Automated Decision-Making in Government: Key Insights from Lord Sales’ Keynote Lecture

On 5 November 2025, Lord Sales delivered a keynote lecture at the Government Legal Department’s Annual Conference titled AI and Public Law: Automated Decision-Making in Government. His remarks explored how artificial intelligence (AI) and automated decision-making (ADM) are reshaping and transforming public administration, and the challenges this poses for administrative law. 

The lecture emphasised that while automation offers efficiency and consistency, it also raises profound questions about accountability, transparency, and fairness. Lord Sales explored this issue through the lens of judicial review methodology, looking at the extent to which this core feature of administrative law is adequately suited to scrutinising the use of AI and ADM in the administrative sphere and suggesting how it will need to adapt.

This article summarises the key themes from Lord Sales’ lecture and considers their implications for the public adoption of ADM in Government.

The benefits and potential of ADM in public administration

After first outlining his understanding of ADM as “covering both AI and also simpler digitised systems and aids integrated in human decision-making processes”, Lord Sales delved into some background context concerning the benefits of ADM and its increasing prevalence in public administration.

He acknowledged the optimism surrounding ADM and noted that automation can deliver efficiency gains, reduce costs, and speed up decision-making. For example, he cited government ambitions to use AI to accelerate housing development approvals, which currently involve “4,000 pages of documentation and takes as long as 18 months from submission to approval.”

Other key benefits which explain why government bodies are increasingly piloting AI systems include:

• Efficiency: Faster decision-making with reduced labour costs (and the potential to feed resources to those who need them the most).

• Accuracy: Algorithms can detect patterns and trends that humans might miss.

• Auditability: Properly designed systems can leave clear audit trails.

• Consistency: ADM can promote the rule of law by eliminating capriciousness through consistent application of rules.

• Public law principles: ADM is capable (when used imaginatively) of making administration more responsive to individual needs. For example, it may allow for systems focused on people’s individual circumstances rather than applying homogenous treatment to categories of people.

Risks and Challenges

Despite these benefits, Lord Sales cautioned against assuming ADM will automatically improve administration. He emphasised several risks, highlighting the tension between technological innovation and the values of administrative law. He cautioned that increasing demands for administrative efficiencies may be accompanied by: 

“...significant risks in terms of enhancement of state power in relation to the individual, loss of responsiveness to individual circumstances and the potential to undermine important values which the state should be striving to uphold, including human dignity and basic human rights”

ADM and the accompanying digital revolution therefore present a challenge for Government to regulate it in a way that allows administrators to maximise the benefits of such technology while also protecting administrative law and the public from its risks. 

Judicial review methodology

In the main body of his speech, Lord Sales set out the processes of administrative decision making and judicial review, explaining how ADM poses novel challenges to some of their more traditional methodological elements. In particular, he highlights the following challenges:

1. Opacity: “Blackbox algorithms have properties that can make them opaque, so that it is difficult to understand not only how but also why a decision was reached.” In relation to requirements to exercise powers for proper purposes and the duty to give reasons (where applicable), the lack of transparency behind the decision-making process may impact the legitimacy of the decision itself.

2. Bias and discrimination: AI trained on statistical data may infer trends that disadvantage certain groups. Combined with the above point on opacity, this may be hard to identify and escape detection.

3. Accountability gaps: Delegation of decision-making to machines raises questions about who is legally responsible. Do public bodies have the legal power to use ADM and is a decision by an ADM system a ‘decision’ legally for the purposes of the source of the power or duty?

4. Fettering of discretion: Algorithms may apply rigid criteria, undermining flexibility in discretionary decisions. There are concerns that ADM algorithmic results may become relied on unthinkingly in the exercise of discretionary power and there is also a question over what happens if someone presents it with factors that are relevant, but which the ADM is not trained for.

5. Relevant and irrelevant considerations: Lord Sales warns that that the nature of ADM systems means they can potentially incorporate factors that are legally irrelevant or fail to account for those that are relevant. 

6. Fairness and dignity: A requirement of procedural fairness will in some circumstances require a public body to hold an oral hearing so that the affected individual can make representations. ADM cannot replicate the value of a human acknowledging and empathising with an individual. ADM also fails to account for the nature of individuals as moral agents who can break free of past behaviours and so Lord Sales calls for human decision-makers to remain in the loop when reviewing ADM system outputs.

7. Evidence and the burden and standard of proof: ADM creates evidential challenges because its processes are often opaque, making it hard for claimants to show what factors influenced a decision. For example, individuals may struggle to discharge the burden of proof without disclosure of algorithmic logic or data inputs, and courts may need to adapt by requiring authorities to provide fuller explanations. 

8. Systemic issues: algorithmic processes can embed structural biases or errors across large volumes of decisions, not just isolated cases. Lord Sales stresses that courts may need to look beyond individual outcomes to assess whether the underlying model or data creates a persistent risk of unlawful decision-making.

9. Remedial discretion and procedural issues: how should a court address a scenario in which it determines that a decision made by an ADM system was unlawful when there may have already been a number of similar unlawful decisions made against other individuals?

Comment

Lord Sales acknowledged that in many ways the existing judicial review machinery is already equipped to scrutinise decisions made by ADM systems as judges have already confronted many of the same issues in the past with other statistical and technical aspects of administration. However, he concludes that without explicit guidance from Parliament, the courts will have to confront and adapt to the above issues so as to maintain key legal principles and values without nullifying the benefits which ADM has to offer. Lord Sales indicated that explicit legislative guidance could help avoid leaving courts to improvise and stressed the importance of meaningful human oversight to prevent abdication of legal responsibility.

For queries or advice on the content of this article, please contact Tom Whittaker, Amanda Leiu or a member of Burges Salmon's Commercial & Technology team. 

This article was written by Fraser Campbell and Amanda Leiu.