Like Covid, we ignore AI exponentials at our peril
The Notebook: Where the City’s movers and shakers have their say. Today, it’s Marc Warner, chief executive of AI firm Faculty
It’s February 2020. A little known disease called coronavirus is spreading exponentially. It seemed early to act – but before long, the NHS risked being overwhelmed and lockdown was unavoidable. It’s almost a cliche to say humans are bad at processing exponentials, but it’s true. I worry we’re similarly underestimating the exponential “rising tide” of AI.
I’m thinking about this because an organisation called METR recently published a paper, Measuring AI Ability to Complete Long Tasks. It’s one of the clearest demonstrations of exponential AI progress yet. Their approach was simple: test AI on how long it takes to complete tasks compared to humans.
The result? AI capabilities have doubled every seven months since 2019. GPT-2 could handle two-second tasks, GPT-3 could do nine-second ones, and current models handle tasks that take humans two hours. If this trend continues, AI could autonomously tackle month-long projects within five to seven years.
........© City A.M.
