During a recent visit to San Francisco, I summoned a driverless car and rode in silence, the vehicle moving without a human behind the wheel. The experience was smooth, and it sparked a conversation with Susan Winslow, chief executive of Macmillan Learning, about how artificial intelligence is reshaping research and education.
What the Waymo effect means
Both of us noticed that the lack of a human driver removed the need for small talk, allowing us to focus on our own thoughts. We call this the Waymo effect: a technology eliminates the friction of interacting with another person, and we perceive that loss as a gain because the effort was obvious while the benefit was hidden.
The idea echoes Tim Wu’s argument about “the tyranny of convenience,” where a frictionless option becomes the default and the value of the removed step fades from view. Albert Borgmann’s device paradigm describes a similar shift: a device supplies a commodity while obscuring the practice once required to obtain it.
Human contact as hidden value
Talking to a taxi driver carries obvious costs—politeness, brief conversation, a moment of social exposure. The upside is less tangible: the driver may be the last stranger you encounter that day, offering a perspective you never asked for. Those spontaneous exchanges can broaden a researcher’s outlook.
Neither Winslow nor I would design a world without such encounters; we simply appreciated the comfort of opting out for that single ride. The pattern, however, repeats whenever convenience outweighs the unseen benefit.
Read Also: Staff shortages hit higher education sector hard
Large language models replace collaborators
Large language models act as the Waymo of intellectual work. They are available at any hour, unlike human co‑authors who juggle teaching, grants, and time‑zone differences. An LLM has no agenda beyond responding to prompts, while a colleague may challenge assumptions you never considered.
If you ask an LLM to critique a paper, it will do so, but only on the material you provide. It will not point out that you are tackling the wrong problem or that a similar effort failed years ago. The inconvenience of a human partner—disagreement, unexpected insight—is precisely what makes collaboration valuable.
When conversations shift from a colleague to a chatbot, a thread of the research community’s fabric is quietly withdrawn. I suggest the term “decollaboration” for this systematic erosion of shared practice.
Incentives that favor isolation
Funding cuts often eliminate travel, workshops, and sabbaticals, activities that support serendipitous encounters. At the same time, evaluation metrics continue to reward rapid output, and LLMs promise speed by drafting literature reviews or sections of a manuscript in days rather than weeks.
Because an AI tool never contests authorship, it appears as a low‑cost partner in a system where credit determines career survival. The combination of tighter budgets, speed‑focused assessment, and ego‑free assistance makes the solo route increasingly rational.
Read Also: OfS risk ratings reveal deeper concerns over university finances
Team‑science research shows that small, unexpected collaborations generate the most novel findings, while early studies of generative AI suggest individual productivity rises as idea diversity narrows. The current trajectory resembles an uncontrolled experiment in trading serendipity for throughput.
Compared with the automation of aircraft cockpits in the 1970s, today’s AI tools risk dulling the researcher’s ability to handle uncertainty. When machines handle routine tasks, the human skill set that once kept projects on course can atrophy without regular practice.
Writing, thinking, and the hidden cost
Writing is not merely a way to share results; it forces authors to confront gaps in reasoning. Removing the effort of drafting, by outsourcing to an LLM, does not accelerate thinking, it bypasses the struggle that reveals hidden flaws.
Psychologists describe this as “desirable difficulties.” Effortful retrieval and delayed feedback create deeper understanding. If the process becomes effortless, the outcome may look polished while the underlying comprehension remains shallow.
As models improve, their prose becomes harder to distinguish from that of a skilled scholar, making the illusion of progress more convincing. Metrics may show faster turnaround and cleaner text, yet the invisible erosion of critical thinking can take years to surface.
Read Also: MIND Education Launches Brain-Based Math Curriculum InsightMath
Balancing tools with human oversight
A March 2026 comment in Nature proposed that AI agents be treated like “airplanes for the mind,” powerful but demanding a pilot in command. The author urged researchers to retain authority over questions, methods, and conclusions.
Designing AI systems that deliberately introduce dissent, multiple models that propose alternative approaches, could restore some of the lost friction. This mirrors Lisanne Bainbridge’s 1983 observation that automation can erode human competence when the automated system is relied upon too heavily.
Funders and institutions should treat collaboration as essential infrastructure, not an optional extra. Supporting workshops, visits, and unstructured time would re‑introduce the frictions that nurture diverse ideas. Evaluation should consider who a scholar thought with, not just what they published.
Winslow and I concluded our meeting after the ride, but the real discussion happened later, in the unplanned hallway chat and the email exchange that followed. Those unscripted moments produced ideas that never entered the driverless car, illustrating the Waymo effect in practice.
Leave a Reply