Benchmarking Human Intelligence versus AI

DeepMind’s Benchmarking of AGI looks like a breakthrough

Tony Czarnecki, Sustensis.

London, 1st December 2021

The release of Google Deep Mind’s paper “Levels of AGI: Operationalizing Progress on the Path to AGI” has quite significantly increased the chances of AGI emerging in the next few years because developers will have to deal with fewer unknowns related to accidentally producing AGI, before global mechanisms of control have been established. The frontier models will be measured and compared against a benchmark once it is globally accepted. It will become a valuable tool for assessing how close AI is to become AGI and what risk the most advanced model may pose to humans. The monitoring of that process should be the key task of the Global AI Safety Institute set up at Bletchley Park Summit in November 2023. However, developers will still have to be prepared for uncomfortable surprises of a sudden increase of near AGI’s capabilities, where none were expected. That is what happened when ChatGPT was released. Read the whole article in PDF here