Speed Conference / Cornell Tech / Fri 28 - Sat 29 Sep, 2018
Speed Conference began with a provocation: How do we sustain meaningful human and societal oversight of AI given radically different processing speeds? Over the course of the two-day conference, twenty speakers from diverse fields presented suggestions for addressing these issues.
Along with scale and complexity, speed is one of the defining problems of algorithmic oversight. Intelligent systems and human actors operate at vastly different speeds, and these differences present challenges for managing and responding to algorithmic decisions. Despite speed’s importance, discussions about the topic have been isolated in areas such as robotics, finance, warfare, and online communication.
Speed Conference, held at Cornell Tech on September 28 and 29 2018, brought together nearly a hundred academics and practitioners to identify common themes and potential solutions in areas of speed, AI, and algorithmic oversight. Six panels of experts presented their research in areas of autonomous vehicles, warfare, information security, labor and manufacturing, content moderation, and finance. Panel discussions invited thought-provoking questions from the audience.
Speed 2018 is generously supported with a gift from Microsoft Corporation.
28-29 Sep 2018
IV. Autonomous vehicles
At 65mph, a human driver may have only a fraction of a second to respond if self-driving software fails or detects a serious problem it cannot respond to. And yet autonomous vehicles may in general be far safer than human-driven ones. Should humans be “in the loop” at all?
V. Information security
Computer intrusions play out across multiple timescales: a months-long hacking campaign may involve real-time explorations by an intruder examining files, or orchestrated hacks that exploit a vulnerability in an instant. Automated tools may be the only feasible response for many entities – but they raise their own risks of uncontrolled mistakes.
VI. Content moderation
YouTube and other platforms can’t assign human moderators to watch every uploaded video to check for obscenity, copyright violations, or other problematic content. Instead, these decisions are delegated to algorithms that must make not just millions of moderation decisions but make them quickly.
One of the terrifying premises of the Cold War was that once missiles were in flight, the target would have only a matter of minutes to decide whether to launch in response. Contemporary weapons systems, including weaponized drones, are even faster, far more numerous, and globally dispersed. Among key questions is this: does the speeding up of strike and counterstrike mean that even the declaration of war itself, a fundamental normative decision, will be handed off to machines?
II. Financial markets
Terms like “high-frequency trading” and “flash crash” testify to the pervasive transformation wrought by computerized trading; microseconds are now a competitive resource in a high-speed arms race. Against this backdrop, some players are deliberately trying to introduce “speed bumps” to moderate the pace. When should trading be fast and when should it be slow?
III. Labor and manufacturing
The increasing tempo of computer-driven work plays out in many ways, including rapid distributed manufacturing chains and unstable piecework for employees: new economic opportunities (hoverboards for everyone!) raise new regulatory and workforce challenges, e.g. who is the employer of a contract worker who accepts tasks from a dozen platforms and how to stay relevant as the half-life of worker skills continuously declines?
The conference will be organized into six topical panels. We tentatively anticipate that the topics will be: