(Roughly) Daily

From Stripe Partners, a framework for rethinking the way we talk about the AI future…

AI is both a new technology and a new type of technology. It is the first technology that learns and that has the potential to outstrip its makers’ capabilities and develop independently.

As Large Language Models bring to life the realities of AI’s potential to operate at unprecedented, ‘human’ levels of sophistication, projections about its future have gained urgency. The dominant framework being applied to identify AI’s potential futures is 165 years old: Charles Darwin’s theory of evolution.

Darwin’s evolutionary framework is rendered most clearly in Dan Hendycks work for the Center for AI Safety which posits a future where natural selection could cause the most influential future AI agents to have selfish tendencies that might see AI’s favour their own agendas over the safety of humankind.

The choice of Natural Selection as a framework makes sense given AI’s emerging status as a quasi-sentient, highly adaptive technology that can learn and grow. The choice is a response to the limitations inherent in existing models for technological adoption which treat technologies as inert tools that only come to life when used by people.

The risk in applying this lens to AI is that it goes too far in assigning independent agency to AI. Estimates on the timing of the emergence of ‘Artificial General Intelligence’ vary, but spending some time with the current crop of Generative AI platforms confirms the view that AI’s with intelligences that are closer to humans are some way off. In the interim using natural selection as a lens to understand AI positions humans as further out of the developmental loop than is actually the case. Competitive forces whether market or military will shape AI’s development, but these will not be the only forces at play and direct interaction with humans will be the principal driver for AI’s progress in the near term.

A year ago we wrote about the opportunity to reframe the impact of AI on organisations through the lens of Actor Network Theory (ANT). More than a singular theory, ANT describes an approach to studying social and technological systems developed by Bruno Latour, Michel Callon, Madeleine Akrich and John Law in the early 1980s. 

ANT posits that the social and natural world is best understood as dynamic networks of humans and nonhuman actors… In our 2023 piece we suggested that ANT, with its focus on framing society and human-technology interactions in terms of dynamic networks where every actor whether human or machine impacts the network, was a useful way of exploring the ways in which AI will impact people, and people will impact AI.

A year on the value of ANT as a framework for exploring AI’s future has become clearer. The critical point when comparing an ANT frame to an evolutionary one is the way in which the ANT framing highlights how AI will progress with and through people’s interactions with it. When viewed as an actor in a network, not a technology in isolation, AI will never be separate from human interventions…

A provocative argument, well worth reading in full: “Why the debate about the future of AI needs less Darwin and more Latour,” from @stripepartners.

Apposite: “Whose risks? Whose benefits?” from Mandy Brown.

* Epictetus

###

As we reframe, we might recall that it was on this date in 1946 that an ancestor of today’s AIs, the ENIAC (Electronic Numerical Integrator And Computer), was first demonstrated in operation.  (It was announced to the public the following day.) The first general-purpose computer (Turing-complete, digital, and capable of being programmed and re-programmed to solve different problems), ENIAC was begun in 1943, as part of the U.S’s war effort (as a classified military project known as “Project PX“); it was conceived and designed by John Mauchly and Presper Eckert of the University of Pennsylvania, where it was built.  The finished machine, composed of 17,468 electronic vacuum tubes, 7,200 crystal diodes, 1,500 relays, 70,000 resistors, 10,000 capacitors and around 5 million hand-soldered joints, weighed more than 27 tons and occupied a 30 x 50 foot room– in its time the largest single electronic apparatus in the world.  ENIAC’s basic clock speed was 100,000 cycles per second (or Hertz). Today’s home computers have clock speeds of 3,500,000,000 cycles per second or more.

source