On its own, “artificial intelligence” is a vague and indecisive term. Despite its many applications, AI is defined simply as “the ability of a digital computer… to perform tasks commonly associated with intelligent beings.” By this definition, artificial intelligence is not even considered a proper noun; it is instead a criterion, a checkmark on a sheet of product function. We consider AI systems such as ChatGPT and Claude revolutionary for their ability to mimic the human ability to synthesize research and mimic our use of language. Yet AI’s revolutionary ability is just that — synthesized mimicry, derived from innumerable sources. By definition, we can imagine it to be Plato’s bird. “Behold,” we could cry out while holding our computer-generated featherless biped, “a man!”
Arguably, true “artificial intelligence” has not yet been discovered or quantified. We have powerful computer programs that can sift through incomprehensibly large sums of data at breakneck speed, yes. In this sense, we have achieved the synthesization of artificial knowledge, as demonstrated by the increasingly common use of ChatGPT and Claude in daily life. However, this knowledge base is markedly not up to the same standards that would define general human intelligence. For instance, current “artificial intelligence” is known to fail at tasks involving complex reasoning. More importantly, despite the common ability to generate human-like text, AI engines arguably do not truly understand the content they process and instead rely on patterns of data to decipher an outcome.
In this sense, the umbrella term of “artificial intelligence” is insufficient in defining the systems that have grown to encapsulate our daily lives. Much of our personal experience with “AI” comes from ChatGPT, Claude, DeepSeek and any other chat-based interface. These engines are referred to as Large Language Models — a type of “artificial intelligence” designed to understand and generate human-like text based on the input they receive. LLMs, while a breakthrough in human ingenuity, come with a stretch of limitations, most notable being their tendency to be incredibly energy-intensive. LLMs are trained on an enormous sum of data, such as the internet, public texts and social media. This expansive neural architecture means that even the simplest query consumes an extraordinary amount of power. The expansion of data centers and the incredible amount of water that is diverted to cool them is largely because of a corresponding expansion in LLM use.
It is important to recognize that “artificial intelligence” is not solely composed of LLMs. The search for a cancer cure, for instance, has seen remarkable breakthroughs through the implementation of non-LLM AI technologies. CleaveNet, developed by the Massachusetts Institute of Technology and Microsoft, is an AI pipeline program and protein language model that uses a chosen set of data, collected by researchers, for the design and prediction of protease substrates. Using this approach, researchers have been able to quickly sort through millions of possible combinations to find usable molecular markers for early stage cancer diagnoses and subsequent therapeutic treatments.
Similarly, DeepVariant, an open-source Deep Neural Network model, transforms DNA sequencing data into images, allowing it to quickly identify genetic mutations at a high precision rate. It is poised to save many lives and has already been used in tumor-cell research and disease diagnostics.
“Artificial intelligence” has also led to breakthroughs in the fight against global warming. The Ai2 Climate Emulator (ACE), though still currently a work in progress, is largely considered a major scientific success for its speed and physical stability when predicting global atmosphere models. ACE, like many new technologies that have adopted the “artificial intelligence” moniker, is not an LLM. Instead, it is an open-source autoregressive machine learning emulator designed to simulate global weather, atmosphere and climate patterns.
Notably, none of these “artificially intelligent” models suffer from the same issues that have caused LLMs to face moral and ethical concerns. They are significantly less carbon-intensive and have no sycophantic tendencies. They also have a clear, single-minded purpose which allows them to perform incredibly well in their designed tasks. Like LLMs, however, these models do not represent true “artificial intelligence.” They are plucked chickens that serve as very useful calculators. They are efficient, yes; knowledgeable, yes; but nowhere are they near true sentience or cognition.
In the face of this calculus, the lexicon of AI must be expanded. We cannot meet anything labeled as “artificial intelligence” with baseless reproach. Yes, some forms of AI, such as LLMs, are harmful and must be met with skepticism. The future of technology, however, rests with the “intelligence” that amplifies the already-intelligent human mind: the kind that crunches numbers faster to solve complex land-sea equations, that synthesizes molecules needed to cure cancer and the many possibilities not yet defined. Progress lies with the separate recognition of the many brands of technology that we call “artificial intelligence.” That begins with the words we use to define it.