Artificial Intelligence and the Past, Present, and Future of Democracy

Langdon Winner’s classic essay ‘Do Artifacts Have Politics?’ resists a widespread but naïve view of the role of technology in human life: that technology is neutral, and all depends on use. He does so without enlisting an overbearing determinism that makes technology the sole engine of change. Instead, Winner distinguishes two ways for artefacts to have ‘political qualities’. First, devices or systems might be means for establishing patterns of power or authority, but the design is flexible: such patterns can turn out one way or another. An example is traffic infrastructure, which can assist many people but also keep parts of the population in subordination, say, if they cannot reach suitable workplaces. Secondly, devices or systems are strongly, perhaps unavoidably, tied to certain patterns of power. Winner’s example is atomic energy, which requires industrial, scientific, and military elites to provide and protect energy sources. Artificial Intelligence (AI), I argue, is political the way traffic infrastructure is: It can greatly strengthen democracy, but only with the right efforts. Understanding ‘the politics of AI’ is crucial since Xi Jinping’s China loudly champions one-party rule as a better fit for our digital century. AI is a key component in the contest between authoritarian and democratic rule. Unlike conventional programs, AI algorithms learn by themselves. Programmers provide data, which a set of methods, known as machine learning, analyze for trends and inferences. Owing to their sophistication and sweeping applications, these technologies are poised to dramatically alter our world. Specialized AI is already broadly deployed. At the high end, one may think of AI mastering Chess or Go. More commonly we encounter it in smartphones (Siri, Google Translate, curated newsfeeds), home devices (Alexa, Google Home, Nest), personalized customer services, or GPS systems. Specialized AI is used by law enforcement, the military, in browser searching, advertising and entertainment (e.g., recommender systems), medical diagnostics, logistics, finance (from assessing credit to flagging transactions), in speech recognition producing transcripts, trade bots using market data for predictions, but also in music creations and article drafting (e.g., GPT-3’s text generator writing posts or code). Governments track people using AI in facial, voice, or gait recognition. Smart cities analyze traffic data in real time or design services. COVID-19 accelerated use of AI in drug discovery. Natural language