The application of traditional tort theory to embodied machine intelligence

摘要

The goal of increasing robot autonomy, or “machine IQ,” is to produce robots that can make real-time decisions in unpredictable environments in order to fulfill a set task. These robots, by definition, will take unpredictable or “unforeseeable” actions in the physical world they share with humans in order to fulfill the human-assigned task. Traditional tort theories of negligence and strict liability are insufficient to impose liability on the legal entities that sell or employ truly autonomous robots. The author defines “truly autonomous robots” as robots that embody machine learning. Where an autonomous robot makes an unpredictable move in order to attain a human-specified goal, liability would not attach to the manufacturer if these changes or methods made after product delivery to the consumer were unforeseeable. Foreseeability is an essential characteristic of the three types of product liability – failure to warn, design defect, and manufacturing defect – and ultra-hazardous activity theory is unlikely to assist us unless we are willing to say that all robotic actions are routinely, foreseeably, dangerous. The author discusses how the 1997 American Law Institute’s Restatement (Third) of Torts shifted the analysis of design defect theory from a consumer expectation analysis to a reasonable alternative design (RAD) test. Nonetheless, the focus continues to be on foreseeable risks: a type of predictable harm to a predictable group of potential victims. Two developments may assist this problem: using a common-sense approach applied to robots akin to the “reasonable person” analysis in tort law, and an increased ability to predict autonomous robot behavior as we continue to interact with them, such the development of reasonable expectations (and rights) regarding robot activity.