Fruitflies, railroad mania, and artifical intelligence.
Technology is moving so swiftly that everyone, from ordinary technology professionals to bystanders, struggles to understand the advances and what they mean to society. What makes artificial intelligence the most challenging technology to comprehend is that it can do fantastic things while at the same time struggling with basic cognition and judgment. If you listen to technology executives, they say we are on the cusp of a revolution. It is a revolution requiring more money, energy, and data to generate Large Language Models that can mimic the human mind. As an AI practitioner bridging the gap between pixels and data, I encourage a critical eye toward all AI claims. Together, we need to understand this technology's promise and limits.
Astronaut Dan Goldin is on LinkedIn, and I follow his messages regularly. He has the drive and can-do attitude that only an astronaut with the 'right stuff' could possess. He is also a thoughtful business leader, pointing out that we can learn much about Artificial Intelligence from fruit flies. The brain of a fruit fly is tiny. Still, it does a remarkable job juggling numerous tasks in a complex environment, including flight, navigation, operating the bodily organs, identifying threats, and finding mates. It does all these complex activities, consuming about a milliwatt of energy. In perspective, many of the lightbulbs in common flashlights are one and a half watts, 1500 times stronger than a milliwat. Fruit flies don't need much energy to power their brains.
Move up the evolutionary scale. Our brains account for over 20% of our daily calorie expenditure, using approximately 20 watts of energy. A typical data center for AI users is between 50 and 70 million watts of energy daily, and it does not come close to what a human brain can accomplish. With five billion years of experience, nature is better at energy consumption than human engineers, which means biology trumps technology.
Science and technology have only had fifty years to work on Artificial Intelligence and Large Language Models, so you can excuse our inability to mimic the energy consumption of living things. We are still new to the effort. Back in May, I pointed out that Open AI CEO Sam Altman said the table stakes for technology have increased. I knew he was not bluffing because building artificial intelligence models requires lots of energy to power the data centers. It requires equal energy to cool these data centers, which costs money and strains local infrastructure.
AI also requires specialized network and artificial intelligence professionals to build and monitor those systems. The amount of time, money, energy, and data is staggering. Someone has to pay for this investment, and those people are large technology companies with a vested interest in getting a return on investment. Large technology companies also count on their customers to defray those costs. Microsoft is achieving this with its Microsoft Azure tools. It is a sound strategy, but the return on investment is elusive.
The May issue of CIO magazine points out that chief information officers struggle to find a use case for Artificial Intelligence. It is not that Artifical Intelligence does not have uses in Fortune 500 companies; the challenge is finding economical uses for Artifical Intelligence. The coding and training of large language models are expensive, and executives are starting to experience sticker shock.
History does not repeat but often rhymes, and I suspect that this enthusiasm for AI resembles the railroad-building craze in Great Britain in the 19th century. The most expensive part of running a railroad is building a track from destination to destination. Railroad companies issued stocks and bonds as IOUs to pay for this construction. Soon, investment and speculation in railroad companies were rampant until the bubble burst, and millions of pounds of wealth evaporated overnight. The risky investments ruined countless investors and ordinary people caught in the frenzy. The 'railway mania' and subsequent collapse built out most of the British rail infrastructure and gave the United Kingdom a head start during the Industrial Revolution. The dot-com bubble's giddy and stupid days provide a parallel tale. Companies helped build out the backbone of the commercial internet we know today, only to fail and be absorbed by other companies at pennies on the dollar. Without Pets.com and MySpace, we would not have Facebook and Amazon today.
So, Artificial Intelligence is in its railway mania phase. It is expensive and consumes gluttonous amounts of energy and money. It is also less intelligent and efficient than a fruit fly. I suspect we will achieve better efficiency and return on investment, but how much money do we need to burn to achieve that goal? I do not know, nor do the big technology companies, which is why all of us should be skeptical.
Until next time.
Adam Clark Estes
Comments ()