Title: Brakes Make Trains Go Faster (and Other Lessons from Safety for Advancing Capabilities)

Making advanced technologies, such as artificial intelligence or autonomous platforms, safe for effective and ethical use is an urgent challenge. Rather than thinking of safety as a barrier to overcome, understanding the limits of when a tool is useful and will perform as expected creates the conditions by which it can be used. History holds many tales of technology not reaching its potential because operators or organizations were unable to trust it. Investing in safety can enhance trust and thus drive usage; for example, the development of coordinated air brakes for trains in the late 1860s led to increased speeds, better service, and a wider base of passengers. But what is safety for AI systems? Traditional safety analysis relates the conditions of a use case to the behavior of a system, but AI methods apply when we do not have full behavioral specifications to work from, relying on implied and not explicit specifications. Safety requires assurance arguments, positive claims that the system will function as intended in a use case while avoiding defined undesired behaviors.

Join via Zoom