AI Unpacked is a series of short explainer videos on artificial intelligence (AI) designed to provide officers and councillors who are new to the topic with a starting point for understanding and engaging with AI concepts.
What is artificial intelligence?
The first video defines artificial intelligence and its role in our daily lives. You'll learn about different types of AI systems, from narrow AI focused on specific tasks to more general-purpose applications, helping you understand how AI can support council services.
"You've probably heard a lot about Artificial Intelligence, or AI, in the last few years. News and social media has been flooded with stories and reports about AI and there’s been a global conversation about the balance between risks and supporting innovation.
But once we sweep away some of the complicated words, cliched imagery and science fiction… What is Artificial Intelligence? The term Artificial Intelligence can mean a lot of things, and there is no single definition that is suitable for every scenario, but one widely accepted meaning is that AI systems are machines that perform tasks that normally require human intelligence, and often learn from and use data to do those tasks.
The idea of AI has been around for a long time. Discussions around Artificial Intelligence began in the 1940s and 1950s with the advent of computers and had already entered homes by the end of the 20th century with the AI systems inside automated vacuum cleaners.
AI systems follow the rules of an algorithm, or a method of data analysis called machine learning, to make predictions and decisions. These systems are often used to do tasks that are traditionally thought of as requiring humans.
So, what are the different types of Artificial Intelligence? Here are a few examples.
Narrow AI is the type of AI that has been around the longest. It is an artificial intelligence system that has been developed or trained to do one specific task. For example, a fraud detection tool might be trained to look for patterns in a set of financial transactions and flag the need for investigations.
General Purpose AI involves using large quantities of data to train a system to perform a wide range of tasks. These AI systems can be used to generate content, translate languages or answer queries. They are often referred to as foundation models because they can provide a base system upon which more specific AI applications can be built.
You may also have heard of the term ‘Artificial General Intelligence, or AGI. This is an idea that computers will soon be able to achieve general intelligence and autonomy. But this is still, as yet, very much theoretical.
Questions:
Has your council started discussions about an AI strategy?
What existing processes in your council could benefit from AI support?
What training or guidance might you and your team need to work effectively with AI?
Watch more of our AI Unpacked series to explore answers to some of the most common questions about AI. These videos explain some of the terms and concepts to help you understand how you and your organisation might respond to the opportunities and challenges that AI brings."
What are algorithms?
Learn how algorithms work in AI systems and their role in council services. This video explains the difference between traditional and AI algorithms, helping you understand how they support decision-making while highlighting the importance of transparency, accountability and human oversight in their implementation.
"A traditional algorithm is a set of rules followed in an order or sequence. They are essentially used to automate a process.
Many council services use algorithms. For example, they are used to manage and sort data to assess eligibility for services such as school admissions or social services. Algorithms can process vast quantities of data, and work faster, than traditional human-led processes.
Traditionally algorithms have been written, or coded, by a human. But these general algorithms are different to AI algorithms. Once an AI algorithm has been coded, it can autonomously process data to identify patterns and relationships in the data.
AI algorithms form the logic by which an AI model operates. Such models can produce impressive outputs, mimic human behaviour and find patterns or trends that may be missed by humans.
While this has many potential benefits there are risks and limitations.
The models created from AI algorithms are very dependent on the data used to train them. So, if they are trained on incomplete or inaccurate data this will be reflected in the models, decisions and suggestions they create.
The complex way in which many AI algorithms and models link the data going in and the new data coming out means it can be difficult to explain in a clear and open way how and why the AI model has reached a conclusion or outcome.
So, while AI algorithms and models can help councils to save time, make decisions and more, it’s important that the data going in and coming out is accurate, reliable and fair.
Reflection Points:
Familiarise yourself with your council’s usage policy for AI algorithms and models, which might have been adopted or in development. Ultimately, staff and organisations are accountable for the outcomes that are created and used.
Revised Questions:
What role do algorithms currently play in your council’s operations?
Are you aware of ways to document and review algorithmic decisions?
What mechanisms exist in your council to oversee the use of automated processes?"
To be added
What role does data play in AI?
Understanding the relationship between data and AI is crucial for council services. This video explains how AI systems use data, why data quality matters, and what councils need to consider when using data in AI applications.
"All AI systems use data. Data plays a fundamental role in training AI models, prompting, and generating outputs.
AI, and in particular machine learning, relies on vast quantities of data to train its models. This data can be anything from text documents and images to sensor readings such as temperature and humidity. By analysing the data, AI systems learn to identify patterns and relationships. They then use these patterns to make predictions, decisions and outputs.
The data used to train an AI system is related to the task for which the AI is developed. A text generation AI system, such as a large language model, requires text data, whereas predictive analytics around road traffic might rely on sensor data.
The quality and quantity of the data an AI system uses has a direct impact on the accuracy, reliability and performance of an AI system. Data must be properly prepared for use in AI applications, no matter the size or complexity of the problem at hand. This can require significant time and effort but is essential for ensuring the accuracy and efficiency of AI systems.
For local government, data quality is important for a variety of reasons.
Much of the data used for AI systems that can help councils is data related to real people and communities.
For example, if data has been collected as part of a targeted intervention focusing on particular socio-economic or ethnic groups, the AI model using the data will be trained on, and find patterns that are reflective of the data. Any decisions and outputs that the AI generate will be reflective of the data used and could lead to biases.
While bias may be less important for data around traffic flow or water usage, to use AI to inform for urban planning, accuracy can matter a lot. For example, school term time traffic data will differ from an annual picture, as will data collected during lockdowns. If data is not consistent, well-structured and well governed, it will be hard to use AI effectively.
AI system development is an iterative process. As AI systems interact with the real-world and collect new data to use in its analysis and outputs, they can be programmed to learn and improve. This means the quality of the data being used needs to be continuously monitored.
Reflection Points:
Familiarise yourself with the principles of data security and privacy. As AI systems become increasingly integrated into the work of local government, it is vital that councils continue to manage data safely and responsibly while ensuring there is the right governance, human oversight and accountability for AI systems using personal data.
Questions:
How would you rate the quality of data in your council for potential AI use?
What steps could your team take to improve data governance and management?
Does your council have data sharing agreements that could affect AI implementation?"
What is generative AI?
Explore how AI systems can create content like text, images, and more. This video explains the capabilities of generative AI and its potential applications in council communications and service delivery.
"Generative AI systems can generate content such as human-like text, images, videos, music and code in response to questions known as prompts. Using these Generative AI systems can help organisations to develop new or faster approaches to the delivery of services.
For local authorities, speed is one of the main benefits. Generative AI systems can rapidly read, comprehend and extract data to generate multiple summaries or reports. Such systems can also accelerate communications workflows. They can create or generate ideas for images and graphic design for web pages, campaigns and brochures.
The abilities of Generative AI systems have increased enormously in the last few years due to an increase in digital data and computing power, improved algorithms and the sharing of computer code. But how do they work?
Generative AI systems use models trained on vast amounts of data. These datasets enable the system to create new content in response to a prompt or request. For example, a Large Language Model may be trained on enormous quantities of textual data, whereas an AI image generator may be trained on a dataset of billions of pre-existing photos or artworks. The outputs that are generated use, and therefore reflect, the datasets used to train the models.
Generative AI systems generate new content in response to a prompt. These prompts can be text inputs, visual inputs, spoken inputs and more.
Generative AI systems are commonly integrated into existing computer software. Recent versions of software such as Microsoft Office and Adobe Creative Cloud have generative AI tools built in, either as an integral feature or an optional extra. These could include tools that translate, summarise or generate images.
Reflection points:
How might your council demonstrate accountability in its use of AI? Generative AI systems are relatively new, and they are powerful tools with many applications. As they continue to develop, we can expect to see many more innovative uses. Councils generating content in this way must foster accountability to respond to a range of questions that exist around the ownership and use of the data and digital content used to train Generative AI models.
Questions on end screen:
Where could Generative AI tools be used in your work?
How might your council use Generative AI tools to innovate?
How might your council build transparency in its use of Generative AI?"
What is a large language model?
Understand the AI systems behind tools like ChatGPT and their potential uses in council work. Learn about their capabilities, limitations, and important considerations for responsible use in local government settings.
"A large language model, or LLM, is a type of generative AI system that can process and respond to human language text and prompts.
The use of AI tools driven by LLM’s, such as OpenAI’s ChatGPT, Microsoft’s Co-Pilot or Google’s Gemini, has increased recently, and there are many benefits that councils can gain from their use.
LLM’s are trained on huge amounts of data, which allows them to ‘learn’ the patterns and rules of language. This data comes from many sources including books, articles, websites, online forums, and code repositories.
Using the data they are trained on, large language models work, in effect, like the predictive text tool you might be familiar with from your phone. When given a prompt, they create a response. This can be surprisingly powerful. For example, an LLM that has been trained on billions of different types of documents may be able to create a convincing first draft of a specific document needed such as a letter, report and presentation. Many LLMs can also be used to extract information from documents and summarise or rephrase text. When trained on data in multiple languages, LLMs may also be able to provide credible translations of text.
There are several ways local authorities could use large language models. They could be used to summarise long reports. They can be used to generate or edit draft text required for communications, reports and documents. They can suggest responses to questions when knowledge is being collected or gathered.
Residents might also use LLMs to help them in their interactions with the council, such as writing applications or letters.
The outputs that large language models generate are very dependent on both the data they are trained on, and the phrases you use to guide the AI model’s output, known as ‘prompts’. Biases in the data used to train a large language model can lead to them generating responses that have explicit or implicit biases such as racial bias, sexism, ageism and ableism.
With some products, data that you input is stored and may subsequently be used in the ongoing training of the model. So being mindful of the data you select to input is very important, especially when dealing with personal or sensitive data.
Reflection Points:
Familiarise yourself with your council’s governing principles surrounding the use of large language models, which might have been adopted or be in development. LLMs are a rapidly developing field and have the potential to change the way councils deliver for their communities. But councils must find ways to demonstrate accountability for their use – including how they have effectively addressed the risks.
Revised questions:
Has your council identified appropriate uses for LLMs in daily work?
What policies guide your team's use of LLM-based tools?
How does your council ensure accountability for LLM-generated content?
Who in your team would be responsible for reviewing LLM outputs?"
- AI Playbook for the UK Government - GOV.UK (Sub-section on Large Language Models)
How can councils use AI?
Discover practical applications of AI in local government, from managing and analysing data to improving service delivery. The video explores how AI can help with tasks like automated text analysis, speech recognition, and image identification, while highlighting important considerations for implementation.
"Artificial intelligence offers councils many potential benefits and opportunities. It could help deliver services in more innovative ways and it could free up time to concentrate on tasks requiring human and real-life interactions. It could help increase productivity while saving time and money. So, what are some of the ways councils can and are using AI?
By using AI systems, councils can manage, sort and analyse vast amounts of data in ways that are much quicker than traditional human methods.
Text analysis systems can automatically sort, summarise and interpret incoming messages, reports, and other complex text.
Speech recognition systems can automatically transcribe meetings, provide transcriptions, and generate summaries or action points.
Image and video identification can be used to recognise number plates for road management, analyse satellite imagery to map green spaces and biodiversity, or track incidences of fly tipping.
More generally, AI can also be used to inform and suggest decisions and policies. Machine learning models, which are computer programmes that learn from data and algorithms, can process much more data than is possible through other means and they may be able to find patterns and trends that were previously missed.
For example, specialised AI systems can use population data, transport patterns and environmental information to help guide town and city planning. They can use data from multiple databases to identify or target potential social care interventions, or analyse population trends to identify communities that may have higher healthcare needs in the future. Some councils are already piloting initiatives like these.
Generative AI systems can also be used for communication. They can help with creating web or social media content, drafting letters and reports, generating images or videos and translating text into multiple languages.
So, as you can see, there are many potential benefits and opportunities for AI systems within councils. When considering AI systems, it is important to see them as powerful enablers, not as a standalone strategy. And, as always, understanding user need is step one of any successful deployment.
It is important to remember that AI systems are not perfect. The tools can make errors, miss key information and provide false information. They are dependent on both the data they use and the algorithms and code used to train them. It’s important that, within councils, AI systems are used to assist not replace human decision-making. Council staff, and councils as organisations, will always be accountable for the accuracy of data that their AI systems use and provide.
Reflection points:
Try speaking to colleagues across your council to find out if and how they’re using AI, and what benefits you might explore by using AI systems. If any of the terms used in this video are unfamiliar, please watch the other videos in the series to learn more.
Questions on screen:
Does your council have an AI strategy?
Does your council have any policies on responsible AI?
How might any existing policies on cyber security and data protection within your council inform the way you engage with AI at work?
Has your council established clear lines of accountability for using specific AI systems in its services?"
What is predictive analytics?
Learn how AI-powered analytics can help councils anticipate future needs and trends. This video explains how predictive analytics works and its applications in council services, from early intervention to resource planning.
"Predictive analytics uses data, statistics and modelling to predict future outcomes. By utilising AI for predictive analytics, councils have the opportunity to unlock greater value from larger data repositories to uncover insights into what might happen in the future.
Some councils have been predictive analytics for many years, particularly in service areas with large data sets, and where decision-making could be enhanced through data science techniques. By analysing data, historic or current, predictive analytic models can be trained to recognise patterns and relationships and make recommendations on courses of actions.
But now these methods are even more advanced. Predictive analytics and AI create a dynamic system that can predict outcomes with greater speed and accuracy by analysing larger quantities of data more quickly. This means AI-powered predictive systems can find patterns and trends and suggest outcomes that may have been slower or harder to find and predict before.
Councils will have many uses for AI-based predictive analytics. For example, to help prioritise proactive interventions for the most vulnerable children and adults, these techniques can be used to identify the young people most at risk of persistent absence. It can also be used to predict which council-managed properties are most at risk of damp and mould to respond quickly to the most vulnerable tenants.
These predictive analytics examples rely heavily on the datasets that are fed into the systems that create them. For example, when predicting which residents may be at risk of experiencing a fall at home, or who may need a more proactive care approach, the model can be tested using existing known examples.
Councils looking to use these techniques require strong data foundations across their organisation to ensure the data used for analysis is complete, up-to-date and good quality. Any predictions will reflect the data fed into the analysis, so if there are inaccuracies or biases in the data, this could be carried across into the outputs, leading to inaccurate or unfair outcomes.
Reflection points:
Councils wishing to use predictive modelling need to invest in good data science capabilities in the workplace. How skilled is your council when it comes to data? Where are the data skills gaps? What could be done to develop greater data literacy?
Questions on end screen:
Are you aware of your council undertaking a data maturity assessment?
If your council has limited in-house data skills, how might it work with partners to access more support?"
- AI Playbook for the UK Government - GOV.UK (Sub-section on Predictive Analysis)
What is human in the loop?
Discover why human oversight remains essential when using AI systems. This video explains different approaches to maintaining meaningful human supervision and establishing effective governance frameworks.
"Human-in-the-loop (HITL) describes a process of how machine learning and AI systems are used. HITL ensures humans have meaningful oversight of an AI system’s role, the data it uses and the outputs it generates in any given task. Essentially, it’s a collaboration between human intelligence and Machine learning and AI systems to achieve a goal.
In HITL, humans play an important role in supervising, guiding, or correcting the machine’s actions. It means there is meaningful human oversight in any use of AI systems. Machines can process vast amounts of data and identify patterns that humans might miss, while humans bring critical thinking, judgment, expertise, and the ability to handle complex situations that machines currently struggle with.
Human in the loop is an important process for local government for several reasons.
HITL processes can improve accuracy and trustworthiness. AI systems and machine learning tools can make mistakes, especially with complex tasks or unexpected situations. Humans can potentially provide a critical safety net in reviewing outputs, correcting errors and ensuring an AI system is functioning as intended.
A ‘human in the loop’ can also contribute to addressing bias and fairness, and protecting against unintentional harm. AI systems inherit biases from the data they are trained on. Given adequate time to consider data inputs and outputs, humans can play a role in identifying and mitigating against these biases. This is especially important in areas where AI is used to assist decision making such as policy decisions, particularly when human life or safety are dependent on it.
AI systems often struggle with complexities and nuance. They are unable to consider real-world context, personal judgements and ethical considerations. Humans bring critical thinking, problem solving and context that can better navigate, and make final decisions on complex issues.
AI systems are complex, and it is sometimes difficult to understand how they have reached decisions. HITL means that humans may be able to help interpret the AI’s reasoning and explain its outputs in a clearer and more understandable way.
HITL can operate at a number of levels. Staff might be involved in approving each individual automated decision. Or staff might review a selection of decisions, selected randomly, or as a result of appeals. And governance boards may provide regular human oversight of the overall functioning of a system, looking at trends and patterns. This kind of governance oversight could involve different stakeholders: including representatives of affected communities.
Reflection Points:
Consider the opportunities in your council for meaningful human oversight in decision making that is assisted by AI systems – this may include conversations with suppliers and partners. By using the strengths of both human and machine intelligence, councils can improve the accuracy, trustworthiness and fairness of their AI use.
Revised Questions:
How does your council maintain meaningful human oversight of automated processes?
What review procedures, if any, exist for AI-assisted decisions in your team?
Has your council established clear accountability frameworks for AI-supported decisions?"
What is responsible AI?
The final video explores the framework for ethical AI use in local government. Learn about key principles including fairness, privacy, transparency, and accountability, helping you make informed decisions about AI implementation.
"Artificial Intelligence systems can impact on people and society in good ways and bad ways. Responsible AI provides a framework for making good decisions on how Artificial Intelligence is used to manage its many impacts.
It is important that councils use AI systems in a responsible way – and know when it’s not appropriate to use AI. This will make sure that any use of AI makes a positive contribution.
So, what are some of the elements of Responsible AI?
Fairness, equality and inclusion must be considered. All AI systems are trained on data, and the quality of this data has a direct impact on the quality of the outputs that AI systems create. AI systems will reflect and reproduce existing biases and inaccuracies in datasets.
Privacy and data protection laws must be considered. Many uses of AI systems might involve processing personal data, such as when AI is used to write tailored letters to residents. Sometimes a resident might be asked to consent to, or opt out of, their data being used in this way. At other times, a public consultation might be needed to justify a council’s use of personal data in an AI system.
Transparency and explainability must be considered. Councils must be transparent with staff and residents about when and how AI systems are being used, and how any decisions enabled by AI have been made. Councils also need to consider what will need to be done if an AI system creates an output or process that does not fulfil its purpose.
Environmental costs must be considered. Many current AI systems have a high carbon and water-use footprint, and AI-based systems may use much more energy than the alternatives. Councils must consider the careful balance between scaling-up AI use and any sustainability impacts, in accordance with the council’s strategy and policies.
While AI systems can support everyday work and help in complex decision making, responsibility and accountability for tasks completed with AI rests with the person or people involved in the sign-off process. It is important to identify who is accountable for each part of AI system design, its deployment and its use.
Reflection points:
Familiarise yourself with the pillars of Responsible AI. Think about how AI is implemented, who it affects and how it affects them. Responsible AI is a framework that must be used in all instances where AI systems are used and integrated into council processes and work.
Questions on end screen:
Does your council have policies on Responsible AI?
How might any existing policies on cyber security, data protection and sustainability guide the way you engage with AI?
Has your council established clear lines of accountability for using specific AI systems in your service area?"
Highlighted pages
Artificial Intelligence Hub
Resources and guidance to help you explore the possibilities of AI, and connect with peers to ensure local authorities are part of an AI powered future.
Cyber, digital and technology
This hub brings together an overview of our support offer, and information on upcoming events, relevant publications and notable examples of practice, to improve the secure use of digital technology by councils and communities.