Democracy on Mars 4: New Tools for State Capacity
Note: this is just a short preview of a full-length post that remains under construction. It is also the fourth post in a series. The earlier posts are here, here and here.
This section will describe how new capabilities, especially AI, could increase state capacity – a prospect with a lot of promise but also significant risk. If not paired with popular sovereignty enhancements such as those described in the previous post, these applications would enable autocracies to wield unprecedented power.
I will outline two main categories of capacity improvements, which can be broadly described as increasing the subtlety and the scope of government power:
Subtlety. Today, encoding laws is mainly done using natural language, but the Martians might instead harness higher-fidelity (but less interpretable) representations produced by neural networks. Natural language is a form of lossy compression that forces legislators and regulators to choose between nuance and specificity: a law that prohibits driving above 55 mph can be enforced consistently, but clearly fails to account for many contingencies that determine safety (rain, lighting, traffic, etc.), whereas a prohibition on “reckless motoring” allows a judge, jury or policy officer to take many factors into account, but it leaves the door open to a near-infinite variety of interpretations, undercutting the premise of equality before the law.1 Natural language is not suited to detailing the interplay between a large number of contextual factors with the precision necessary for consistent implementation. This section will relate these issues to classic framings in legal philosophy (e.g., between rules and standards), and then describe how AI could bypass natural-language descriptions and allow lawmakers – or the general public – to craft far more precise norms that still have the scalability of legal rules. Innovations in this direction have the potential to produce more consistent, nuanced legal decisions, but they also raise major issues in terms of interpretability and equity.
Scope. Many phenomena lie beyond governments’ ability to control fully because they are too complex, distributed and fast-evolving, but AI – combined with other technologies, such as cheap sensors and real-time data sharing – could mitigate these constraints. For example, many historical attempts at price-setting have failed, but with access to more information and computational power, new forms of centralized economic management may become feasible. Similarly, efforts at creating planned cities have been hamstrung by crude simplifications of human behavior, but AI could provide far richer models of urban population adaptation. Such capabilities could improve infrastructure and facilitate fairer economic outcomes, but autocracies could also harness them for propaganda, cronyism and social control. Even if aligned with uncontroversially positive goals and tethered to detailed democratic input, such systems create questions about what level of statism people may find desirable, and how this trades off against other values such as various definitions of economic and social freedom.
Next post: Philosophical questions (still under construction, the linked page is a preview)
Vague formulations also shift power from legislators to generally unelected organs of the state (law enforcement, the judiciary, etc.), which attenuates the connection between the will of the public and the effects of laws.