Chatbots are not AI

---

Issue Areas

---

Public discourse around AI or “Artificial Intelligence,” is usually implicitly about chatbots, specifically consumer facing versions such as ChatGPT and Claude. However, chatbots are only an interface, not AI itself. Chatbots sit within an AI and technological ecosystem where they rely on advances from other subfields, and their breakthroughs in turn feed back into those same areas. As a result, overregulating them risks stifling innovation across the American technology industry more broadly. Such poor policy choices would risk missing out on a chance to create unprecedented wealth and prosperity for Americans. This is not because chatbots will do everyone’s work, but because the innovations that improve their capabilities will also spread across many fields—boosting productivity, lowering prices, and creating entirely new industries and jobs that don’t exist today.

AI is a general purpose technology that can be applied to every industry from healthcare to construction to agriculture. In fact, chatbots are likely not even the space where AI will generate the most value. Many other countries have realized this. China has already developed the world’s first AI hospital. Qatar has sped up building permitting from 30 days to two hours with AI. India has used AI tools to double the income of certain farmers and lift yields by 21 percent. America should not be left behind.

By focusing only on chatbots when we think about how to regulate AI, we miss its most important roles as a general purpose technology such as boosting productivity, lowering costs, and enabling entirely new kinds of work.

It is key to understand that AI technology has existed long before OpenAI released ChatGPT in 2023. It goes back nearly a century to Alan Turing’s computing work in the 1940s, with the terms “artificial intelligence” being coined as early as the Dartmouth Conference in 1955. Many of the core ideas behind today’s AI systems, including the backpropagation methods used to train neural networks, were developed as early as the 1970s and 1980s. These ideas sat dormant for decades, limited by data and compute, until cheap GPUs and the explosion of internet scale data unlocked them in the 2010s. What followed was a wave of breakthroughs across the field.

DeepMind, the AI lab acquired by Google in 2014, is a clear example. Its mission was, and still is, to develop AGI (“Artificial General Intelligence”) as a tool for scientific inquiry. Their approach began by applying AI to games—most famously the strategy board game Go. In 2016, long before ChatGPT’s breakout moment, DeepMind trained a model, AlphaGo, which beat the world champion by playing millions of games against itself and finding techniques that were so advanced that the best human players could not understand its strategy. The next phase of DeepMind’s work applied the techniques they learned from training AlphaGo to advance Demis’ mission to help humans understand fundamental questions in science, starting with biology.

AlphaFold, the project that won two of its researchers a Nobel Prize, is AI. But no words go into it. It is not a chatbot. AlphaFold works by predicting which amino acids go together to determine possible protein structures, a technique which has valuable applications in medicine. This innovation was so important because previously it was incredibly expensive to go through the process to predict the structure of a protein. Because of AlphaFold, new medicines can be made much more quickly, and the science that follows may one day allow for personalized medicine that can treat any individual for any disease.

AlphaFold is one of the many examples of AI applied to improving human health and flourishing. Another subset of AI, computer vision, is useful for analyzing images and is already making a difference in healthcare. For instance, AI can now better identify cancer in CT scans—in some instances three years earlier—than human doctors by picking up on slight patterns that humans miss. Creating such a model uses many of the same techniques and algorithms required to make large language models, but clearly, such systems are not chatbots.

Outside of medicine, non-chatbot AI will have massive deflationary effects on prices, driven by increased productivity rather than job loss, if the technology is allowed to develop. One clear example is robotics. Right now, in factories across the world, robotic arms assemble components to make products from cars to iPhones to prefabricated homes. One of the major limitations of robotic arms, however, is that they historically have operated deterministically, meaning that there was little room to adjust to real world conditions on the fly. While companies have made some progress on this front, robotics by and large is limited to highly controlled environments like factories.

But what happens when robots can operate with high variation across a wide range of conditions? We are not fully there yet, but many companies see this future coming, driven by the same breakthroughs in algorithms and learning methods that have powered modern chatbots and coding models. At scale, AI will enable robots to take on the dangerous manual labor that society depends on.

In agriculture, automation could dramatically decrease the global cost of food. In housing, it could enable continuous construction using methods too dangerous for human workers. The result is a textbook case of Jevons paradox: as costs fall, demand and usage expand. Lower costs unlock new development projects, new types of food production, and entirely new technologies, driving wealth creation, economic growth, and jobs that we don’t even know exist today.

AI will also drive efficiency in the systems that quietly move the economy, like logistics. If you drive from Austin to Atlanta, Apple or Google Maps gives you a near-optimal route using heuristics, since computing the truly optimal route is often too expensive to do at scale. The problem gets much harder with the vehicle routing problem, like a UPS van making 120 stops, where the whole sequence has to be optimized together. Traditional operations research handles small cases, but as routes grow more complex, those methods no longer suffice. AI excels here, learning from operational data to weigh traffic, weather, closures, and driver behavior, then adapting in real time. The payoffs are already large: Uber Freight has used machine learning to cut empty truck miles from ~30 percent (the U.S. average) to 10–15 percent. Because freight movement and last mile delivery (accounting for roughly 40 percent of logistics costs) are both heavily route-dependent, improvements like these stack across the supply chain and ultimately lower prices for consumers. Cheaper transportation makes previously unprofitable production viable, bringing more factories online and expanding economic activity. Chatbots and large language models are just one interface within the broader field of artificial intelligence, and not even the one likely to create the most value. They emerge from an ecosystem where a breakthrough in one area becomes an unlock in another. The methods behind AlphaGo led to AlphaFold; the training techniques behind chatbots are now teaching robots to handle the physical world. This flow runs in both directions. Regulating “AI” as though it were only chatbots misjudges the technology based on its most visible form, and risks slowing down the medical, industrial, and logistical breakthroughs that depend on the same underlying research. Right now much of this innovation is happening in America, and if we want to stay the dominant global power in the 21st century we must make sure that our economy can take advantage of the increased productivity, lower costs, and entirely new industries that AI technology promises.