In place of the traditional epistemological view of knowledge as justified true belief we argue that artificial intelligence and law needs an evidence-based epistemology according to which scientific knowledge is based on critical analysis of evidence using argumentation. This new epistemology of scientific evidence (ESE) models scientific knowledge as achieved through a process of marshaling evidence in a scientific inquiry that results in a convergence of scientific theories and research results. We show how a dialogue interface of argument from expert (...) opinion, along with its set of critical questions, provides the argumentation component of the ESE. It enables internal scientific knowledge to be translated over into a wider arena in which individual nonexpert citizens and groups can make use of it. The external component shows how evidence is presented and used in a legal procedural setting that includes fact-finding, weighing the credibility of expert witnesses, and critical questioning of arguments. The paper critically reviews the standards of admissibility of scientific evidence using the ESE. (shrink)

This paper argues for a coherentist theory of the justification of evidentiary judgments in law, according to which a hypothesis about the events being litigated is justified if and only if it is such that an epistemically responsible fact-finder might have accepted it as justified by virtue of its coherence in like circumstances. It claims that this version of coherentism has the resources to address a main problem facing coherence theories of evidence and legal proof, namely, the problem of the (...) coherence bias. The paper then develops an aretaic approach to the standards of epistemic responsibility which govern legal fact-finding. It concludes by exploring some implications of the proposed account of the justification of evidentiary judgments in law for the epistemology of legal proof. (shrink)

While courts depend on expert opinions in reaching sound judgments, the role of the expert witness in legal proceedings is associated with a litany of problems. Perhaps most prevalent is the question of under what circumstances should testimony be admitted as expert opinion. We review the changing policies adopted by American courts in an attempt to ensure the reliability and usefulness of the scientific and technical information admitted as evidence. We argue that these admissibility criteria are best seen in a (...) dialectical context as a set of critical questions of the kind commonly used in models of argumentation. (shrink)

This paper puts forward three critiques of pardo’s second-order proof rules thesis. The first criticism states that these rules are not suitable to guide the interpretation of standards of proof rules because they confuse matters of legal interpretation with matters of epistemology. The second criticism states that second-order proof rules are affected by the same indeterminacy problems they are designed to resolve, thereby rendering them unsuitable for the task they are purposely designed for. The third criticism renders pardo’s proposal redundant. (...) a reconceptualization of second-order proof rules is offered. (shrink)

The danger of human operators devolving responsibility to machines and failing to detect cases where they fail has been recognised for many years by industrial psychologists and engineers studying the human operators of complex machines. We call it “the control problem”, understood as the tendency of the human within a human–machine control loop to become complacent, over-reliant or unduly diffident when faced with the outputs of a reliable autonomous system. While the control problem has been investigated for some time, up (...) to this point its manifestation in machine learning contexts has not received serious attention. This paper aims to fill that gap. We argue that, except in certain special circumstances, algorithmic decision tools should not be used in high-stakes or safety-critical decisions unless the systems concerned are significantly “better than human” in the relevant domain or subdomain of decision-making. More concretely, we recommend three strategies to address the control problem, the most promising of which involves a complementary coupling between highly proficient algorithmic tools and human agents working alongside one another. We also identify six key principles which all such human–machine systems should reflect in their design. These can serve as a framework both for assessing the viability of any such human–machine system as well as guiding the design and implementation of such systems generally. (shrink)