Education for Versatility and Virtue

Credit: Greg Rakozy/Unsplash.

How can higher education best prepare graduates for professional life? Recent articles in the mainstream press [1–2], higher-ed trade journals [3–4], and education-industry partnerships [5] reflect the broad interest in this question. For decades, attention has centered on the supply of STEM (science, technology, engineering, and mathematics) graduates, and whether this was sufficient to preserve American dominance in those fields. Today, there seems to be a more diffuse anxiety. Funding for scientific research is at historic lows, talented technology workers are more numerous, so people are left to wonder: What kind of education will maximize my chances of securing a good job that can grow into a fulfilling career?

My purpose here is to promote a kind of education — less of a subject, more of a style — that I think will serve as the best preparation for the jobs of tomorrow and for the most urgent and vital questions of today. This will require confronting questions concerning technology, the modern university, and an imposing tableau of social, economic, and environmental challenges. All this I will do. But first, I need to talk about thermostats.

From thermostats to metacognition

In 1979, the Stanford computer scientist and artificial intelligence pioneer John McCarthy suggested that an old-school, mechanical thermostat might properly be said to have beliefs [6]. Really, one very simple kind of belief: that the temperature in the room right now is X. It seems obvious that a thermostat can represent the temperature, and indeed can even act on it by switching on the heat or the air conditioning…but why should we think of this as a belief?

McCarthy was not perverse, nor was he tempted to attribute intelligence to thermostats. Thermostats are, without a doubt, dumb. McCarthy was exploring the conceptual framework we use to discuss our own mental interiors — we believe X and know Y, we want Z, etc. — to see if these ideas could reasonably be applied to, or extended to, the artificial intelligences that he imagined would one day be invented.

Fast forward to 2016, when DeepMind’s Go playing computer AlphaGo defeated Lee Sedol — at that time, ranked number two in the world. Everyone acknowledged the astonishing computational ingenuity of AlphaGo’s engineers. Some observers, including players that AlphaGo defeated, were willing to attribute actual insight to the machine itself. But no one seriously believed that it compared favorably to the overall prowess of a human mind. It may not have been dumb, but it was at best an idiot savant. This assessment still feels true, but what is it missing? Why is such a powerful engine, able to defeat grandmasters at one of the world’s subtlest games, excluded from the ranks of thinking beings?

An influential answer to that question was presented years earlier by John Haugeland, a philosopher at the University of Pittsburgh [7]. Continuing McCarthy’s line of inquiry, Haugeland argued that in order to think, the thinker has to be aware of, and care about, the possibility of error. The thinker must recognize the possibility that something might have gone wrong in their perceptual or conceptual apparatus. They must be capable of thinking, “Hang on, that simply cannot be right,” and of taking steps to uncover the source of the mistake.

When we deliver this challenge to the thermostat, it fails badly. Whatever the sensor reads, it believes. It is simply not in the thermostat’s repertoire to say to itself, “My sensor indicates that the living room is six million degrees, but that is absurd; it must be broken, or I must be dreaming, or…”. To think that would require the thermostat to have beliefs about beliefs, as well as a commitment to understanding the world properly. It has neither. Similarly, AlphaGo can represent the illegality of a certain move, but only in the shallowest way. Because AlphaGo’s entire mental universe is circumscribed, it does not know what “illegal” means, or that Go is a game, or that it is playing anything. It cannot rise above the rule set that defines its mental universe to see that there is a larger context to its own thinking — a context necessary to understanding higher-level truths.

Part of what made AlphaGo possible was the finiteness of the game’s rules. Profound complexity can still be found in the consequences of those rules, but somehow, their boundedness released the engineers from having to tackle the vastly greater challenge of creating a general artificial intelligence. In a way, similar engineering advances had become visible on factory floors decades earlier, where robots replaced humans in repetitive, finitely describable, manual work. We might venture to predict, then, that any problem that can be described clearly will eventually be automated. In the long run, machines will perform the tasks needed to solve these problems, while humans work on problems that cannot be — or have not yet been — described clearly.

By now, the point of the thermostat story should be clear. A key shortcoming of machine intelligences, so far, seems to be their lack of a capacity for metacognition: the ability to think about their own thoughts. And from this shortcoming comes a profound lesson for humans: If you cannot see the boundaries of your own mind; if you do not recognize that there are such boundaries, whether you see them or not; if you cannot acknowledge, or do not care about, the possibility of error in your perceptions or judgments; if you cannot rise above your context — then you are simply a glorified thermostat. And you may soon be replaced by a better, faster, ‘smarter’ one.

Curriculum, meet pedagogy

If we were to survey students in the waiting rooms of academic career services offices, I doubt we would find many hoping to develop their metacognitive skills. Instead, I imagine that when most students approach the question of career preparation, the spotlight is on their course of study, especially their major(s). Presumably, this reflects their expectation that future employability depends upon the knowledge and skill they will acquire. In other words, the curriculum. Here, I will make recommendations about curriculum as it relates to job preparation, and also take up the question of pedagogy — that is, the way that classes are taught.

Consider the following thought experiment, devised by philosophers Andy Clark and David Chalmers [8]. A young woman named Inga hears of a new exhibit at her favorite museum. Knowing the address, she sets out and soon arrives. In the same city lives an old man named Otto. Otto has Alzheimer’s disease, but he compensates for some of his memory loss by keeping a notebook with frequently-used information. The notebook is so integral to his daily life that he is never without it. It includes directions to the same museum. When Otto hears of the new exhibit, he sets out, frequently consulting the notebook as he goes. He too soon arrives without trouble.

Most of us would say that Inga knows how to get to the museum, but Otto does not. The notebook is a prosthesis that allows Otto to get around effectively, but in the end he does not really know where the museum is. Clark and Chalmers disagree. They argue that Inga and Otto are in the same mnemonic situation: they both know where the museum is; it is just that their systems for memory storage differ. Inga’s memory is encoded in her brain, while Otto’s is in his notebook. You might object that Otto could lose the notebook. True, but Inga could have a stroke; either one could lose access to the knowledge at any time.

This idea has become known as the extended mind thesis, and it presses us toward a startling new view of mental activity. Thinking is not just something that happens inside skulls. Thinking — not only remembering, but the actual process of working through ideas — loops out into the world, involving notebooks, sketch pads, whiteboards, the internet, even other people. By cultivating the metacognitive ability to recognize when we lack knowledge, seeking it out, and using it, we enhance our mental capabilities. According to this thesis, using external resources in this way is not a cheat; it is not “almost” thinking. It is actual thinking.

The implication for students is that if they are sufficiently self-aware to recognize the limits of their own thinking; they have reliable access to books and journals, the internet, diverse colleagues, etc.; and they know how to vet and use those resources, then they will have effectively the same abilities as if they had memorized all that information, but with almost none of the upfront effort. If you happen to be in a profession where the problems are highly variable (which, in the long run, I think will be typical), that differential advantage becomes magnified. Besides, if certain elements turn out, again and again, to be essential to your work, you will find them committed to memory almost automatically, without much conscious effort.

What kind of curriculum, specifically, strengthens these mental muscles? Differing answers can be found in the general education programs of institutions around the world, but the consensual view reveals clusters, including critical and creative thinking, communication, teamwork, and ethics. These skills likely need to be more highly resolved and explicated in order to assess accurately, but a full discussion of that point is beyond the scope of this essay.

Now, if you know that a job will require discipline-specific knowledge immediately, fine. But beware: many employers report that academic institutions fail to keep pace with changing professional landscapes [10]. In the time it takes a student to work his or her way through a major — the median time to complete a bachelor’s degree is 4.3 years [11] — some of what was learned may be irrelevant. And it is far from assured that academicians are sufficiently connected with practitioners to ensure what they teach is current to begin with, much less that they can rapidly redesign courses if they discover a mismatch.

I’ll illustrate the point with a metaphor, based on a wise tip from travel writer Rick Steves, who forcefully recommends packing light [9]. Why? It’s liberating. Your stuff is less likely to get lost. And it’s cheaper. Obviously, he doesn’t recommend traveling so light that you lack essentials, but rather, packing a versatile, layerable wardrobe, planning for the best rather than for every contingency. He writes, “Rather than carry a whole trip’s supply of toiletries, take enough to get started and look forward to running out… Then you have the perfect excuse to go into a Bulgarian department store, shop around, and pick up something you think might be toothpaste.” Translation: The best bet for most students is to master a compact set of powerful, highly repurposable capabilities — a layerable mental wardrobe — and only the specific disciplinary knowledge that is certain to be critical at the start of their journeys.

As for pedagogy, the most important thing for students is to engage deeply and effortfully with whatever they are learning, so that they know what they know and can use it to maximum effect. Broadly speaking, this approach to education falls under the heading of active learning, and it comes in many styles, from problem solving in peer groups to reflective solo work to critical debate with an expert instructor. Many of these are intended to promote near and far transfer — the application of knowledge to contexts distinct from those in which it was originally learned. This ability is crucial for the transition from school to the working world.

Active learning is fairly easy to recognize when contrasted with the paragon of passive learning, which is lecture. Lecture focuses on the dissemination of information from an authoritative source to the student. While this can be a good way of exposing students to new ideas, mere exposure is not conducive to understanding. It is not even very conducive to remembering, let alone the sort of fluid, synthetic cognitive processes that support complex abilities such as those that will help students in their future jobs.

You cannot get a good education — that suggests passive receipt — but you can build one. And, like all construction projects, a solid foundation is critical.

Professional choice and global prosperity

Although I have promoted a specific approach to education, I have made no recommendation concerning any particular job or career. I will remain neutral on that, but in light of my earlier exhortation to step outside of one’s own context, I feel compelled to comment on the global environment in which students make their career choices.

It would be a shame if the only thing to come from a reimagined school-to-work pipeline was a system that allowed everyone to afford the house and car of their choice, but did nothing to address the prevailing strategic challenges facing human civilization as a whole. It is too easy to lose ourselves in family-sized consumeristic bubbles, to feel that success and security in those micro-environments somehow insulates us from the storms outside. As climate change, ecological collapse, economic injustice, and political upheaval rattle our walls, the fragility of that feeling becomes increasingly difficult to ignore.

Should educators encourage students to pursue careers that address such challenges, rather than merely self-interested pursuits? I think the answer has to be ‘yes’. As long as the motives are transparent, I don’t see any ethical problems with such encouragement. Moreover, so enormous is the sweep of these challenges, so great the effort required to tackle them, that there is room for every interest: There would be no need to pressure students into a course of study that they would not otherwise find rewarding.

Perhaps the most comprehensive map of current global challenges is the United Nations Sustainable Development Goals (SDGs), adopted in 2015 after almost thirty years of multilateral, international work. Here I wish to address just one curious, and as far as I know, unremarked feature: Almost every SDG aims at an emergent property of a complex system. By this, I mean that the problems they address have no single cause, no one locus of control, but are instead dynamic and difficult to predict consequences of interactions among many actors and systems. Not all problems are like this, not even all global challenges. In the early twentieth century, for example, many worried that agricultural production would fail to meet the demand of a growing population. It was a conceptually simple problem: how to produce more food? The solution turned out also to be conceptually simple: the Haber process, which allowed the industrial-scale production of fertilizer. More fertilizer meant more food. Of course, we are now reckoning with many unintended consequences of agricultural industrialization, but the basic solution was clear.

Contrast this linear problem structure with the most closely related SDG, which is to “End hunger, achieve food security and improved nutrition and promote sustainable agriculture.” One immediately sees that a solution to this problem requires more than mere production and must address, for instance, food distribution, which in turn requires consideration of infrastructure, politics, and economics. The sustainability of agriculture is also, all by itself, an extraordinarily complex problem, touching upon land and water use, ecology, technology, intellectual property (e.g., of genetically engineered crops), and more. Such webs of interdependence are apparent in most of the other SDGs as well, including poverty, education, climate action, responsible consumption, peace, and sustainable economies.

The reason I highlight this feature of the SDGs is that solutions to complex problems like these will not be crafted by experts in any single discipline. They will instead require creative thinking, problem solving, participation in and coordination of diverse teams, superlative research and analytical abilities, and a keen awareness that the paradigms responsible for the problems may need to be disrupted in order to solve them. These are precisely the types of abilities developed in the model general education program I am encouraging. Of course, domain specific knowledge will be required. It simply will not be sufficient. To solve these complex challenges, to stitch the disparate knowledge together, will require something more than single-domain expertise, no matter how impressive [12–13].


  1. Marcus, J. (2020, February 20). How technology is changing the future of higher education. The New York Times.
  2. Svrluga, S. (2019, November 14). Is college worth it? A Georgetown study measures return on investment — with some surprising results. The Washington Post.
  3. The Chronicle of Higher Education. (2019). Broken ladder: Higher education’s unfulfilled promise.
  4. Mintz, S. (2019, November 6). Career preparedness. Inside Higher Ed.
  5. National Association of Colleges and Employers. (2019). Measuring competency proficiency: the career readiness pilot project.
  6. McCarthy, J. (1979). Ascribing mental qualities to machines. In Martin Ringle (ed.), Philosophical Perspectives in Artificial Intelligence. Humanities Press.
  7. Haugeland, J. (2002). Authentic intentionality. In Matthias Scheutz (ed.), Computationalism: New Directions, pp. 159–174. MIT Press.
  8. Clark, A. & Chalmers, D. (1998). The extended mind. Analysis 58,10–23.
  9. Steves, R. (n.d.) Packing smart and traveling light.
  10. Wilkie, D. (2019, October 1). Employers say college grads lack hard skills, too. Society for Human Resource Management.
  11. U.S. Department of Education, National Center for Education Statistics. (2019). Baccalaureate and Beyond (B&B:16/17): A First Look at the Employment and Educational Experiences of College Graduates, 1 Year Later (NCES 2019–106), Table 2. Retrieved from
  12. Murray, S. (2019, November 6). Is ‘AQ’ more important than intelligence? BBC.
  13. Dede, C. (2017, December 11). Students must be prepared to reinvent themselves. Education Week.

I am Vice Provost of Academic Innovation at the Minerva Schools at the Keck Graduate Institute.

Get the Medium app

A button that says 'Download on the App Store', and if clicked it will lead you to the iOS App store
A button that says 'Get it on, Google Play', and if clicked it will lead you to the Google Play store