The tax system incentivizes automation, even in cases where it is not otherwise efficient. This is because the vast majority of tax revenue is derived from labor income. When an AI replaces a person, the government loses a substantial amount of tax revenue - potentially hundreds of billions of dollars a year in the aggregate. All of this is the unintended result of a system designed to tax labor rather than capital. Such a system no longer works once labor is capital. Robots are not good taxpayers. The solution is to change the tax system to be more neutral between AI and human workers and to limit automation’s impact on tax revenue. This would be best achieved by reducing taxes on human workers and increasing corporate and capital taxes.

This chapter explains the need for AI legal neutrality and discusses its benefits and limitations. It then provides an overview of its application in tax, tort, intellectual property, and criminal law. Law is vitally important to the development of AI, and AI will have a transformative effect on the law given that many legal rules are based on standards of human behavior that will be automated. As AI increasingly steps into the shoes of people, it will need to be treated more like a person, and more importantly, sometimes people will need to be treated more like AI.

This chapter defines artificial intelligence and discusses its history and evolution, explains the differences between major types of AI (symbolic/classical and connectionist), and describes AI’s most recent advances, applications, and impact. It also weighs in on the question of whether AI can “think,” noting that the question is less relevant to regulatory efforts, which should focus on promoting behaviors that improve social outcomes.

AI has the potential to be substantially safer than people. Self-driving cars will cause accidents, but they will cause fewer accidents than people. Because automation will result in substantial safety benefits, tort law should encourage its adoption as a means of accident prevention. Under current laws, suppliers of AI tortfeasors are strictly responsible for their harms. A better system would hold them liable for harms caused by AI tortfeasors in negligence. Not only would this encourage the use of AI after it exceeds human performance, but also the liability test would focus on activity rather than design, which would be simpler to administer. More importantly, just as AI activity should be discouraged when it is less safe than a person, human activity should be discouraged when it is less safe than an AI. Once AI is safer than a person and automation is practicable, human tortfeasors should be held to the standard of AI behavior.

The impact of artificial inventors is only starting to be felt, but AI’s rapid improvement means that it may soon outdo people at solving problems in certain areas. This should revolutionize not only research and development but also patent law. The most important requirement to being granted a patent is that an invention must be nonobvious to a hypothetical skilled person who represents an average researcher. As AI increasingly augments average researchers, this should make them more knowledgeable and sophisticated. In turn, this should raise the bar to patentability. Once inventive AI moves from augmenting to automating average researchers, it should directly represent the skilled person in obviousness determinations. As inventive AI continues to improve, this should continue to raise the bar to patentability, eventually rendering innovative activities obvious. To a superintelligent AI, everything will be obvious.

This chapter starts Part II (domain dependent feature engineering) by describing the creation of the base WikiCities dataset employed here and in the next three chapters. This dataset is used to predict population of cities using a semantic graph built from Wikipedia infoboxes. Semantic graphs were chosen as an example of handling and representing variable-length raw data as fixed-length feature vectors, particularly using the techniques discussed in Chapter 5. The intention with this dataset is to provide a task that can be attacked with regular features, with time series features, textual features and image features. The chapter discusses how the dataset come to be, an Exploratory Data Analysis over it, resulting in a base, incomplete featurization. From there, a first featurization was produced, with an error analysis process including feature ablation and mutual information feature utility. From this error analysis, a second featurization is proposed and an error analysis using feature stability concludes the exercise. All insights are captured in the two final feature sets, one conservative and other expected to have higher performance.

This chapter closes Part I presenting advanced topics, including dealing with variable length feature vectors, Feature Engineering and Deep Learning and automatic Feature Engineering (either supervised or unsupervised). It starts bridging the pure domain independent techniques to start drilling into problems of domain-specific importance. Variable length feature vectors has been a problem for fixed-size vector ML ever since. In general, techniques involve truncation, computing the most general tree and encoding paths on it or just destructive projection into a smaller plane. The chapter briefly delve into some Deep Learning concepts and what it entails for feature engineering. Automated Feature Learning using FeatureTools (the DataScience Machine) and genetic programming is covered, also Instance Engineering and Unsupervised Feature Engineering (in the form of autoencoders).

This chapter concludes by responding to some of the controversies about artificial intelligence and possible criticisms of AI legal neutrality. It argues that AI legal neutrality is important regardless of whether AI broadly achieves superhuman performance, and that the law would not want to constrain AI development for protectionist reasons. It further argues that AI legal neutrality is a coherent principle for policymakers to apply, even though it allows the law to treat AI and people differently and will sometimes be at odds with other regulatory goals. Finally, it discusses some of the risks and dangers of AI and argues these are susceptible to management with appropriate legal frameworks.