Forecasting TAI with biological anchors: Part 4: Timelines estimates and responses to objections

Abstract

This report presents a framework for predicting the approximate timeline for transformative AI (TAI), defined as the advent of an AI capable of solving a wide range of important tasks across different domains at a level comparable to or exceeding human performance, resulting in major social and economic changes. It uses a quantitative model based on the concept of biological anchors, particularly the evolutionary hypothetical idea that the total number of floating-point operations (FLOPs) a biological organism can perform across its lifespan is constrained by the amount of energy it can consume and the efficiency of its neural architecture. By estimating hardware trends and algorithmic progress, the model extrapolates the amount of computation required to train transformative AI with similar amounts of FLOPs per second to biological organisms. This allows us to predict the approximate time when the computational cost of training TAI becomes technologically feasible. The report also addresses concerns and limitations of the framework, including data and environment bottlenecks, algorithmic breakthroughs, and the possibility of alternative paths to TAI. – AI-generated abstract.