The future of automation
What activities remain un-automatable? Of course, there are tasks that require human empathy, trust, accountability, or authority, for example aspects of healthcare and of government. There will always be a market for the hand-made, the artisanal, and the bespoke, as long as potential buyers are not themselves impoverished by automation.
Then there are tasks that require highly general and varied abilities, comparable to humans in ability and generality, and these are farther in the future. Mental tasks with this property are called “AI-complete”. For example, carrying on an extended, human-like conversation, as in the Turing test (4). We could coin a term such as “android-complete” for physical tasks with the same property. An example I like, perhaps a counter-intuitive one, is the work of a jobbing carpenter. A robot capable of the necessary range of dexterous action, use and choice of tools, flexible work in difficult spaces, use of multiple senses, and choices of action depending on experience and physical intuition – I think such a robot would be capable of working in many different areas also.
Estimated timelines for general AI vary, but have shortened recently, with median estimates probably in the 2030s–2080s. When this happens, we’ll see automated businesses, automated science – the sky is the limit. We won’t be able to ban automation but we have to protect human interests. We have a lot of thinking still to do.
(1) Acemoglu and Restrepo, “Artificial Intelligence, Automation, and Work”, 2019
(2) Nakamura and Csikszentmihalyi, “The Concept of Flow”, 2014
(3) Whitelaw, “Metacreation: Art and Artificial Life”, 2004
(4) Turing, “Computing Machinery and Intelligence”, 1950