Delivering public services through the use of Artificial Intelligence (AI) has clear benefits, as does the increased use of data to inform decision making. These benefits have to be balanced with the recognition that they come with risks and challenges to governance processes, not least the ethical risks.

The Committee on Standards in Public Life fairly recently reported on the use of AI in public service delivery and decision making, and how such processes fulfil or otherwise the Nolan Principles. The findings were relatively reassuring, noting that AI and data can potentially lead to improved decision making ethically, and more objective decisions, alongside the obvious gains in efficiency.

As a case study, the use of data and AI in higher education has provided significant benefits to students, building a profile of individuals that enables universities to better teach and support each student. Automation can improve speed and accuracy of certain decisions, with chatbots for example providing swift answers to potential students.

However, AI systems are dynamic, especially as they become more sophisticated. Governance of these systems therefore requires continued checks on progress and a strong element of challenge.

The impact of diversity is one area of concern. AI systems can inherit biases that can exacerbate inequality. It is noticeable that the tech industry, who often have a big role in the design and implementation of these systems, is not very diverse and may not appreciate the biases being built in.

A proportion of individuals will always be better at understanding how systems work and taking advantage. AI systems can often have an approach where a large proportion of cases are decided automatically with a smaller proportion flagged for individual attention, perhaps to the detriment of those not assessed in such a way. This issue is another that needs to be continuously monitored and checked to ensure important cases are not being missed.

It is advisable to thoroughly investigate if an algorithm will work before putting it in place. Coming back to the example of higher education, while automation may work with the initial contact made with potential students, fairly soon students prefer the personal contact of an in person interview. Designing an automated, and expensive, interview system might improve efficiency but would likely discourage the bigger aim of increasing student recruitment.

There is a clear impact on the staff and the skills required to operate AI systems.

One issue is the likelihood that implementing AI will usually lead to redundancies. This is of course a challenge to how the workforce is managed and supported. However there are opportunities. Implementing such systems can take time, during which staff can be upskilled to better prepare them for the future and ensure their evolving skills don’t become obsolete.

A particular concern is that AI can often be best implemented to replace the work done by early entry and trainee staff. This clearly presents a challenge to career progression within sectors if opportunities to start out on a certain career path are greatly limited. Those training students need to ensure honest conversations about how required skillsets of the workforce are evolving.

The public perception of how AI is being used is vital to its success.

The public have to be able to trust the systems, for example to be assured that biases are not unfair, while acknowledging that all decision making has some form of bias whether using AI or not. Key to achieving this is to ensure that people feel they can have control over decisions being made. The onus on those operating the systems therefore is less in understanding the design of any algorithm and more knowing how the outcomes are derived.

There is a difference between complicated systems and complex ones. Complicated systems often rely on highly integrated systems combined to make something work, such as an airplane. People generally trust such systems to work without much investigation. Complex systems, often involving decisions about people, often require ethical decisions that people are less likely to readily trust.

There is clearly a requirement for the use of AI to be regulated, although currently some debate about how best to do this.

There is potentially a huge opportunity for the public sector to embrace AI, and the government has an opportunity to demonstrate what good looks like. The role of the regulator(s) will be to balance innovation over ethics, while noting these can often both be achieved. There is also the balance between collecting significant quantities of data to ensure AI systems can work effectively over the requirement not to overburden organisations and the public.

As noted above, all systems, involving AI or not, lend themselves to being gamed, and this extends to those organisations being regulated. There is a concern that this could be exacerbated by the use of AI in decision making that could lead to a scaling up of the ‘arm’s race’ between the regulated and the regulators.

There is some debate about how much of an impact AI is having, and whether decision making really is evolving that much.

It is clear that there is data on a much larger scale than before and a much larger use of self-learning machines. However, there is some dispute about how far AI will impact professions in the future, say for example the legal profession. Using computers is not always preferable and can lead to errors as described above. Implementing large scale IT based systems always bring risk, and require an understanding of ethics that many are not familiar with. It therefore requires some bold leaders and investment decisions to be an early adopter in new areas of AI use.

The event took place on 18 March 2021. Thank you to Lord Jonathan Evans KCB DL, Chair of the Committee on Standards in Public Life, Roger Taylor, Chair of the Centre for Data Ethics and Innovation and Professor Shân Wareing, Deputy Vice Chancellor at the University of Northampton for joining our panel.