CDEI publishes reports on online targeting and bias in algo decisions

The government-backed Centre for Data Ethics and Innovation (CDEI) has published two linked reports into online targeting and bias in artificial intelligence (AI)-powered decision-making.

The first, the 25-page Review into online targeting, focuses on the “ad-tech ecosystem” and states: “As targeting has become more accurate in its predictions and more powerful in its ability to influence our behaviour, concerns have grown about the extent to which we understand the way it influences our individual decisions and the impact it is having on our society.”

The review looks at public attitudes towards online targeting, regulation and governance of it, and possible solutions. So far, it has found “that people’s attitudes towards targeting change when they understand more of how it works and how pervasive it is”.

As regards governance, the report says: “Any changes to oversight mechanisms need to take into account how responsibility should be split between different actors.” And in terms of solutions, it says: “In some areas, stronger regulations may be needed. In other areas, greater transparency and visibility of how targeting operates may be more useful. Giving individuals stronger controls or rights over how data about them is used may provide both protection from harm as well as opportunities for innovation.”

The second report, the 25-page Review into bias in algorithmic decision-making, looks into data, tools, techniques and governance in respect of AI-powered decision-making in policing, financial services, recruitment, and local government.

So far, the review has identified a “tension between the need to create algorithms which are blind to protected characteristics [such as gender or ethnicity], while also checking for bias against those same characteristics”.

With respect to tools and techniques, the CDEI’s early work has found “there is limited understanding of the full range of tools and approaches available (current and potential) and what constitutes best practice. This makes it difficult for organisations that want to mitigate bias in their decision-making processes to know how to proceed”, says the report.

The CDEI has also zeroed in on the fact that algorithmic decision-making is bereft of the human ability to trade off competing values. “Effective human accountability for the use and performance of algorithmic tools will be critical regardless of context”, it says. “However, the form of that accountability and the mechanisms required to make it effective will differ. Our sector approach will allow us to test this hypothesis.”

In a statement accompanying the reports, Roger Taylor, chairman of the CDEI, said: “These reports set out the issues we are addressing and our approach, an update on our progress to date and how we will spend the next few months as we work up towards final recommendations to the government.”

In the foreword to the reports, Taylor said: “There are significant rewards for societies that can find the right combination of market-driven innovation and regulation to maximise the benefits of data-driven technology and minimise the harms. The UK, with its robust legal and regulatory systems, its thriving technology industry, and its leading academic institutions is well placed to achieve this.”

He added: “While data-driven technology continues to develop at great speed, there is no shortage of predictions about its future impact. But some of the challenges are with us now and are not merely theoretical.”

The CDEI will publish its final reports with recommendations to the government in December 2019 for online targeting, and in March 2020 for the bias in algorithmic decision-making.