Automated Decision-Making (ADM) systems range from social media to traffic management, and are populated by machines – arrangements of software, rules, protocols and devices that identify, acquire, collect, manage and process data.
As part of their operations, these machines make sequences of choices and predictions that may have far-reaching consequences. For example, such machines may post anti-vaccination or pro-vaccination information to your Facebook feed, or diagnose that you have cancer, or predict that your child is at risk of abuse from your family.
ADM machines are incorporating extraordinary developments across a range of technologies. In addition to the explosion and distribution of data sources and operations described in the Data research program, communications infrastructure, distributed sensors and processing, web-controlled devices and protocols for verifying transactions have all evolved rapidly.
Simultaneously, major advances in machine learning, deep learning and artificial intelligence, along with new generative approaches to synthesising new content, have dramatically expanded the scope of ADM across a range of social domains accompanying considerable legal, ethical, social and political implications.
The Machines research program will focus on four distinct examples of decision-making machines – search engines, risk profiling systems, recommendation systems, and smart contracts, also known as automated agreements – all with distinct social and technological contexts, histories and capacities for ethical design.
We will conduct a systematic overview of the ADM landscape to develop an innovative three-dimensional taxonomy of decision-making machines and provide a categorisation of ADM that will support work across the Centre. A classification of ADM design and deployment will enable us to measure the expansion and diversity of ADM in comprehensible and analysable terms. We will analyse patterns of observed effects to disentangle the different outcomes of ADM, enabling us to develop common approaches to problems across different circumstances and contexts. We will work with industry partners to study how machines are developed and deployed and how they are embedded in human and organisational processes, and analyse the encoding of fairness, accountability, and transparency in the technical infrastructure of automated systems. We will explore how probabilistic ADM differs from human reasoning, and the implications for the measures of equity, inclusion, and accountability in decision-making.