Developing Law And Regulation For Technology Presented by: Mrs. Williams Date: March 3, 2024 The Coming Challenge - Even lawyers in conventional practice will increasingly need to understand Al and other forms of technology, and the impact these technologies will have on the existing law. - In the third section of our Law & Computer Science course (about which more next week) we go through a different area of law each week, examining how that area will need to adapt its rules designed for humans in order to ensure that the rules operate effectively in relation to technology. Areas Covered: - Cryptocurrencies and property law - Algorithmic collusion and competition law - Algorithmic discrimination and employment law - Algorithmic decision making and public law - Privacy, Security and Identity, - Criminal liability and digital evidence - Tort liability for autonomous systems. Impact Beyond The Course - R Williams and T Melham, 'Automated decision-making in the public sector' Thomson Reuters Practical Law Public Sector 8/12/20 - R Williams, 'Rethinking Administrative Law for Algorithmic Decision Making' 2021 OJLS (forthcoming). - Briefing paper for the Legal Education Foundation's Response to Law Commission consultation on its 14th Program of law reform (with Reuben Binns and Lilian Edwards) - Presentation for Law Commission/SLSA 14th Program Event - Civil Service Live Event - One unit of the OLTEP 'AI University' Module in conjunction with the Government Legal Department, Non Departmental Public Bodies and, shortly the Office for Al. Algorithmic Decision Making and Public Law The grounds of Administrative Law - It must follow a fair procedure: not be biased, hear from the right people, give reasons for its decision(?) - The public authority must have jurisdiction to make the decision. - It must take into account all the right considerations and only the right considerations. - It must not fetter or delegate its discretion. - Its decision must be reasonable or proportionate. Does ADM present new challenges? - In some ways no: • Humans are not perfect in their decision- making (see, e.g. Algorithmic Transparency for the Smart City' by Brauneis and Goodman; 'Emotional Judges and Unlucky Juveniles' by Eren and Mocan). • Administrative errors and 'algorithms' can apply in an analogue context too, see e.g. EK (Ivory Coast) [2014] EWCA Civ 1517 and R(Guittard) v Secretary of State for Justice [2009] EWHC 2951 respectively. • - shouldn't just assume that because a case involves ADM it is inevitably novel. But in some ways yes • Transparency  Automated systems can be simultaneously more and less transparent than the human equivalent.  Less transparent: non-linear systems (machine learning).  More transparent: specificity of instruction, metrics of accuracy, training method, even forms and mechanisms of discrimination are more predictable and transparent. Greater surface area for admin law to bite. • Metrics of accuracy  A number of different metrics can be used to assess a system, e.g, how many of the target instances did it detect? How many of the positive results it generated were actually positive? How many of its negative results were true negatives? How many times out of the whole sample did it correctly classify a result as positive or negative?  This is of course not unique to ADM. We're very used to dealing with this in the context of COVID-19 tests at present.  But the question of which metric makes a system, or deployment of that system, e.g. reasonable, is not one that public law has yet had to answer (though see Bridges [2020] EWCA Civ 1058 on PSED) • Correlation as opposed to causation  NB that ML relies on statistical inferences, not reasoning, so to what extent is mere correlation sufficient to justify the use of a factor in making a prediction/determination?  What if an ADM system makes decisions on the basis of a feature which correlates with a particular outcome, but is not causative of it?  Would such a feature be a relevant consideration? The Good News - Although there are some challenges, there are clear ways in which public law is already well set up to help. Ways In Which Administrative Law Can Help - GDPR Arts 13(2) (f), 14(2)(g) and 15(1)(h) - stat that the data subject has the right (inter alia) to 'meaningful information about the logic involved in ADM. - The meaning of this phrase has been the subject of a lively academic debate (Goodman and Flaxman, Wachter, Mittelstadt and Floridi, Selbst and Powles). Notice - Duty to give notice well established at common law. - Even in CMPs D must often be told the 'gist' of the case against him/her, (AF (No 3) A v UK). - In Bourgass Lord Reed held that 'a prisoner's right to make representations is largely valueless unless he knows the substance of the case being advanced... that will not normally require the disclosure of the primary evidence... [but] what is required is genuine and meaningful disclosure of the reasons why [the decision was made].' - General statements about the prisoners' behavior, or risk etc. was not sufficient. - So could this be what 'meaningful information about the logic involved' means? Cf Selbst and Powles. - NB also the ICO guidance: 'It is vital that individuals understand the reasons underlying the outcome of an automated decision, or a human decision that has been assisted by the results of an Al system. If the decision was not what they wanted or expected, this allows them to assess whether they believe the reasoning of the decision is flawed. If they wish to challenge the decision, knowing the reasoning supports them to formulate a coherent argument for why they think this is the case.' Other Examples - The GDPR also refers to the concept of 'proportionality', e.g. processing of Art 9 sensitive data must be, inter alia, 'proportionate to the aim pursued'. • Again, Administrative law has a wealth of case law on the meaning and applicability of proportionality. - Over-reliance on automated systems and over-rigidity of decision making • Rules on fettering and non-delegation. The Public Law Toolkit Beyond Public Law? - But actually, there is a separate question whether, if they are developed properly, these tools could provide techniques for use outside public law. - One of the key determinants of the application of public law is the imbalance of power between the decision-maker and the private individual, which also applies more generally to the use of ADM... - NB public law is 'leaking' outside traditional boundaries anyway: Braganza v BP Shipping Ltd & Anor But... - Lawyers and computer scientists will need to work closely together to understand and set standards for, e.g. the choice of particular systems for particular contexts, the metrics used to assess the performance of those systems. - And this will need to happen both when the systems are deployed in the first place and when their deployment or decisions are reviewed. - We can meet these challenges, but we can't do so alone, we need a fully interdisciplinary approach.