Before the Roman Empire and the Greek philosophers, before records of Chinese dynasties and Hindu kings, there were algorithms.1 Babylonian mathematicians discovered that a series of well-organized and repeatable steps—an algorithm—could be devised to accomplish a particular task, beginning with some input and concluding with some output. If the input and the steps were the same each time, the output would also be the same. This idea was essential to order and structure in mathematics and all that was built upon that foundation. Two plus two will always equal four.
Profound as this insight was on its own, the crux of algorithmic thinking was that different sets of steps could be devised for different tasks, yielding different results even with the same inputs; the answer to the calculation two
minus two will always be zero. The Babylonians crafted one process for multiplying numbers and an inverse one for dividing them.2 Greek thinkers used more complex algorithms in their quest to find prime numbers. Islamic scholars in the ninth century developed innovative algorithms as they discovered algebra. In the 1800s, Charles Babbage and Ada Lovelace began to imagine what algorithms could do as part of general-purpose computing machines, which had not yet been invented. Lovelace in particular recognized that the numbers in algorithms didn’t have to represent quantities but could represent more abstract concepts; images or sounds, for example, could be converted to a numerical form and then given to a machine.3
The AI advances described in the previous chapter continued this history, first with expert systems and then with machine learning. While AlexNet and other data-centric supervised learning systems attracted interest in machine learning in 2012, and while GANs broadened the conception of what the technology could do, it was the subsequent algorithmic innovation described in this chapter that offered a second vital spark for the new fire. With ever more powerful and more efficient algorithms, AI could accomplish greater feats.4
Most significantly, algorithmic breakthroughs cemented the salience of AI to geopolitics. In changing what the technology could do, these algorithms also changed who cared about it. AI had once primarily been the sphere of technical engineers striving to maximize the performance of their systems. Now, it became the domain both of evangelists, who applied it to notable scientific problems, and of warriors within governments, who aimed to gain a strategic edge over rival nations. The gap between these two worldviews began to widen as it became more obvious just how powerful AI’s algorithms would be.
CHAPTER 2 1. This comparison does not include the Xia dynasty, a part of traditional Chinese historiography but one for which there are no contemporaneous records.
2. In essence, these new operations were new one-step algorithms.
3. L. F. Menabrea and Ada Augusta Lovelace,
Sketch of the Analytical Engine Invented by Charles Babbage,
Esq, vol. 3 (1842; repr., London: Scientific Memoirs, 2014).
4. Danny Hernandez and Tom B. Brown, “Measuring the Algorithmic Efficiency of Neural Networks,” arXiv, May 8, 2020, http://arxiv.org/abs/2005.04305.
Copyright © 2024 by Ben Buchanan and Andrew Imbrie. All rights reserved. No part of this excerpt may be reproduced or reprinted without permission in writing from the publisher.
