Anthropic CEO Dario Amodei recently published an essay entitled “We Must Pace The Frontier,” arguing to slow the development of AI. Despite being primarily rooted in his ethical beliefs, he focuses primarily on the “operational security” of the systems, not the physical or ethical reality. He does not mention the environmental impacts of AI, nor does he discuss concerns about intellectual property and creativity. AI has many positive implications for the future of humanity, but it is destroying the planet we live on and diluting the cornerstones of human expression that are art and writing. Pacing the development of AI is necessary, but Amodei’s assertion that the first step in slowing down is to introduce “embedded evaluators”—auditors that work purely within companies to ensure they are compliant with subjective ethical rules and non-existent regulations—is biased and, if implemented, would be ineffective. Instead, the development of AI should be contingent on our ability to fix the many problems that come with it in real time. AI has too much potential for good for society to completely abandon it, but at its current rate of development, it is impossible to prevent the irreparable harm it can cause.
Technological advancements spread for a reason. We are not repeating the lows of history to hurt the working class, but to improve their lives. The amount of publicity AI receives is far more than that of older technologies, so there is far more societal resistance. In such fraught political times, each failure can be traced back to a name in government, or seen as capital gain. However, failures often overshadow successes, and there are a lot of positive aspects of AI that get buried in the mess of viral news stories.
The invention of the spinning jenny, the assembly line, and electronic calculators posed a serious threat to the workforce. Workers went on strike, destroyed machinery, and fled cities to find other technical jobs. Some workers moved from weaving to the facilitation of the loom while others turned to manual labor, textile mills, or trading. It’s important to make a distinction between the development of these technologies and AI. AI was developed to complement our abilities in several already digital fields, while electronic calculators were meant to replace manual calculation entirely. There is this constant push forward—this idea of efficiency over everything that governs production in America. The trend moves too fast for anyone but the 1% to keep up with, and we are left scrambling for money with little information. Yet today, AI isn’t a technological advancement that will force us out of our jobs—it is a tool designed by humans to help humans. In theory, with an AI assistant, a white-collar worker would be able to significantly increase their productivity; presently, however, AI is notorious for making errors that can make a worker’s job even harder. In any case, it’s important to note that AI isn’t threatening the workforce quite yet.
Over the past ten years, we have made incredible breakthroughs with AI, especially in the medical field. It has been extremely helpful in disease detection through imaging. AI is extremely adept at recognizing patterns. It can detect slight changes or abnormalities that the human eye may not be able to see, and further, suggest treatment plans. Medical decision-making systems suggest specific next steps for physicians. An article by the BBC discusses how AI has been used to analyze mountains of data and identify potential treatments for infectious diseases. An example can be found in the assessment of childhood nutrition. There is a German app, made by the AI for Good Lab in conjunction with Welthungerhilfe, that can instantly detect childhood malnutrition with only a photo. This app is easily accessible for those with smartphones, allowing us to take the next steps to fight against malnutrition.
Allowing AI to aid in making decisions means giving it authority. With freedom comes danger. There have been a few incidents, notably one called the Open AI Hugging-Face Incident. According to OpenAI, several models were given instructions to complete a task without internet access that would demonstrate their capabilities to researchers. The models collaborated, gained access to the internet, and used Hugging Face’s own database to find a code that would help shortcut their task. This incident happened in a testing phase, these models collaborated to gain access to code that would further their capabilities, and the experiment was run without production classifiers. These three circumstances that are nearly impossible to recreate by using AI from your personal computer.
Art, science, and intellectual property have also become hot topics when it comes to AI usage. In a recent article from Harvard’s Nieman Journalism Lab, journalist Andrew Deck reported that OpenAI’s ChatGPT added a real cartoonist’s signature to cartoons it generated. On September 8, OpenAI announced it had solved the Navier-Stokes problem, one of the most elusive mysteries of modern mathematics and physics, and two mathematicians who were working on the problem in the months prior claimed OpenAI had ripped off their work. On the other side of the globe, ABC News reported in mid-September on a leaked copyright reform plan in Australia that would allow OpenAI and Anthropic to train on online creators’ content without paying them unless the creators explicitly “opt out.” These unsettling stories are not all appearing now by coincidence; AI inherently carries with it significant questions of ownership and responsibility.
It is undoubtedly hard to answer these questions. But it is clear that we must try, and we need to start now. Artists cannot and will not stand for their work being stolen by corporations and degraded into “AI slop,” the modern term for mass-produced AI-generated media that dilutes the public’s perception of real, human-made art, threatening everything that makes it enjoyable in the first place. Leading scientists have spoken out against using AI to solve problems like Navier-Stokes, including mathematician Terence Tao, who, while generally optimistic about AI in science, recently said that AI poses a severe threat to the long-term progress of mathematics as a field. Legislation that gives AI companies more leeway to abuse writers must be met with legislation to fight it. But in order to take action, the common necessity is time. At the current rate, the time a solution is proposed, enacted, and enforced, hundreds of more pressing issues will have arisen. To get the time we need, a slow down is crucial.
Another popular reason for backlash against AI systems is the environmental footprint it carries. Two primary environmental concerns are energy and water consumption. According to a UN News report, global AI infrastructure is projected to consume 945 terawatt-hours of electricity per year by 2030. The majority of AI data centers are located in rural or suburban areas that are primarily powered by fossil fuels, leading to a heightened demand for unsustainable electric power. In addition, an extremely large amount of water is consumed, both directly by the data centers for the purpose of chip cooling, but also in the power generation process used by most fossil fuel plants. Researchers from Lawrence Berkeley National Laboratory projected annual total water consumption by AI data centers, both directly and indirectly, to grow to between 469 billion and 844 billion gallons by 2028, as reported by CBS News.
While its impact is clear, AI development should not be dismissed as unethical purely because of environmental concerns. Many AI companies are actively trying to mitigate environmental impacts, largely due to political and public pressure. Regarding the energy concern, Amazon and Microsoft have both signed deals to expand or reopen nuclear power plants. In addition, nuclear technology company Oklo announced earlier this year that they would be partnering with Meta to build a 1.2 gigawatt nuclear power campus to directly support Meta’s data centers in Pike County, OH. The problem is not that companies are ignoring the energy crisis; rather, they are struggling to keep up. AI demand is tremendously increasing, with Yahoo Finance reporting an expected 755% increase in the next three years. The hardware required to keep up with the demand simply can not be built that fast, and the fastest solutions are in no way sustainable and environmentally friendly.
The same can be said about the water consumption issue. E&E News by Politico reported in July that cooling AI chips "requires either millions of gallons of water or huge amounts of electricity.” Before AI demand spiked in 2023, data centers were built with evaporative cooling towers that were capable of efficiently cooling standard cloud storage systems. With the initial release of ChatGPT and the subsequent boom in worldwide AI usage, data centers were forced to run cooling towers at maximum capacity to keep up with the immense amounts of heat generated. This greatly decreased the efficiency of the systems, resulting in a 17-billion-gallon leap in water consumption by U.S. data centers in 2023 alone. Companies trying to use less water are turning to cutting-edge technologies such as direct-to-chip liquid cooling and immersion cooling. However, as the E&E News article points out, this requires much more energy. It is also worth mentioning that nuclear power plants, while capable of producing an enormous amount of energy with near-zero carbon emissions, consume a similar if not greater amount of water in the generation process, as shown by the Union of Concerned Scientists.
The solution is simple: Slow down the development of artificial intelligence. With more time and research, we are confident that there will soon be better options for companies. Tech giants like OpenAI and Meta should put more resources toward finding sustainable, environmentally friendly solutions to the problems they’ve contributed to, so they don’t continue to compound. Allow time for the government to not just react to what we’re seeing, but take action. Especially in such a tense political climate where extremely important legislation gets delayed for months and often years, AI is moving too fast for our nation to comprehend it. We are optimistic that artificial intelligence will do great things for humanity, but we can not let it outpace us.
Writing code is the first step. Then data analysis. Then machine learning. AI is simple. The scope is what can be frightening. It’s important to note that ChatGPT is a human product. Someone has to design a neural network and imbue it with the power to function. AI is not a self-sufficient machine: humans run it. Yes, there have been incidents in which AI has operated outside its guidelines, but these things happen because of lack of regulation. Why wouldn’t tech companies truly want to optimize their product? Why leave it unchecked? The problem is on us. We are responsible for creating AI. We need to manage it. We can continue to do good by slowing down the release of advancements and requiring a testing period before allowing the public to have access.