AI computing power likely to be next battleground in fight against environmental impact … [+]
Artificial intelligence has moved from science fiction fantasy to mainstream reality. AI now powers online tools from search engines to voice assistants, and it’s used in everything from medical imaging analysis to autonomous vehicles. But advances in AI will soon run into another pressing problem: energy consumption.
A bit like cryptocurrencies today, AI risks becoming a target for criticism and regulation due to its large appetite for electricity. Proponents are grouping into camps, with AI optimists touting continued progress with more computing power, while pessimists are beginning to describe AI’s use of energy as wasteful, even dangerous. . Attacks echo those targeting cryptocurrency mining in recent years. There is no doubt that further efforts will be made to stifle AI innovation by reducing its energy supply.
The pessimists raise valid points. Developing ever more efficient AI requires vast computing resources. For example, the amount of computation used to train OpenAI’s ChatGPT-3 would have been equal 800 petaflops of processing power, the equivalent of the 20 most powerful supercomputers in the world combined. Similarly, ChatGPT receives on the order of hundreds of millions of requests every day. Estimates suggest that the electricity needed to meet all these demands could be about 1 GWh per day, or enough to power the daily energy consumption of about 33,000 U.S. homes. Demand is expected to increase further in the future.
In a sense, none of this is new. Almost since the dawn of the Internet, there have been concerns on energy consumption. As the network of servers, data centers and connected devices has grown, so has the demand for electricity. Environmentalists have long pointed out that the continued operation of these energy-intensive facilities and devices, necessary to maintain the 24/7 availability of online services, results in environmental impacts.
Much like cryptocurrencies and the early internet, AI’s energy consumption can seem shocking without proper context. Critics will inevitably characterize the meteoric growth of AI computing as excessive, unnecessary and risky. However, if AI makes workers more productive, for example by saving programmers time writing code, researchers saving time researching and reading articles, and saving ordinary office workers time to engage in repetitive tasks and write documents, AI will invariably save a lot of energy as well.
Additionally, these energy savings do not take into account other direct benefits of AI. Although this is not the only factor, adding calculate enhances AI capabilities, from natural language processing to computer vision. While simply running calculations on problems doesn’t guarantee progress, it reliably brings significant improvements. Computing serves as the fuel for AI innovation.
Unfortunately, just like cryptocurrency mining, nuances are lost as different factions clash in the public arena. These debates can quickly turn emotional. Faced with major concerns related to climate change, restrictions on electricity consumption will likely increase. as a tactic employed by activists to oppose disadvantaged industries. Think “Operation Choke Point” for energy consumption.
We have already seen officials target industries they deem objectionable by putting pressure on the banks and payment networks that provide them with financial services. Sectors such as gun dealers, payday lenders and fossil fuel companies have become targets. Despite any evidence of illegality, these industries face pressure from politicians and activists who seek to change their behavior by cutting them off from traditional financial institutions.
Today, as questions arise as to whether the development of AI should be limited, similar pressure campaigns in the context of AI computing seem inevitable. Just as banks are forced to restrict quid pro quos for political reasons, power companies could soon face similar pressures on their most energy-hungry customers.
Although activists’ fears are not entirely unfounded, there are real risks of AI that need to be considered and addressed – blanket restrictions on electricity consumption are not a good approach. Basing the rights to use power on political favor puts us a slippery slope. Despite everything that divides Americans recently, access to electricity fortunately remains relatively uncontroversial.
Rather than picking winners and losers, policymakers should focus on providing reliable, affordable electricity to all law-abiding, bill-paying customers. Industries like cryptocurrency and AI are a natural fit improve energy efficiency over time, as this is already a priority within these industries. You have to give them the space they need to mature.
We should not fall prey to attempts to stifle the supply of electricity to emerging technologies simply because some industries have gotten caught up in current politics. Through smart, innovation-friendly policy, meeting our energy needs and our environmental goals can go hand in hand.