Carnegie Mellon University historian Christopher Phillips and University of Pittsburgh’s Alison Langmead challenge conventional thinking about artificial intelligence with research tracing current AI rhetoric back to the 1950s. Their new paper in IEEE Annals of the History of Computing argues that discussions about AI reveal more about how humans discuss technology than the technology itself.
“We use words like ‘smart,’ ‘read,’ ‘write’ and ‘think’ very differently when we’re talking about humans than when we’re talking about computers,” Phillips said, questioning why the same terms are applied to both. Langmead and Phillips suggest a broader conversation is needed to clarify what computers can and cannot do, beyond the current hype cycle.
Early Computing Debates Shaped Current AI Language
The ambiguity surrounding artificial intelligence isn’t a recent phenomenon; its roots extend back to the earliest days of computing, according to new research. Christopher Phillips of Carnegie Mellon University and Alison Langmead of the University of Pittsburgh examined historical conversations surrounding computing, revealing a pattern that emerged as early as the 1950s. This practice, they argue, involved using language that held precise meaning for computer scientists while simultaneously suggesting broader capabilities to the public, a tactic that continues to shape perceptions of AI.
Early computing pioneers faced a choice: embrace human-centered language or adopt more precise descriptions of machine function. Computer scientist Norbert Wiener, for example, used the term “learning” to describe computers implementing successful rules, a technically defined term that carried broader connotations for non-experts.
Phillips and Langmead contend this distinction is crucial, as strategic ambiguity allows technical language to be accessible while potentially overstating a technology’s humanlike qualities. “Why can’t we say the computer is executing a particular set of instructions?” Phillips asked. “Why do we have to call it thinking?” This historical pattern extends to modern AI benchmarks. Tests like MMLU and “Humanity’s Last Exam” are often framed as measuring machine knowledge or reasoning, but the researchers argue they more accurately assess classification accuracy, how well systems identify correct answers, rather than genuine understanding.
Langmead emphasized the importance of a broader conversation. “The work Chris and I are doing focuses on how human beings have allowed computers — tools of our own invention — to become an integral part of our daily life,” she said. “Because they are now enmeshed in our social world, how we talk about them and imagine them matters a great deal. We would like a larger, ongoing conversation that transcends the hype cycle about precisely what computers can and cannot do.”
This timely study reminds us that language does not simply deliver scientific or technological ideas: It changes these ideas and it transforms our understanding of them.
Andreea Ritivoi, William S. Dietrich Professor of English and associate dean of research in Dietrich College
Norbert Wiener’s “Learning” and Strategic Ambiguity in AI
The deliberate use of ambiguous language surrounding artificial intelligence stretches back to the field’s earliest days, a pattern identified by researchers examining historical records. This early choice established a precedent for where technical precision is sometimes sacrificed for broader appeal. “Computational algorithms and mathematical derivations rely on precision.
Everyday language, however, makes heavy use of individual words that have multiple context-dependent meanings: they are often understood in a particular way by small groups of people within a limited setting, but others may easily attribute to those words very different meanings,” explained Rob Kass of Carnegie Mellon’s Department of Statistics & Data Science. This historical pattern continues to influence modern AI discourse, obscuring the distinction between computational processes and human cognition. Phillips questions the necessity of labeling a computer’s execution of instructions as “thinking,” asserting that such language risks misrepresenting the technology’s capabilities.
Computational algorithms, and mathematical derivations, rely on precision. Everyday language, however, makes heavy use of individual words that have multiple context-dependent meanings: they are often understood in a particular way by small groups of people within a limited setting, but others may easily attribute to those words very different meanings.
Rob Kass, Maurice Falk University Professor of Statistics & Computational Neuroscience in the Department of Statistics & Data Science and the School of Computer Science Machine Learning Department
MMLU and “Humanity’s Last Exam” Measure Classification, Not Knowledge
The increasing reliance on benchmarks like Massive Multitask Language Understanding, or MMLU, and “Humanity’s Last Exam” misrepresents what these tests actually measure, according to research from Carnegie Mellon University and the University of Pittsburgh. Rather than assessing genuine knowledge or reasoning, these evaluations primarily demonstrate a system’s ability to accurately classify answers within a standardized framework.
Phillips highlights that reducing complex human activities like creativity and learning to mere computational outputs overlooks the crucial role of lived experience and judgment. “Most of us read poetry because we’re interested in the person who wrote it,” Phillips said.
We use words like ‘smart,’ ‘read,’ ‘write’ and ‘think’ very differently when we’re talking about humans than when we’re talking about computers.
Christopher Phillips, professor and head of the Department of History in CMU’s Dietrich College of Humanities and Social Sciences
Interdisciplinary Origins of Computing Demand Broader AI Conversations
They discovered that some early computing figures intentionally employed imprecise language, while others did so unintentionally. The team found that during the mid-20th century, discussions about computing’s role involved historians, psychologists, and engineers, fostering a more interdisciplinary approach. Phillips emphasizes that framing AI as “thinking” obscures the fundamental differences between computation and human cognition.
The work Chris and I are doing focuses on how human beings have allowed computers – tools of our own invention – to become an integral part of our daily life.
Alison Langmead, University of Pittsburgh
See today’s quantum computing news on Quantum Zeitgeist for the latest breakthroughs in qubits, hardware, algorithms, and industry deals.
