Go back

Public Sector AI Strategies: Consideration for a Public Value Approach


Developing strategies for a public value approach to public sector Artificial Intelligence, the sixth paper of the Digital Revolution and New Social Contract research project.

Artificial Intelligence (AI) is becoming an increasingly common part of public sector technology and digitalization agendas. With rising investment and usage has come a parallel interest and demand for evaluating AI solutions from an ethical standpoint, identifying not only the social risks but the risks to fundamental rights, values, and obligations between citizens, society, and the state. While ethical reviews are increasingly common within public sector AI programs, researchers and policymakers need to better understand how the model of public administration impacts what kind of functions AI performs in the public sector and its impact on how the public sector can address public needs over time. 

This paper provides a brief introduction to a public value approach to public sector AI strategy development.

The author argues that AI solutions deployed within neoliberal and new public management contexts will reinforce neoliberal and new public management ends.

For instance, whether AI solutions, designed with public value intentions and operational context, may end up facilitating the reduced capability and privatisation of public sector and welfare-creating activities. 

New public sector programs need to better understand both what kinds of problems can be feasible and desirably addressed through AI, as well as what kind of AI solution best works when AI is identified as a means of addressing the problem. Some authors argue that efforts to apply technical solutions to social and technical problems without either understanding the problem or while over-prioritizing the solution can end up creating undesirable unintended consequences, as well as undesirable restructuring of social relationships. AI is not a solution for every problem; instead, what matters is understanding what problems may most benefit from a given solution or solution mix, for which AI may be one piece. 

There are different perspectives to public value and how AI generates that value. They are important because they raise questions that decision-makers need to address when designing policies for AI development and deployment in the public sector and beyond.

From a Moore perspective, the first area of interest is to understand how AI can improve the organisational capabilities and orientation of an organisation towards value creation. Booze says that what should be prioritized is the evaluation of what kinds of normative demands are implicated in or created by the increasing use of AI. The ethics within AI agendas are increasingly relevant, but AI ethics teams may serve as ‘ethics washing’ instead of endowing actors capable of identifying and detecting potential social harms from AI usage with the power to either change the deployment or fully alert public sector actors to secure larger reviews or algorithmic recalls.

Mazzucato and Kattel argue that the distribution of AI in society is a function of how the public and private sector relate to shape the existing market structures, such that some kinds of AI solutions are developed and diffused, and others are disincentivized and shut down. A pre-distributive approach to AI development is needed which better aligns the existing incentives of a given national innovation system, and the sub-systems around AI development as a technological innovation system. 

Out of considerations for public value and public value failures related to AI, six challenges emerge:

Capacity, political economy, structural inequality, dual-use, non-market, and epistemic challenges.

To address them, the paper proposes six policy recommendations. They include the establishment of a public value framework for public sector AI development and deployment; a clarification of the relationship between AI procurement and adoption and the pursuit of value-creating activities; a suitable framework to evaluate how and when the public sector may be subsidising a rent-extractive AI industry; a review at different administrative levels to identify the loss in public sector capabilities through direct automation, an EU conference and research engagement to align technical developments with social science research; and a European AI commons for best-in-class algorithms for public value oriented contexts.


Josh Entsminger,  DPhil Candidate in Innovation and Public Policy, University College London Institute for Innovation and Public Purpose

To read more about the topic and download the full paper, click here.