Definitions of Intent for AI Derived From Common Law

EasyChair Preprint 4422

Abstract

The concept of intent is left undefined by the judiciary for jurors to decide. This poses a problem when considering intent in AI. There exist laws which are defined by intent, crimes which are intent specific (notably murder) and inchoate offences (intent to do x offences) which rely on establishing intent. As Autonomous Algorithmic agents (A-bots) grow in complexity, they will behave in ways deemed illegal if repeated by humans. AI designers must ensure that their A-bots do not intend to break the law. Equally prosecutors must develop tests to test intent in A-bots, both to establish whether they inherit intent for their owners or in the case when AI-personhood exists, whether the A-bot has the required mens-rea.

Keyphrases: AI, Intent, Reinforcement Learning, autonomous agents, causal reasoning, legal reasoning, mens rea

@booklet{EasyChair:4422,
  author    = {Hal Ashton},
  title     = {Definitions of Intent for AI Derived From Common Law},
  howpublished = {EasyChair Preprint 4422},
  year      = {EasyChair, 2020}}