Beyond Intellect and Reasoning: A scale for measuring the progression of artificial intelligence systems (AIS) to protect innocent parties in third-party contracts

By Acres A Stowe

The purpose of this book is to draw readers’ attention to various legal intricacies associated with deploying self-directed artificial intelligence systems (AIS), particularly emphasizing the limits of the law, vis-à-vis liability problems that may emerge within third-party contracts. With the advent of today’s ostensive “Amazon Halo or Alexa,” consumers are having to conclude contracts (e.g., sale of goods and distant financial services) in much more complex (cybernetic) environments. Generally, with one party acting in the capacity of a human being while the other (as an autonomous thing/device [AIS] with capabilities well beyond that of humans) representing the interests of others (not just other humans). Yet traditional jurisprudence is limited in scope for holding these systems legally accountable if they were to malfunction and cause harm. Interestingly, within the judicial system itself, the use of AIS is more prevalent now, including within the criminal justice system in some jurisdictions. In the United States, for instance, AIS algorithms are utilized to determine sentencing and bail processing. Still, jurists find themselves limited to traditional legal methodologies and tools when tackling novel situations brought about by these systems. For example, traditional strict liability concept, as applied in tort law, typically ties responsibility to the person(s) (e.g., AIS developers) influencing the decision-making process. In contract law, particularly where third parties are concerned, AIS are equated to tools for the purposes of traditional strict liability rules. Thus, binding anyone on whose behalf they would have acted (irrespective of whether such acts were intentional or foreseeable).

• Language: English
• Publisher: Page Publishing, Inc.
• Release date: Mar 31, 2022
• ISBN: 9781662466472

Book preview

Chapter 1

Introduction

The area of AIS is still considered a novelty devoid of any specific definition, although the concept was initially conceived as early as the 1950s.¹ Throughout the 1960s and over three decades later, these systems (considered as robots at the time) were reprogrammable and controllable.² However, it is widely accepted that the aptitude and intellectual capacity (albeit artificial) of present day systems can be equated to that of humans—superior in most cases, thus rendering them potentially unpredictable.³ Moreover, some of these systems (autonomous or semiautonomous) are thought to be dangerous and for legal reasons, hinge on whether the capacity exists for them to mimic the so-called reasonable person.⁴ AIS are taught (machine learning) via patterns initiated through training data (e.g., historical data relating to breast cancer screens validated by human professionals) in order to provide a model that will allow the system to eventually compare and analyze new similar input data and then determine the appropriate output.⁵ Based on that process, data is transformed into knowledge as a source of prediction—decision-making.⁶ Significantly, scholars tend to describe machine learning (deep learning) algorithms as black boxes because AIS autonomous processing ability is beyond the grasp, even of designers.⁷ Deep learning relies on a hierarchy of representation learning, producing different level of abstractions.⁸ Therefore, the outcome of applying one particular machine learning methodology from another, heavily rests on the training data (see Fig 1). So while on the one hand AIS possess invaluable capabilities, on the other hand, the consequences of intelligibility from training data can be cause for concern (legal and otherwise).

Fig 1.0 Deep Learning process—Goodfellow, Benjo and Courville (2016)

Undoubtedly, twenty-first century transformative technological advancements of AIS have improved efficiencies in numerous sectors globally. In healthcare practice, for example, the so-called Corti system is utilized to improve patients’ care (managing emergency dispatch processes),⁹ as well as Google’s algorithm developed to detect diabetic retinopathy in retinal fundus photographs.¹⁰ Nonetheless, the extent to which those systems would have considerably outperformed humans was first illustrated in a 2011 Jeopardy competition.¹¹ In that face-off (IBM AIS WATSON versus the two greatest Jeopardy champions), Watson behaved exactly as its human counterparts.¹² The system strategically pondered all the relevant questions (on topics such as: literature, pop culture, and sports). Once it had seemingly searched its database, it then indicated (by triggering a buzzer) and revealed the answer.¹³ Again, in 2016, two prominent AI systems (Google’s AlphaGo and DeepMind) competed in a prehistoric Chinese game—Go (similar to CHESS but much more complex) competition against an expert player.¹⁴ It is an extremely complex game because unlike CHESS where each player only has four hundred potential moves available after the first two turns, in Go, a player has 130,000 potential moves. Despite those intricacies Google’s AlphaGo faced, the AIS prevailed by imitating the behavioral patterns of its human opponent (the expert) and adjusting its strategic moves appropriately.¹⁵ Most remarkable though is the idea (albeit hypothetical) that AIS may eventually seek to invoke rights, for instance, the right not to be destroyed.¹⁶ A notion that was portrayed in a 2003 mock trial, where a computer sort legal redress to protect its right to exist.¹⁷ Even if this scenario seemed farfetched a decade ago, today the question is much more pressing as AIS become more