324 | Elizabeth Mynatt on Universities and the Importance of Basic Research
What this covers
Elizabeth Mynatt, Dean of Computer Sciences at Northeastern University, discusses the architecture of American research alongside Sean Carroll in this episode. The conversation centers on why universities—despite their apparent inefficiency as institutions—have become the primary engine of technological breakthrough. Mynatt traces this to the postwar system designed by Vannevar Bush, which paired peer review and federal funding with a dual mission of research and education. She argues that this combination creates a distinctive ecosystem: researchers accept lower salaries than industry offers in exchange for freedom to pursue long-horizon questions, while students gain exposure to cutting-edge work that later propagates across industries. The episode ranges widely through her own research in human-centered computing and technology design for aging populations, grounding abstract arguments about innovation in concrete examples.
The substance of the argument unfolds through specific cases. Mynatt traces how military-era RFID technology, developed for aircraft identification, was repurposed by Cornell researchers into "cow tags" for dairy monitoring and eventually enabled robotic milking—showing how Universities as Intellectual Cross Streets allows ideas to migrate unpredictably across domains. She contrasts this with industry's narrow profit motive, noting that Public Good Research like assistive technologies for people with disabilities would never attract venture capital alone, yet paired with regulation and institutional commitment, generates widespread benefit. A recurring theme is the Resurgence Pattern—how unfashionable fields like neural networks and mRNA vaccines survived academic "winters" that industry would have abandoned, only to become foundational when computational or biological conditions aligned. She also turns to the friction between technology design and human behavior, arguing that the human element remains harder than engineering, and raises the ethical stakes of ubiquitous computing systems that collect intimate data about decline and autonomy.
Mynatt argues that the unique synergy of universities—combining education and blue-sky research under a publicly-funded, peer-reviewed system established in the 1950s—is what drives long-term technological and economic breakthroughs, and that current funding cuts threaten an irreplaceable ecosystem.
- Universities pursue long-horizon, public-good research that industry's short-term profit motive will never fund, yet this research yields trillions in downstream value.
- The dual business model of research plus education makes universities strangely efficient and lets insights cross-pollinate across disciplines.
- Peer review, open publishing, and federal funding provide objectivity and reliability that private funding cannot match.
Universities' dual business model—creative research at low salaries plus education—generates enormous economic return and cross-pollinating innovation industry alone cannot achieve.
- Universities are strangely highly efficient because two business models sit side by side: a creative-research model that convinces people to work long hours at much lower salaries than industry out of commitment to the mission, and an education model where people pay to learn and be exposed to new ideas—this combination has been the foundation of American innovation and economic growth since the 1950s.
“Universities are highly efficient because we have these dual business models sitting right next to each other. We have the business model of creative research, that convinces people to work long hours at much lower salaries than they could get in industry.”
- Universities generate innovation because they sit at intellectual cross streets where people are exposed to research from many sources (e.g., military-funded RFID), work in laboratories that are 'time machines into the future,' and then graduate into diverse industries carrying ideas that combine in unexpected ways—as when cow-tag RFID migrated into home and family applications.
“And that's the magic of universities. Universities sit at these intellectual cross streets... they work in laboratories that are essentially time machines into the future, and then they graduate out and go into all sorts of industries”
- The system of government funding university research is a 20th-century invention codified by Vannevar Bush, who—struck by how technology won World War II—laid down rules at the close of the war to push research out to conveniently distributed universities, pair research with workforce education, set national priorities through mission agencies while remaining hands-off, and bake peer review into the NSF as part of the 'secret sauce.'
“Vannevar Bush was a scientific advisor during the war and famously wrote a number of articles coming at the close of World War II... he laid down the rules of how this would work.”
- Talented researchers often take pay cuts to move from industry to academia because they gain freedom to pursue blue-sky work and openly discuss their research, as Mynatt did leaving Xerox PARC when she could no longer twist human-technology interaction research to fit the document company's needs—and the best faculty thrive on rotating between industry product work and returning to creative scholarship with students.
“I wanted to work on how people interacted with these everyday technologies, and I ran out of ways to twist it into the pretzel of the document company”
- The economic return on university research is measurable and dramatic—annual investments in the low billions correspond to annual revenue in the low trillions—because the dual-business-model university produces research at low salaries that cross-pollinates across industries in ways that segmented, doubly-paid industry researchers never could.
“Universities are highly efficient because we have these dual business models sitting right next to each other. We have the business model of creative research, that convinces people to work long hours at much lower salaries than they could get in industry.”
- Universities and industry operate on opposite economic models: industry follows a narrow path of bringing something to market for profit, while universities are rewarded for thinking big over long horizons, doing foundational and creative work, and—crucially—for sharing ideas through publication and citation, making openness a virtue in academia and a danger in industry.
“Universities are the opposite of that, right? They are rewarded for thinking big, thinking long horizons, doing something that is foundational... And we get rewarded on how much other people read about it. So it only counts if you get it published”
- Radio frequency identification, originally a World War II friend-or-foe transponder technology, was repurposed and miniaturized by Cornell researchers for cattle—becoming so dominant in dairy that the tags were known as 'cow tags'—enabling tailored feed, milk-production monitoring, and eventually robotic milking machines that let labor-short dairy farms stay economically self-sufficient.
“this new technology called radio frequency identification. It had been used in World War II... some researchers at Cornell said, this could be useful for cows. And they developed and miniaturized the technology to the point that RFID tags for a long time were just known as cow tags”
Federal research funding is a strategic national asset that develops foundational knowledge decades before impact and cannot be regrown once destroyed by budget cuts.
- Because no funding system can be perfectly optimal or forecast which research will pay off, agencies manage research like a venture-capital portfolio—balancing criteria such as geographic distribution (the 'missing millions'), programs for junior researchers (NSF CAREER awards), and targeted societal-impact institutes (AI for aging, agriculture, education)—laying down diversified bets to mature the portfolio over time.
“at this point, you're managing a portfolio, right? Venture capitalists are managing a portfolio, and they're taking risk, and over time, you understand kind of like how to lay down different types of bets”
- Over the past decade-plus the National Science Foundation, under director 'Paunch,' deliberately pushed research investment across the United States to the 'missing millions' rather than concentrating it in the top ten universities, on the belief that bringing research and student education as close together as possible is where the magic happens.
“Ponch, the previous NSF director, would talk about the missing millions and really pushed that research investments went out across the United States, not just to the top 10 universities, but it was getting out everywhere”
- Industry's heavy AI investment cannot substitute for federal research funding because, as in a gold rush where the people who profit are those selling the shovels (explaining NVIDIA's success), industry investment is narrow and short-term—lacking the scale, breadth, and community ownership needed to decide what is worth investing in.
“remember the people who made money in the gold rush were the people who sold the shovels?... So this explains NVIDIA... Industry is very narrow and very short term in terms of what they can invest in.”
- Stubborn university researchers sustaining unfashionable fields—neural networks through AI winters, number theory before e-commerce existed, mRNA vaccine techniques—lay the theoretical foundations decades in advance, so that breakthroughs that feel instantaneous (like generative AI) were actually decades in the making once enough data and compute arrived.
“What just appeared was decades, decades and decades in the making... Number theory, right?... people working on the mathematical foundations of information science, they weren't thinking about Amazon”
- Proposed cuts of up to 60% to the National Science Foundation are clear-cutting a flourishing research forest on the myth that something else will grow back, but you cannot regrow a forest just because an administration changes; talent is already leaving the country to set up research elsewhere and—because inertia is not frictionless—will not flock back, doing dramatic, lasting damage at the very moment AI makes the research ecosystem more needed than ever.
“One of the versions out there is 60% cut in the National Science Foundation... it feels like, we've got this flourishing forest and we're clear cutting it with the myth that something else is going to grow and it's not going to grow.”
- The university system has natural forcing functions that distinguish ahead-of-its-time research from genuinely failed programs: researchers must keep publishing and attracting students to participate in the academic economy, so when an area stops gaining traction people shift to areas where they can sustain that currency, and fallow fields like virtualization get reinvented (e.g., for cloud computing) when someone finds a creative new use.
“there's natural limiters in the system... at some point, it's like, all right, I have to keep publishing and I have to be able to attract students because that's my currency. And people will shift into areas where they can continue to participate”
Human-centered computing requires understanding what people actually do with technology over decades, not optimizing narrow technical metrics or imposing designs unsuited to their needs.
- The standard model of pushing out technology innovations—frequent interface updates designed mostly for 20-year-olds—imposes a measurable psychological and cognitive cost on older adults, who must repeatedly relearn tasks like online banking, making this the wrong model for designing successful solutions for them.
“we can measure the psychological impact, the cognitive impact, every time a new update comes out... our model of how we spew technology innovations out into the world, which is mostly based on 20-year-olds, is not the right model for how we would design successful solutions for older adults.”
- Early ubiquitous-computing researchers misjudged mobile devices, treating them as interchangeable 'digital Post-it notes' scattered around the environment, and failed to anticipate that they would become so deeply personal that people would share information and use them in entirely unpredicted ways.
“we really saw these as interchangeable. They were like electronic Post-it notes, digital Post-it notes... but we never really thought of them as so deeply personal as the way they landed up becoming.”
- Quoting Fred Brooks—'never underestimate the laziness of the user'—Mynatt argues that brain-computer interfaces and similar effortful technologies will fail to gain mass adoption because they are more work than they are worth, since people get sufficient information from low-effort channels like texting.
“Fred Brooks, a famous scientist, software engineer who worked in virtual reality, said never underestimate the laziness of the user.”
- The most counterintuitive lesson of human-centered computing is that people—not technology—are the hardest part of the equation; understanding what people will do and why they will do it is far more challenging than the technical engineering.
“I think people is always the hardest part of the equation. Everyone assumes technology is the hardest part. Understanding what people will do and why they will do it is really challenging.”
- Human-centered computing is roughly a 20-year-old field that arose as computing moved from workplaces into everyday life; rather than contorting humans to fit machines, it designs computing systems people actually want to use, integrating psychology, human factors, ethnography, qualitative research, and ethical/privacy trade-offs.
“instead of contorting the human being to how they would use computers, how do we actually start designing computers and computing systems such that people would actually want to use them?”
- Technologies often take decades to have impact, and the eventual high-impact use is frequently a completely different application than originally imagined—VR, for example, became foundational in automotive design rather than the hyped consumer headset wandering the streets.
“sometimes it is decades in the making in terms of understanding when these technologies will have such an impact. And oftentimes it will be a different application of that technology that really takes hold.”
- Human capability decline appears to happen in sudden step functions where life falls apart, but in reality it is a long slow decline that people compensate for so well that the gradual reality is masked; ambient home technology could detect early cognitive decline that humans hide.
“human behavior tends to show up in step functions. You seem to have perfectly reasonable control of your life, and then things fall apart... And in reality, it's actually a long, slow little decline, but we compensate so well for that that it masks what's actually happening.”
Universities investigate failures and public goods industry will not—privacy leaks, insecurity, assistive technologies—that regulation and ethics frameworks must address together.
- Because human-centered computing involves ethical trade-offs of risk, autonomy, and control of information—such as who counts as the 'user' when an older adult's children pay for and set up the technology—a background in ethical frameworks should be a requirement for every student in the field and foregrounded in research.
“having a background of ethical frameworks and foregrounding that into technology innovation, I feel should be a requirement of every student studying in this field.”
- Calls for no regulation of AI are dangerous because decades of technology innovation show that violations of trust, privacy, and security can be managed with regulation in place while still innovating quickly—and faculty embedded in the White House, Senate, and DOJ work precisely at this intersection of business models, regulation, and technology.
“that's why I'm very distressed over these conversations around no regulation around AI because we have to be able to innovate as quickly as possible. Like, are you kidding me?”
- Universities investigate questions industry never will: the ways AI fails (privacy leaks, insecurity, data poisoning, censorship) which industry won't publish even if studying them, and public-good technologies like assistive technologies for people with disabilities, whose markets are too small to exist alone but which—paired with regulation like the ADA and Section 508—profoundly improve millions of lives and reliably end up benefiting everyone.
“universities will poke at and embrace questions that industry never will... we have folks doing research about all the ways that AI fails... if you're not publishing, then the rest of the world can't learn and improve.”
- For older adults at greater risk of fraud and scams, human-centered design may justifiably 'turn the knob' toward a little less privacy and more control by family or institutions—something inappropriate for someone in their 30s but reasonable in their 70s—because deployment of monitoring is fundamentally a human ethical question, not a technical one.
“maybe turn the knob a little bit of a little less privacy and a bit more control by family members or other institutions to keep a tighter rein on things, something you wouldn't do to someone in their 30s, but now in their 70s, it's like, okay, there's more risk here.”
- Whereas computing students in the 1990s embraced technology determinism and dismissed ethics as someone else's problem ('I just do the coding'), by a decade later—after Facebook and similar events—students recognize technology is a 'hot mess' and actively ask for ethical tools and guidance, though a large gap remains between heady ethical understanding and the tools needed when actually writing code.
“When I started teaching ethics and technology... the students hated it. They were complete technology determinism... And the students are like, yeah, no, technology is a hot mess. Give us tools and guidance.”
Interdisciplinary crossing points where researchers encounter diverse fields and ideas are where breakthrough innovation happens and universities deliberately reward such work.
- The most exciting innovation happens where disciplines cross—the 'white space' between fields—and universities have responded over the past decade by lowering barriers and increasing rewards for interdisciplinary work, evidenced by over 50% of Mynatt's students pursuing combined majors and the 1990s shift in grant data from single-investigator to multi-investigator, multi-discipline proposals.
“It is where the disciplines cross that things get really exciting... we saw it in the data in the 90s when it had flipped to multi-investigator, multi-discipline focused.”
- While industry rides the current tall tidal wave of generative AI, universities should surf ahead to the next wave—smaller, more efficient models applied to specific domains through cross-disciplinary partnerships—and Mynatt bets that much of the eventual impact of today's AI will actually come from these next-generation systems that universities can begin building now.
“the next wave is around these smaller models. It's more efficient models. It's models that are going to be applied into very specific domains where universities can do this cross-disciplinary partnerships.”
Early technologists often fail to predict how people will spontaneously adopt and repurpose technologies in entirely unexpected social and practical ways.
- At a National Academies broadband-future meeting, Mynatt was treated as heretical for predicting that people would start getting their news from social feeds rather than traditional news organizations, because committee members—viewing the internet through a cable-broadcasting lens—could not imagine why anyone would trust random social feeds over major channels.
“I said people are going to start getting their news from their social feeds as opposed to traditional news organizations. And they all but kicked me out of the room for being so heretical and so ridiculous.”
- As soon as connective affordances appear on a network—even crude ones like renaming oneself in a 1993 chat app—people spontaneously begin sharing their status and finding interesting social uses, a natural human inclination that preceded products like Facebook by many years.
“as soon as those little baby affordances were there on the internet, people started doing that. And so Facebook was many, many years later after that. But it was just this natural inclination.”
- Machine-learning approaches to predicting technology impact tend to use too narrow a funnel—tracking metrics like model accuracy—while neglecting contextual factors such as business readiness, hardware availability, and regulation; the real difficulty is predicting when all these stars align.
“those predictions tend to have a very narrow funnel... But what they need to also take into account are all the contextual factors, right? So the business readiness and the availability of hardware and even the particular regulatory side of things.”
Peer review and the publishing system are evolving toward open access because researchers demand it, the pandemic accelerated the shift, and this change is irreversible.
- Scientific publishing paywalls are falling because researchers want open access but are constrained by journal economics; the pandemic accelerated open preprint sharing (e.g., arXiv) in an all-hands-on-deck emergency, and professional bodies like the ACM are now shifting to open models and finding alternative revenue, making the open-access genie impossible to put back in the bottle.
“The pandemic did quite a bit for this... pre-publishing in my field, it's in archive. You're getting results out there... You're not going to get the genie back in the bottle after that.”
- Peer review is not merely a gatekeeping mechanism but an integral part of the research process itself: the act of writing, reviewing, and arguing about proposals gives researchers valuable feedback that improves their thinking even when they aren't funded, making the system more objective than alternatives where someone simply hands money to favored institutions.
“those discussions of the act of writing proposals and reviewing proposals and arguing about proposals, that is part of the research process, because it allows people to, you may not get funded the first time... but you get really good feedback”
Ubiquitous computing systems face security and privacy as weakest-link problems where one compromised device breaks the entire network's safety.
- Internet of Things security and privacy is fundamentally a weakest-link challenge: a home full of well-behaved internet-connected devices can be compromised by a single bad actor that grabs information from everything else and broadcasts it to the world, breaking the entire safety net.
“it's in some sense a weakest link challenge from a security and privacy point of view... you get one bad actor in the home, and it's grabbing information from everything else and then broadcasting out to the world. Like, just one bad actor breaks the security and privacy safety net.”
- The maxim 'if you think it's free, you're the product' captures how ostensibly free technology services monetize users' data, but Mynatt is hopeful corporations will eventually recognize simplicity and privacy as selling points once they have extracted as much data as possible.
“There's the old saying, right, nothing's free. If you think it's free, you're the product, right? And it's your data.”
Public messaging about research value should emphasize concrete societal benefits—farms, healthcare, cancer prevention—not abstract computer science.
- Advocates have done an insufficient job explaining research's value because the public hears 'computer science research' and thinks of Meta, Google, and Amazon—companies they assume need no help—when in fact federal research dollars go to farms, healthcare, safe automobiles, and preventing cancer deaths; the 'losing talent' and 'cows not Google' framings are the messages that ring true in Washington.
“most folks, when they hear computer science research, they think meta, Google, Amazon... And I'm like, no, it's actually about farms and health care and safe automobiles and preventing cancer deaths.”