How Can I Tell If My Algorithm Was Reasonable?
45 Pages Posted: 12 May 2020
Date Written: April 17, 2020
Self-learning algorithms are gradually dominating more and more aspects of our lives. They do so by performing tasks and reaching decisions that were once reserved exclusively for human beings. And not only that—in certain contexts, their decision-making performance is shown to be superior to that of humans. However, as superior as they may be, self-learning algorithms (also referred-to as artificial intelligence (AI) systems, “smart robots”, or “autonomous machines”, among other terms) can also cause damage.
When determining the liability of a human tortfeasors causing damage, the applicable legal framework is generally that of negligence. To be found negligent, the tortfeasor must have acted in a manner not compliant with the standard of “the reasonable person”. Given the growing similarity of self-learning algorithms to humans in the nature of decisions they make and the type of damages they may cause, several scholars have proposed the development of a “reasonable algorithm” standard, to be applied to self-learning systems.
To date, however, the literature has not attempted to address the practical question of how such a standard might be applied to algorithms, and what the content of analysis ought to be in order to achieve the goals behind tort law of promoting safety and victims’ compensation on the one hand, and achieving the right balance between them and encouraging the development of beneficial technologies on the other.
This paper analyses the “reasonableness” standard used in tort law, as well as the unique qualities, weaknesses and strengths of algorithms versus humans, and examines whether the reasonableness standard is at all compatible with self-learning algorithms. Concluding that it generally is, the paper’s main contribution is its proposal of a concrete “reasonable algorithm” standard that could be practically applied by decision-makers. Said standard accounts for the differences between human and algorithmic decision-making, and allows the application of the reasonableness standard to algorithms in a manner that promotes the aims of tort law and at the same time avoids a dampening effect on the development and usage of new, beneficial technologies.
Keywords: algorithms reasonableness, the reasonable algorithm, AI and Tort Law, Damage by self-learning algorithms
Suggested Citation: Suggested Citation