Main image: Cohorts 1-5 AI Challenge Day 2024
The annual ART-AI ‘AI Challenge Day’ was held on Monday 29th January 2024 at the APEX City of Bath Hotel. This is an all day event where we and external partners get together to have roundtable discussions on specific issues of interest to each partner.
We are extremely grateful to the following companies and organisations who spent the day with us presenting an ‘AI Challenge’ and hosting table discussions: ATLAS ELEKTRONIK UK, BLAST Fest, BMT, CMS, Deloitte, Domin, Global Digital Foundation, GXO, Method Grid, Tobacco Control Research Group and UK Safer Internet Centre.
At the end of the day, each table had a student representative present their feedback on the discussion to all attendees. This presentation highlighted the main challenge(s) of the company, along with possible solutions, next steps and future outcomes.
We are already thinking about the 2025 AI Challenge Day, so do get in touch if you have a challenge to suggest!
‘Table Hosts’ and AI Challenges
How can AI be used to enable autonomous platforms to comply with the principles of Mission Command?
The practice in the UK’s Armed Forces of devolving responsibility down to low levels of command is known as mission command. The commander’s intent is shared with subordinates, who are told what to achieve and why, but are then left to decide how to achieve it. Subordinates are encouraged to use their judgement, initiative and intelligence in pursuit of the commander’s goal.
With the increased use in defence of robotic and autonomous systems to take human operators out of harm’s way, we need those systems to be able to respond in accordance with commander’s intent, especially when the tactical situation requires platforms to maintain radio silence. How can AI best be utilised on autonomous platforms to use judgement, initiative and intelligence to support the commander’s goal and comply with the principles of mission command?
Community-Driven Development in STEM and Creative Industries Through Data Use & AI-Enhanced Digital Platforms
This session will focus on leveraging AI and technology to collect insights from participants before, during and after live events in an unobtrusive manner to efficiently guide future public engagement.
Morning Session: Understanding BLAST’s Data Needs & Brainstorming an AI-Enhanced Platform
In the first half of the session, we will discuss how to develop a strategy to gather and use data to help BLAST achieve its aims. This will involve exploring which data is most important, where this data can be gathered from and how it could be used.
In the second half of the session, we will brainstorm AI solutions that streamline digital interactions, audience insight gathering, and encourage a dialogical approach to information exchange. The aim is to conceptualize an AI-powered digital platform that not only gathers and analyses data but also fosters deeper community engagement.
Afternoon Session: Further Brainstorming & Building
In the afternoon, we will continue to brainstorm these solutions and begin to build out a platform using open source tools. One possibility for this could be a platform that records audience questions and feedback, and then analyses the text for themes and sentiment.
Key Questions:
1. How can AI facilitate better documentation, analysis, and sharing of community insights in a creative and engaging manner?
2. What features should an AI-enhanced digital platform have to support live feedback and interactive dialogue before, during and after community events?
The use of explainable AI in the engineering design process to help identify design options
Utilising machine learning techniques in the engineering design process can be a great way to condense and embed previous examples and historic data. However, in engineering design where everything is about compromise rather than just providing a solution, the trade off and optimisation space needs to be described. How can explainable AI techniques be used to provide engineering designers with a virtual “co-pilot”? How can the solution trade off space be described?
Morning: Frontier AI safety
What technical means could be developed to prevent AI algorithms from self-replicating?
What technical means could be developed to prevent kill switches from being bypassed or deactivated by AI algorithms?
Afternoon: Hallucinations
What technical means could be developed to enable hallucinations to be identified in outputs of generative AI algorithms?
How do hallucinations manifest themselves in different modalities?
Creative industries
How will AI affect the creative industries (film, games and TV) in the short, middle and long-terms? Can we predict future trends, threats, and opportunities? How can the creative sector adapt to developments in generative AI?
At the closing of COP28 governments around the world have agreed, for the first time ever, to “transition away” from fossil fuels to avert the worst effects of climate change. Achieving transition presents a wicked challenge. Policy makers will need to balance multiple competing and conflicting factors to identify interventions to support change. They will need to consider and balance societal ambition, industrial ambition, and Net Zero/SDG goals, whilst also considering the wider national, and global, impact of the choices made.
Policies to achieve fair transition will need to be accepted by communities and able to endure through successive governments. Investment decisions will need to be made that ensure each ambition can flourish separately and collectively.
Imagine if we had an AI model that could be used to predict the interactions between competing placemaking priorities and identify the most socially acceptable outcome. What might be the challenges? Through whose lens do we optimise ‘socially acceptable outcomes’? What are the biases inherent in the decision making and are these better or worse than humans?
How the world currently uses energy is inefficient, expensive, wasteful, and has a high carbon output. Meaningful change to reduce the amount of energy that we consume is needed urgently. Achieving more with less is how we make this possible. The solution is digital.
Combining, optimising, and integrating diverse and challenging disciplines such as metal additive manufacturing, advanced electronics, digital control techniques and connected technologies, Domin are developing groundbreaking motion control technologies (including servo valves, active suspension, braking systems, flight actuators) that are revolutionising the industrials, aerospace and automotive sectors.
Together, we will save one gigatonnes of CO2 every year by 2030.
Topics where AI can benefit Domin include the following:
- Using large language models to build a valve concession bot. An automated human-in-the-loop concession bot will improve productivity during the manufacturing and test process.
- Using generative AI (not necessarily language models) to help with CAD design and model simulation. This will enable faster design iteration and product validation, and help us deliver solutions to our customer sooner.
Global Digital Foundation is acting on the growing consensus across industry that effective tools and services are now needed to operationalise assurance and build trust and confidence in AI. To demonstrate regulatory compliance, build public trust, and ultimately unleash the potential of AI, suppliers and users understand the need to operationalise regulations and standards through AI assurance. Launched in May 2022, the AI Assurance Club is a Global Digital Foundation initiative. The Club brings together representatives from across the AI ecosystem. We do this by convening events and working groups with leading industry experts, AI assurance providers, regulators, policy makers, standards and certification bodies, and academic experts. We share key insights into the rapidly evolving AI assurance ecosystem, and offer unique opportunities for members to shape industry’s response to these developments. Our growing membership currently numbers around 250 professionals working in AI technology, international standards development, public policy, and universities.
Of particular current interest is the challenge of transparency through the value chain; essential for risk assessment and mitigation, and importantly to enable information to be compiled to demonstrate regulatory compliance and standards conformity. This challenge has many facets including: the wide variety of value chain configurations, the power dynamics between large and small actors, the value chain ‘visibility horizon’, the ‘technology exceptionalism’ of ‘big tech’ foundation model providers, and the current trend to push down ambiguities in implementation for resolution by standards bodies – who have yet to complete their work. At a more detailed level, issues also arise due to the variety of AI technologies. For example, continually updating AI systems make it harder to envisage processes for assurance information sharing through the value chain.
Our roundtable discussions will share our current analysis, and consider some facets of the AI assurance challenge in more detail, with a view to refining and producing new ideas that can be incorporated into future AI policy and standards contributions.
In our role as a supply chain organisation, we are responsible for the storage and management of inventory for a multitude of clients, each facing unique seasonal demand fluctuations. These variances present a complex challenge in optimising warehouse space, manpower, and resources. We believe there is an opportunity to leverage historic data, such as past inventory levels specific to each client, regional weather conditions, and economic indicators, to build an AI-driven predictive model that can more accurately forecast these seasonal demands.
Addressing this challenge is not just about improving our internal efficiencies; it’s about providing increased value to our clients. By integrating and analysing multiple data streams, we aim to optimise inventory storage, reduce costs, and minimise stockouts or overstocking for our clients. The end goal is a more agile, responsive supply chain that can adapt to seasonal changes in demand, thereby improving customer satisfaction and our competitive edge. We are seeking innovative solutions that can help us turn this vision into a reality.
Lessons Learned Discipline in Major Project Delivery
Lessons Learned typically refers to the structured process of identifying, analysing, and documenting key insights gained through a project’s implementation and execution. The goal is to reflect on what went well, what could be improved and to capture effective practices that should be sustained or shared across the organisation.
For major capital projects, an effective lessons learned practice is critical given the complexities, budgets, and long timelines involved. A recommended approach involves conducting periodic lessons learned sessions throughout the project lifecycle to capture real-time feedback.
Challenge:
Key issues with current lessons learned practice:
- Lesson Learned practice is not integrated into day-to-day P3M (Portfolio, Programme and Project Management). Lesson capture and playback are often done at prescribed project milestones, rather than when the lessons are most relevant.
- Lesson content is hugely variable, leading to high-quality insight being lost in the noise.
- Lack of technical tools to support Lesson Learned practice means that even when motivated to capture or retrieve a relevant lesson, project team members find it difficult to record or search for relevant lessons.
There are two primary research areas within machine learning that are especially relevant to our vision for how Lessons Learned can be improved:
- Behaviour detection: Discerning patterns in project team behaviour to infer intention or predict outcomes in order to detect the optimum time to capture or play back a relevant lesson, thereby embedding Lesson Learned practice into day-to-day P3M.
- Natural Language Processing (NLP): To automatically extract and categorise lessons from relevant project documents to improve lesson searching and, in combination with behaviour detection, enable automatic lesson playback to the relevant person at the relevant moment.
Whose Line is it Anyway?
One of the challenges in the modern information space is mapping the messaging around certain topics: what are the dominant messages and who is propagating these. This is particularly true in contested domains such as discussions around climate change or tobacco regulation and control. Often ostensibly “independent” actors promulgate messages that are supportive of positions taken by, say, the fossil fuel industry or the tobacco industry: these actors are often funded, directly or indirectly, by these industries. In today’s fast moving information space, the challenge is how do we harness AI to quickly map the messaging around certain topics and readily identify the actors that may be aligned with certain interest groups so that these can be researched further.
- In a world of synthetic images, how can AI identify and prevent Non Consensual Intimate Image abuse?
- Can AI help corroborate those experiencing harmful sexual behaviour?
