A fruitful AI Challenge Day 2023

The annual ART-AI 'AI Challenge Day' took place on Monday 30th January 2023 at the APEX City of Bath Hotel.

Main image: Cohorts 1-4 AI Challenge Day 2023

The annual ART-AI ‘AI Challenge Day’ was held on Monday 30th January 2023 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 very grateful to the following companies and organisations who spent the day with us presenting an ‘AI Challenge’ and hosting table discussions: BMT, Civica, CMS, Costain, Mystic AI, ONS, Techmodal and The London Data Company. The students really got involved with the problems presented to them. “Discussions were very interesting” one of our students wrote in the feedback, and another said “I liked the project diversity”.

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. Some of the next steps presented were ideas for academic papers, white papers and PhD projects.

Overall, the feedback was that the attendees “appreciated the time to network and collaborate” and there is a call for similar events in the future. If you are interested in coming along to present an AI Challenge next year at the AI Challenge Day, please do get in touch [email protected].

Programme 0900-17:00 (GMT)

09:00-09:30 Arrival Tea/Coffee

09:30-09:45 Introduction to the day

09:45-10:45 Presentation of AI Challenges from the table hosts

10:45-11:00 Mid Morning Tea/Coffee

11:00-12:30 Table discussions

12:30-13:30 Lunch

13:30-15:00 Table discussions

15:00-15:15 Afternoon Tea/Coffee

15:15-15:45 Prepare presentation of table discussion

15:45-16:45 Presentations/Feedback from table discussions

16:45-17:00 Closing the day

17:00 Drinks reception

‘Table Hosts’ and AI Challenges

Autonomous maritime navigation is set to transform the maritime industry providing consistent, continuous and cost-effective navigation free from human error. However, there remain a number of barriers to its full adoption including trust and regulation acceptance. One such regulation-based restraint is the International Maritime Organisation’s (IMO) Collision Regulations (COLREGs), written to ensure safe navigation through open water. These COLREGs require a “look-out”, a slightly ambiguous term that could be used to describe a human or equivalent automated system. Currently, this function is performed on almost all vessels by a human watchkeeper, but in the future could this be replaced with a computer?

Conversational AI: Conversational AI systems are gathering a lot of attention and there is discussion of their use for lonely or isolated individuals to provide interaction, our first question is: What are the use cases and implications of these solutions in the public sector?

Our second question relates to conversational AI and the principal of gamification, this being the use of interactive reward and scoring systems to make apps more engaging to the user. Conversational AI apps have started encouraging an emotional connection to the user (such as Replika) and used memory systems and avatars to anthropomorphise the solution. The creation of this attachment drives user engagement in a similar fashion to gamification and encourages users to return to the app – is emotive gamification ethical? Particularly for commercial apps?

As artificial intelligence continues to become increasingly more sophisticated how do we, as a society, attribute ownership (intellectual property rights) from the works created by AI whilst ensuring i) we promote fairness and future development and ii) do not, inadvertently, disincentivise collaboration.

Title:  Distributed sensor environment for anomaly detection


Multi-static Electro-Magnetic (EM) sensing has been demonstrated as means of passive radar sensing that is able to sense air movements at very low altitudes.  The ability of these sensor networks to be geographically distributed enables a variety of commercial applications in addition to its primary sensing function. 


There are number of challenges in engineering and deploying such systems:

  • Rapid deployment-Many applications of this system will require rapid deployment to support an operational need
  • Anomaly detection – Data collected from sensors will need to identify and classify objects base on a set of rules
  • Physical architecture optimization – Depending on the operational environment, the physical distribution of sensors will impact the performance of the overall system.  A variety of factors may need to be considered to optimally architect the system
  • Rapid training data synthesis – Data sets needed to train the anomaly detection algorithms will need to be rapidly determined in order to support the rapid deployment of the overall system

Title:  Automated assurance packages for infrastructure development


Major complex delivery programs in infrastructure reduce a single scheme design into smaller units called work packages.  These work packages enable the detailed planning and execution of construction processes at the most granular level.  Another major technical activity on complex delivery programs is Technical Assurance and the generation of assurance packages.  These assurance packages contain evidence that all requirements related to a given work package have been met.  Assembling assurance packages can be a time consuming and costly activity for each work package.  Programs with hundreds of work packages will have significant schedule challenges in delivering hundreds of assurance packages.


Multiple assurance packages that are related and interdependent can be combined into a single package without degrading the detail or value of its contents.  However, this is a time consuming and manual process.  The factors that determine if and how individual packages can be combined may be simplified algorithmically. 

Mystic AI makes it easy to work with machine learning models throughout the whole ML lifecycle and to deploy AI at scale. Our customers use our Pipeline AI toolset to upload and manage their models which they can then deploy on our serverless cloud. We are constantly looking for ways to improve the efficiency of our serverless cloud of GPUs. Our challenge is to optimise the decision process algorithm (task allocation) so that we can handle a higher volume of workloads across a fixed array of GPUs, reduce latency and minimise the ‘cold start problem’.

Our question is:

  • Is it possible to use AI to help with this optimisation?
  • If so, what are some suggested ways to approach the challenge?

The how we, whether we, and should we of Admin Data and AI: The ONS has long relied on the census to provide information on England and Wales’ populations. However, their infrequency leaves holes in the statistics in this rapidly changing world. Therefore, the ONS is turning to Administrative Data (a by-product of people’s interaction with public services, including sources from health, education etc.). Using this data brings its own problems, to discover patterns over time, the datasets must be linked longitudinally, but for multiple reasons this can be tricky (e.g., people who die abroad for instance are often not removed leading to 130-year-olds). AI would make finding patterns overtime easier and faster, to predict what could happen next and guide government spending, but there are three sticking points:

– How – Technically how would this be possible? Do the nuances in the datasets mean AI cannot be used or that multiple rules or caveats would be required to draw useable conclusions?

– Whether – The ONS is allowed to do its work thanks to the Digital Economy Act which allows the use of data to produce statistics, but nowhere does it mention the use of AI, would more legislation be needed first to say whether it is allowed? How does GDPR play into this?

– Should – How should it be communicated what is happening to public data and could people opt out, but would this ruin the statistics?


Techmodal work with a variety of organisations at differing levels of technical maturity. Even the most advanced organisation typically has a large volume of unstructured legacy information held on paper records. Being able to ingest those records and convert into data for processing in natural language models would be of significant advantage to our customers.

Limited digitisation has taken place, often scanning hand-written documents. However, this isn’t done with onward analysis in mind and typically serves only to reduce storage requirements. As a result additional pre-processing is needed to make these images suitable for conversion. The scanned images will be skewed, often with folds or creases and using a large amount of acronyms and other non-standard text.

Given that fully digitizing an organisation is difficult, we are hoping to demonstrate benefit from a hybrid approach where the initial capture point is still paper, but where the form is designed to make follow up analysis far easier.


Techmodal would like to address these legacy documents with technical innovations. Specifically:

A new data capture approach, that would improve the digitisation process. This may include printing additional features onto a hard-copy document for example use of corner markers to allow for simpler image correction or QR codes to pass meta-data through to an OCR model.

At which stage in the life-cycle management of autonomous assets could AI be of most benefit?

  • In the development of a product?
  • Business/marketing strategy?
  • Monitoring health of the product? (In service? On the market?)
  • Improvements? Technically? From a customer perspective?
  • End of life of products?

How could AI be developed for the identified topics above? What would be the benefits? Disadvantages?

Decision makers are often provided the outputs of multiple models to provide insight on a problem and facilitate decision making. At different decision points the insights from different sources can provide more or less accurate insights. How do we communicate to these non-technical users when they should trust an insight from one AI model over another at a given time?

Keep Reading

The 22nd International Conference on Autonomous Agents and Multiagent Systems

The 22nd International Conference on Autonomous Agents and Multiagent Systems (AAMAS) will be in person and takes place 29 May – 2 June 2023 in the London ExCeL conference centre. ART-AI staff members Marina De Vos & Julian Padget have had a paper accepted for this conference.

Read More

Joint NII/ART-AI Policy Modelling and Reasoning Seminar

Two ART-AI Staff members, Marina De Vos & Julian Padget are co-organising an invitation-only joint NII/ART-AI workshop at Shonan on Policy Modelling and Reasoning.

Read More

Alan Turing Institute’s AI UK 2023 

AI UK 23 will take place at the Queen Elizabeth II Centre in Westminster, London.

Read More

AI Challenge Day 2024

The annual ART-AI 'AI Challenge Day' will be held on Monday 29th January 2024, 09:00-17:00 (GMT) at the APEX City of Bath Hotel, James St W, Bath BA1 2DA.

Read More