Canberra rolled out the red carpet this week to one of the AI overlords whose technology is driving the world down the path of creative destruction. Anthropic’s CEO Dario Amodei, the putative “good” tech oligarch, was spinning his version of a machine-driven future with the elan of a man who has untangled the mysteries of the universe – or at least built a predictive text model that can scrape the output of humanity and spit out compelling summaries of our collective consciousness.

He regaled the prime minister, assorted elected officials and the tech sector’s glitterati with his pitch for good AI that would transform the economy, before becoming the first to sign up to the government’s new datacentre principles, conveniently released just a week earlier. It was compelling shill and, to be fair, Amodei is not the worst of the gods. He created Anthropic after leaving Open AI when the company dispensed with its not-for-profit, “safety first” mission. He regularly shares thoughtful essays on the path of technology and has been open about his fears for the impact of his own products. He broke with the Trump administration over the limits to how his technology would be used to spy on citizens and enable autonomous weapons, turning himself into an enemy of the state.

But as Toby Walsh, a scientia professor of artificial intelligence at the University of New South Wales, reminds me: there is no “good AI” because AI is both good and bad. It can unlock new connections and knowledge by synthesising huge amounts of information, but it depends on the extraction of huge amounts of energy to create tools that replace human workers with machines. Like other models, Anthropic has been trained on the stolen work of creators. Indeed the company settled a $US1.5bn claim by authors in the US. Amodei has blithely predicted his technology will see up to half of all white-collar entry level jobs destroyed, yet carries on regardless.

It’s been instructive to watch the reception Amodei has received from our government, which has proclaimed itself to be all in on AI and its siren song of a fast track to productivity, albeit based on industry-sponsored modelling. This despite the concerns of its core union constituents, the righteous outrage of artists and the well founded concerns of parents. As the government wrestles with these contradictions, the idea of a “progressive AI” appears manna from heaven. The memorandum of understanding struck with Anthropic ticks all the right boxes from “tracking frontier AI progress and........

© The Guardian