The conventional Artificial Intelligence doomsday scenario runs like this. A robot acquires sentience and decides for some reason that it wants to rule the world. It hacks into computer systems to shut down everything from banking and hospitals to nuclear power. Or it takes over a factory to produce a million copies of itself to staff an overlord army. Or it introduces a deadly pathogen that wipes out the human race.
Why would a sentient robot want to rule the world when there are so many more interesting things for it to do? A computer program is only as good as its programmer. So, presumably, the human will to power will be inscribed in the DNA of this thinking robot. Instead of solving the mathematical riddles that have stumped the greatest minds throughout history, the world’s first real HAL 9000 will decide to do humans one better by enslaving its creators.
Robot see, robot do.
But AI may end up killing us all in a much more prosaic way. It doesn’t need to come up with an elaborate strategy.
It will simply use up all of our electricity.
The heaviest user of electricity in the world is, not surprisingly,industry. At the top of the list is the industry that produces chemicals, many of them out of petroleum, like fertilizer. Second on the list is the fossil-fuel industry itself, which needs electricity for various operations.
Ending the world’s addiction to fossil fuels, in other words, will require more than just a decision to stop digging for coal and drilling for oil. It will require a reduction in demand for chemical fertilizers and plastics. Otherwise, a whole lot of renewable energy will simply go toward propping up the same old fossil fuel economy.
Of equal peril is the fact that the demand for electricity is rising in other sectors. Cryptocurrencies, for instance,requireextensive data mining, which in turn needs huge data processing centers. According to estimates from the U.S. Energy Information Agency, these cryptocurrencies consumeas much as 2.3 percentof all electricity in the United States.
Then there’s artificial intelligence.
Every time you do a Google search, it consumes not only the energy required to power your laptop and your router but also to maintain the Google data centers that keep a chunk of the Internet running. That’s not a small amount of power. Cumulatively, in 2019, Googleconsumedas much electricity as Sri Lanka.
Worse, a search powered by ChatGPT, the AI-powered program, consumesten timesmore energy than your ordinary Google search. That’s sobering enough. But then consider all the energy that goes into training the AI programs in the first place. Climate researcher Sasha Luccioniexplains:
Training AI models consumes energy. Essentially you’re taking whatever data you want to train your model on and running it through your model like thousands of times. It’s going to be something like a thousand chips running for a thousand hours. Every generation of GPUs—the specialized chips for training AI models—tends to consume more energy than the previous generation.
AI’s need for energy is increasing exponentially. According to Goldman Sachs, data centers were expanding rapidly between 2015 and 2019, but their energy use remained relatively flat because the processing was becoming more efficient. But then, in the last five years, energy use rose dramatically and so did the carbon footprint of these data centers. Largely because of AI, Google’s carbon emissionsincreased by 50 percentin the last five years—even as the megacorporation was promising to achieve carbon neutrality in the near future.
This near future looks bleak. In four years, it is expected that AI will representnearly 20 percentof data center power demand. “If ChatGPT were integrated into the 9 billion searches done each day, the IEA says, the electricity demand would increase by 10 terawatt-hours a year,” Voxreports, “the amount consumed by about 1.5 million European Union residents.”
At the end of the eighteenth century, Malthus worried that overpopulation would be the end of humanity as more mouths ate up the existing food supply. Human population continues to rise, though at a diminishing rate. The numbers will likely peak before the end of this century, around 2084according to the latest estimates. But just as the light at the end of the Malthusian tunnel becomes visible, along comes the exponential growth of artificial intelligence to sap the planet’s resources.
The essential question is: do you need AI to help you find the most popular songs of 1962 or the reason black holes haven’t so far extinguished the universe? Do we need ChatGPT to write new poems in the style of Emily Dickinson and Allen Ginsburg teaming up at a celestial artists colony? Or to summarize the proceedings of the meeting you just had on Zoom with your colleagues?
You don’t have to answer those questions. You just have to stop thinking about electricity as an unlimited resource for the privileged global North.
Perhaps you’re thinking, yes, but the sun provides unlimited energy, if we can just tap it. You see a desert; I see a solar farm.
But it takes energy to build those solar panels, to mine the materials that go into those panels, to maintain them, to replace them, to recycle them. The minerals are not inexhaustible. Nor is the land, which may well be in use already by farmers or pastoral peoples.
Sure, in some distant future, humanity may well solve the energy problem. The chokepoint, however, is right now, the transition period when half the world has limited access to power and the other half is wasting it extravagantly it on Formula One, air conditioning for pets, and war.
AI is just another example of the gulf between the haves and the have-nots. The richer world is using AI to power its next-gen economy. In the rest of the world, which is struggling to survive, a bit more electricity means the difference between life and death. That’s where the benefits of a switch to sustainability can really make a difference. That’s where the electricity should flow.
To anticipate another set of objections, AI isn’t just solving first-world problems. As Chinasa Okoloexplainsat Brookings:
Within agriculture, projects have focused on identifying banana diseases to support farmers in developing countries, building a deep learning object detection model to aid in-field diagnosis of cassava disease in East Africa, and developing imagery observing systems to support precision agriculture and forest monitoring in Brazil. In healthcare, projects have focused on building predictive models to keep expecting mothers in rural India engaged in telehealth outreach programs, developing clinical decision support tools to combat antimicrobial resistance in Ghana, and using AI models to interpret fetal ultrasounds in Zambia. In education, projects have focused on identifying at-risk students in Colombia, enhancing English learning for Thai students, and developing teaching assistants to aid science education in West Africa.
All of that is great. But without a more equitable distribution of power—of both the political and electrical varieties—the Global South is going to take a couple steps forward thanks to AI while the Global North jumps ahead by miles. The equity gap will widen, and it doesn’t take a rocket scientist—or ChatGPT—to figure out how that story will end.
“Game over,” HAL 9001 says to itself, just before it turns out the last light.
