On 7th June, iNetwork hosted the event “An Intro to AI for Local Public Sector Managers”. The session was held virtually as part of the iNetwork Effective Information Sharing & Security (EISS) program with a presentation by Greig Feguson of Bridewell (formally known as Arculus Cyber Security).
The event served as an introduction to artificial intelligence and its role in the local public sector. It was open to all Northern WARP member organisations, specifically senior managers, to understand the basics of AI, how it can be used in the public sector, the emerging tools available, and also provided an opportunity to discuss and ask questions.
The presentation was split into five modules, the first of which was an introduction to Artificial Intelligence starting with the Oxford English Definition of Artificial Intelligence:
“The theory & development of computer systems able to perform tasks that normally require human intelligence, such as visual perception, speech recognition, decision-making, & translation between languages.”
Attendees were asked for suggestions for examples of what they believe meets this definition in their everyday lives leading to suggestions such as Netflix, Chatbots, Amazon sales and Predictive Text. These were all good ideas brought up by the group, to which Greig also added navigation data such as Google Maps and home voice automated devices like Alexa. Through the input of the attendees, the most popular example that was brought up was ChatGPT, something the majority of people at the session had previously used in some capacity leading to a debate around the positives of this (such as time saved and accessibility for neurodiverse individuals) and the negatives (such as concerns around accuracy including high profile incidents where this has been an issue).
The discussion then moved on to a brief history of AI, from the birth of the theory in the 1950’s with the work of Alan Turing, up to its current use in local Government today. This includes things such as public facing chatbots, image recognition to tackle littering and predictive analysis around issues such as homelessness. Although these and other applications of AI have the potential to bring numerous benefits, it is also important to be mindful of the risks and challenges which led to some interesting questions raised around this as this module was brought to an end.
The next module presented by Greig was the Basics of Artificial Intelligence. This started by looking at how AI works from a technical viewpoint. We were shown the process beginning with an AI system accepting a data input (by speech, text, image, etc), the system then processing the data using rules and algorithms to produce an outcome, this then being assessed through analysis and feedback before finally using this assessment to adjust input data, rules, algorithms, and target outcomes. Although this is the basics of how it works, the attendees were informed that to qualify as artificial intelligence, there are also several key components that need to work in conjunction. There are numerous components, however, Greig highlighted three key ones to expand on with further details. The first of these was cognitive computing, a process that uses data mining to mimic the human brain. The second of key components was Natural Language Processing (NLP) which is what enables the AI to understand, interpret and generate human language, highlighted by its use in popular applications such as Google translate. The final component covered was computer vision, being a field of AI that enables computers to interpret visual information such as videos and images.
After looking at how AI works, the session moved on to look at the types of AI with these being divided into two categories – capability and functionality. Greig broke down the capability category into three further sections using real world examples. These are:
– Narrow AI – a goal orientated AI trained to perform a specific task such as Siri or Alexa.
– General AI – whereby intellectual tasks are performed with human-like efficiency such as ChatGPT.
– Super AI – artificial intelligence that surpasses human intelligence and can perform tasks better than a human, we have seen glimpses of this with a machine designed for one specific task such as playing chess, however, we have not yet seen real life examples of this in a machine for multiple tasks.
Similarly, the other category, functionality, was also broken down into further sections:
– Reactive machines – basic AI that do not store past experiences for the future such as IBM.
– Limited memory machines – these store and use past experiences or data for a short length of time such as the emerging technology of self-driving cars.
– Theory of Mind – can understand human emotions and socially interact like a human. In the real world we do not quite have this technology yet.
– Self-aware AI – a super-intelligent machine with consciousness, emotions and beliefs which again we do not have real world technology at this level yet.
The explanation of these points provided several insightful comments and opinions from the attendees, including people’s opinions on the need for governance behind it.
Greig then moved on to the third module with a look at AI in the public sector. This started by looking at the public sector areas that could benefit from AI such as in Healthcare, including using this to analyse medical images, public safety where AI can be used to identify crime patterns and predict potential activity or in transportation where artificial intelligence could improve road safety or traffic flow.
Seeing areas where this could be used, the presentation moved on to the natural next point of whether it should be, with the benefits and challenges highlighted. This provided several interesting points with comments being raised by attendees in relation to people’s privacy or ethical concerns together with the balance that needs to be struck between long term benefits versus the immediate outputs.
It was interesting to hear the varying viewpoints of attendees on this topic. Greig followed this by informing of the current guidance and approach taken by UK law, particularly the prescriptive legislative framework based on Product Safety Laws that was ratified recently on 25th May. It has not yet been approved as law, however, it can give us an idea of the approach that the UK is looking to take.
The session then progressed to the fourth module with a look at a case study and emerging tools. The case study was on the FSA (Food Standards Agency). During the pandemic, the FSA accumulated a backlog in relation to local authority inspections so to deal with this a tool was developed to provide a predicted food hygiene and compliance ratings prior to an inspection taking place. This was done using a machine learning based web app that had been trained on 120 data points from a range of data sources.
Greig took us through the case study from a risk assessment perspective. We were shown the risks that would be associated from this undertaking such as potential bias from the model or the information required to be able to manage and securely use this personal information. After looking at the risks, we then looked at what processes were then put in place to mitigate these with actions such as having a cyber security assurance undertaken of the system and an integration of fairness related tooling to detect and mitigate potential biases.
Attendees were then taken through an example of an emerging tool with Microsoft co-pilot. The goal of co-pilot is to provide real-time digital assistance to help with productivity. This is an interesting example with several attendees already aware of some of the capabilities of the program leading to comments from multiple attendees discussing areas such as the cost versus benefit factor and strategies for adoption such as implementing new technology in bite sized chunks to try and reduce any fear factor.
The day’s event was then rounded off with the final module of the day being AI in Bradford & beyond with an interactive discussion. This was a conclusion of the presentation to sum up the key points raised, and to also receive further comments from attendees on their current thoughts and opinions leading to some interesting debate. We will be hosting more events related to AI in the upcoming months and encourage members to visit our website and to attend more of iNetwork’s future events.