Data 2020: AI and algorithmic accountability

Fri Feb 21, 2020
$download_content = get_field('download_content');

AI and machine learning algorithms are increasingly being used to make decisions – including decisions about us. But how are these decisions made?

Data 2020: AI and algorithmic accountability

AI and algorithmic accountability is one of the areas we’ve identified in our Data 2020 landscape review to help organisations understand hot topics in the world of data in 2020 – from digital competition to data rights

AI and machine learning algorithms are increasingly being used to make decisions – including decisions about us. But how are these decisions made? How can algorithms be interrogated or understood, and how do we ensure that unfair bias isn’t being built in – even unintentionally? More fundamentally, should some decisions be automated at all?

The UN Special Rapporteur on extreme poverty and human rights has highlighted the impact of AI and digital technologies on people’s lives. They are being used in crucial decisions such as eligibility assessment, fraud prevention and detection, and risk scoring and needs classification.

AI Now has detailed the range of algorithmic systems used by the public sector. The UK’s Centre for Data Ethics and Innovation is examining bias in algorithmic decision-making in financial services, crime and justice, recruitment and public services, and is creating recommendations about how any potential harms can be identified and minimised.

Transparency and an organisation’s ability to explain the AI algorithms it uses builds trust. It also enables teams to monitor how decisions are made, and if necessary address failings, bias or problems in the system. Understanding and auditing automated decision-making is critical for society – not just in terms of ensuring decisions are accurate, non-discriminatory and fair, but to ensure people can maintain their autonomy.

Upturn and Omidyar Network have described how to design systems for accountability. The Alan Turing Institute and the UK’s Information Commissioner’s Office have also developed guidance for improving the explainability of AI.

  • Regulating accountability and transparency mechanisms for private sector use of AI
  • Engaging citizens around public service automation, including the use of facial recognition and biometrics
  • Integrating AI into human decision-making
  • Improving the explainability of algorithms created by machine learning
  • Monitoring and responding to the impacts of algorithmic decision-making

This is not an exhaustive list of resources. If you provide tools or resources in this topic, please let us know by emailing [email protected]

  • AI Now
  • Ada Lovelace Institute
  • Centre for Data Ethics and Innovation
  • Etalab
  • Nesta
  • Open Data Institute
  • Partnership On AI
  • The Alan Turing Institute
  • The Institute for Ethical AI & Machine Learning

This is not an exhaustive list of all organisations working in this area. If your organisation is working on this topic and you’d like to be included in this list, please let us know via [email protected]

Want some strategic data advice for your organisation? Want to find out more about our research and development in this area? Fill in our form below and someone will be in touch soon…