AI UK 2024

It was a fantastic opportunity to attend this year's AI UK conference at the Queen Elizabeth II conference centre.  There were so many discussions, exhibitions, and visions for the future of artificial intelligence. The conference underscored the gap between those with access to technology and those without, a divide that threatens to leave many voices unheard in AI training. There were some thought-provoking sessions and exhibitions.
These are the potential paths AI might take in the coming years.

Day 1

The New Real: Opening Provocations

 "Artistic and Queer Visions of AI Futures" Most refreshingly the session kicked off with a bang; we were delighted by speakers from the creative industries who have been making AI work for them.

• An AI drag queen trained by a real drag queen,
• A robotic arm and trained to paint by an artist.

Here is Me the Drag Queen with her AI double

The speakers challenged us to rethink our understanding of AI's mistakes—are they mere errors, or could they be considered new creations? This session pushed the boundaries of our conventional perspectives on AI's role in creativity and society.

Putting LLMs to Work

A session dedicated to "Putting LLMs to Work" posed critical questions about the utility and challenges of generative AI.  For example, it is now super easy to generate versions of scientific papers. All submitted papers are reviewed but what happens when the volume submitted becomes untenable?

People worry that AI replace them. Maybe, but the amount of legal challenges that AI is bound to spin up will keep lawyers busy; unless they start using large language models!

We laughed at the idea of AI replying to our emails and the recipient AI continuing the conversation back again; much like AI email ping pong!

Data, Labour, and AI

The conference also tackled the ethical considerations of AI development, particularly the exploitation of gig workers in training LLMs to filter harmful data. Mophat Okinyi spoke to us from Kenya giving us a brief insight into the harms that have been done to teleworkers working for companies training LLMs on social media data. 

Just imagine having to look at all of the most distressing and offensive content every day for hours upon hours!

This session illuminated the critical role of these workers, who possess an intimate understanding of the business far beyond that of many CEOs.

Exhibition Stands

There are already some amazing innovations being created.

This is CEMRG who are creating Cardiac Digital Twins

 

Diegetic GUI

This robotic arm is controlled by your eyes!

 

University of Exeter

Really interesting to see that AI understands images in a much different way than we do. For example, whereas we concentrate on the head of an animal, the AI judges the image quite differently.  In this example, the chicken in the middle picture has some very light pixellation around its head and the AI labelled it as a fox. In the picture on the right, the middle area of the man is pixellated so the AI was unable to recognise him.

 

Day 2

Safety First: AI Innovation and Deployment

The second day opened with a session on AI safety and innovation, reflecting on the lessons learned from the use of surveillance during Covid-19 and the importance of integrating human intuition with AI in complex systems. This discussion emphasized the necessity of designing AI systems with redundancy and human oversight to enhance both safety and trustworthiness.

Just like systems on an aircraft, AI should be one of many so that if something goes wrong, you can switch over to another.

The Secret Session: AI in Defence

• Command & control
• Surveillance
• logistics
• Cyber security
• Countering Misinformation - DSTL & DARPA to detect deep fakes

"The Secret Session" offered insights into the use of AI in defense, ranging from command and control to cyber security and countering misinformation. This session shed light on the cutting-edge research being conducted by DSTL (Defence Science and Technology Laboratory) and DARPA (Defense Advanced Research Projects Agency), underscoring the strategic importance of AI in national and global security contexts.

It's hard to deny that enhancing a fatigued military person tasked with surveillance with an AI assistant is going to greatly improve detection of enemy activities. It's also really reassuring to know how much work is going into making any AI developments safe. 

AI will be able to launch much more efficient cyber attacks that we will need to be super quick to respond. The question is how much autonomy can we allow our counter cyber AI to have?  It can be anything from just notifying us to taking action. Notification only may be far too slow and AI action could cause more damage than the original attack.

Handling Disinformation

Misinformation is a bigger problem than the disinformation itself. The original false post is not the biggest problem, it is the re-posts that cause the biggest threat. Some innovations include the use of cryptographic embedded information to show any changes to the original image algorithimically assessing the degree of manipulation.

Imagined Futures: A Visionary Dialogue

An imaginative session invited speakers to project themselves into the future, speaking from the vantage points of 2034, 2052, and even 2107, to discuss how the decisions we make today can shape a desirable future. This exercise in imaginative foresight urged participants to consider long-term impacts and ethical considerations in AI development.

Autonomous Cyber Defence: from Lab to Ops

Visualisation to express computer models makes them easier for us to understand.  Interestingly, artists are needed to scrutinise the output.  We also need local observers to counter disinformation.

What can Children Teach us about AI?

It was both interesting and reassuring to know that researchers have been working with 4 primary schools in Scotland to find out how children feel about the use of AI.  To ensure that AI remains safe, it is essential to include all voices, not just those from people working in the space.

The children were invited to be the interviewers, posing their questions and giving their opinions.  They were very concerned that AI should be fair, that all children should be offered the same advantages.  Most interestingly was their response to the proposal to train AI on video of children so that it could be used to predict seizures.  The children were very clear; they did not wan their data to be stored. This influenced the project and work to further anonymise the data is being undertaken.

Overall there was a big emphasis on making AI safe.