On Regulating the Unknown: The 1865 Locomotives Act and AI

In 1865, the British Parliament passed the Locomotives Act. It required that any 'self-propelled vehicle' be manned by three people, one of which had to precede the vehicle on foot and carry a red flag to warn horse-drawn carriages. The vehicle was limited to a maximum speed of 4 miles per hour in the countryside, and 2 miles per hour in towns. The law was designed to manage a new technology that was poorly understood and feared by the horse-drawn carriage industry and the public. Parliament responded with this detailed Act, that remained in force until 1896, to manage something which it did not understand.

By then France and Germany had developed motor industries, and Britain, constrained by this legislation, had fallen behind in an industry it helped pioneer. When Parliament finally amended the Act, raising the speed limits and abolishing the red flag requirement, it was too late to recover lost ground.

The EU AI Act followed a similar pattern. Formally proposed in April 2021, it took four years to draft at a time when the dominant AI model was narrow and task-specific. The EU's risk-based classification was built around this understanding. Between the proposal and its approval, the AI landscape changed entirely; GPT-4 was released and general purpose AI systems (GPAI) became common. Technologies that did not exist when negotiations began, had to be accommodated in a final text that was not designed for them, with the GPAI provisions sitting uneasily next to the original architecture of the Act.

The concerns that have motivated the EU are real, but AI is a technology that is highly reflexive and adaptable, and its governance needs to match those qualities. Indeed, in 3 to 4 years we might not be talking about AI governance at all because AI will simply be present in everything, just like we don't talk of electricity policy, because electricity is a given in industrial, infrastructural and economic policy. Legislation specific enough to describe what AI is today, might not survive what the technology becomes tomorrow.