“Artificial Intelligence is the mantra of the current era.”
Machine-learning pioneer Michael I. Jordan began his 2019 Harvard Data Science Review essay with this observation. “Mantra” signals repetition, power, and a loss of precision. By the late 2010s, artificial intelligence had become a label for statistical learning, optimization, data systems, robotics, and speculative humanlike intelligence. Jordan argued that grouping those ideas together can obscure the engineering and social work needed to build dependable services. The phrase does not dismiss progress. It asks whether a fashionable name is helping people understand what has actually been achieved and what remains unsolved.
Jordan distinguishes several ambitions often compressed into AI. Human-imitative AI seeks machines with capabilities resembling individual cognition. Intelligence augmentation uses computation to extend human decision-making. Intelligent infrastructure coordinates people, services, markets, and machines at large scale. The last two may transform daily life without producing a machine that thinks like a person. A transportation network that matches riders and drivers, for example, depends on prediction, incentives, reliability, and governance. Its intelligence is distributed across the whole system, not located inside a single model.
This systems view changes the research agenda. Accuracy on a benchmark is only one component. Deployed technology must handle changing data, strategic behavior, privacy, security, uncertainty, and the consequences of error. It must give people reasons to participate and ways to challenge outcomes. Statistics helps quantify uncertainty; computer science supplies scalable mechanisms; economics and the social sciences explain incentives and institutions. Jordan’s essay argues for a discipline broad enough to combine those fields rather than treating a learning algorithm as the entire solution.
The quote remains relevant whenever “AI-powered” is used as a substitute for explanation. Labels can inflate expectations, blur responsibility, and make ordinary automation appear autonomous or magical. More exact descriptions support better governance. A language model, risk score, recommender, or robotic controller has a different purpose and failure pattern. Each should be evaluated in its context, with evidence suited to its use. The revolution worth pursuing is not a contest to imitate every feature of human intelligence. It is the careful construction of systems that help people coordinate, learn, decide, and create without hiding their institutional design. The line encourages precise language. Before judging an “AI” system, ask whether it is a prediction model, an adaptive service, an optimization process, or something else. Clear names lead to clearer tests, more realistic expectations, and better decisions about where human judgment belongs. Precision improves public choices.
Repeated use of the AI label can make distinct technologies and ambitions appear to be one coherent thing.
Jordan asks readers to look past the name and identify the actual system, task, and evidence.
Human imitation, intelligence augmentation, and intelligent infrastructure involve different scientific and social questions.
Important progress does not require every useful system to resemble an individual human mind.
Reliable deployment combines machine learning with statistics, computing, incentives, institutions, and human oversight.
Benchmark accuracy cannot by itself establish that a service will work safely in a changing social environment.
Explore more "Quotes of The Day"
Continue exploring how writers, thinkers, artists, and advocates have described the choices that shape an everyday life.