend all of the talks I wanted to – it was a huge event! Plus many of the workshops overlapped or were run at the same time.

In this post i’ll summarise a few of the key themes that emerged from the event.

I must admit, the conference has got me feeling somewhat philosophical about the future of AI and the roles we need to play in it’s evolution. I say we – because it’s apparent that we all need to be involved in inevitable adoption of AI into every aspect of our lives.

AI is a broad term, encompassing multiple methods – as with most new or trending technologies, once you get past the initial knee jerk reactions and voices of doom, there is something actually quite interesting going on under the hood.

Bioinformaticians will already be aware of some of the concepts used in AI applications. Techniques, such as Hidden Markov Models (HMMs) and clustering algorithms, were developed long before the term ‘artificial intelligence’ gained popularity.

As AI has evolved to encompass a wide range of data-driven methods these approaches have become closely associated with AI applications, most notably in speech recognition and natural language processing.

Moving away from a more technical discussion – the adoption of AI tools in research and public services brings with it an entirely new set of challenges for society as a whole.

Interestingly these challenges hinge on how humans are going to develop, modify – and perhaps most importantly – interact with AI technologies.

Foundation models as public assets

In a number of talks emphasis was made on a need to invest in foundation models as public assets. The general consensus was that funding for AI technologies in the UK should focus on developing AI technologies from the ground up.

Foundation models are general purpose models, such as LLM’s like GPT-4, that are applicable to many use cases.

The argument follows that general models, which can be later fine tuned for specific purposes, are more valuable to us than highly specific AI models. This seems like a ‘no brainer’ but there is another, less obvious, point attached to this idea.

Large, pre-trained AI models (like GPT-4, BERT, or DALL·E) have been trained using publically available data and as such should be treated as publicly accessible resources that can benefit all, rather than commercial products.

Amidst the AI hype there is demand for more niche, highly specific AI tools that can be commercialised.

Thank fully, there is already a thriving open-source community for AI applications, where thousands of models are made freely available on Hugging Face.

Maintaining Human connection

Another key theme emerging from the meeting is to ensure humans are involved in the ultimate decision making process when relying on AI to perform automated tasks.

It was encouraging to see that the general consensus solution was to enhance human connection and interaction with AI workflows, rather than separating ourselves from them.

Another hope is that AI tools can reduce the cognitive overload we have created for ourselves through technological advancements and ever increasing admin, ultimately empowering us to spend more time communicating with each other.

A nice example of this was the idea of an ‘ambient conversation tool‘ which could have a considerable impact on healthcare, empowering healthcare professionals to spend less time typing up patient notes and more time communicating to their patients.

It’s clear that there is little desire for AI to replace human creativity and ingenuity, perhaps it’s time to shift our priorities to developing AI tools with a more far reaching impact that assists us in our day to day lives.

Of course, these ideas are far less ‘trendy’ – but their potential for improving our quality of life is huge.

AI and the future of research in the UK

AI applications are already having an influence on the research community, for better and for worse. AI tools such as Chat-GPT are now routinely being used to help write manuscripts, generate code and workflows.

If we set aside the panic of what this represents – continual reliance on tools such as Chat-GPT will no doubt lead to change in the way we think and tackle problems – it does make me wonder if such tools are ultimately a power for good.

Rapid technological advancements and a pressure to publish ever more manuscripts make it increasingly difficult to keep up to date with the latest knowledge. Certainly GPT based models can assist us with this task, with their accuracy ever improving.

As with the ‘ambient conversation tool’ applications that can reduce the cognitive burden on research departments could be, in my opinion, more valuable that the adoption of AI methods in scientific works themselves.

Ultimately, the power of AI might lie in it’s ability not to replace us but to share the load, to unburden us and allow us precious time to get creative.

For research and problem solving in general, this could be invaluable.