From birth, parents raise us in different ways, teachers teach us in different styles, and doctors treat us with different medicines and give us different advice. These experiences set us upon our paths in life, sculpting how we learn and how we stay healthy. We can think of each interaction between a teacher and a student, or between a doctor and a patient, as a little miniexperiment, with an outcome that can be evaluated: Did the student learn and become able to use the information to do creative and useful things throughout life? Did the patient improve in health and develop proactive health-related behaviors? With almost one million physicians, and about four million educators of children, in the U.S. alone, we are as a society conducting millions of perturbations of behavior every day. However, we do not take advantage of the enormous amount of empirical data that, in principle, could be collected and analyzed in the process. A tool for generating and mining such a data set could not only reveal general empirical facts and principles about how best to teach, or to prevent and treat disease, but also allow individuals to monitor their own personal parameters that govern how they best operate, empowering them to better themselves.

Consider the idea of an ongoing clinical trial. Currently a clinical trial for a drug involves, typically, a blinded test of a treatment versus a control, which lasts a certain amount of time, and progresses in multiple stages, increasing the number of people each time, and looking for certain outcomes. Then, if the trial ends successfully, the drug can be sold. However, it's been observed that a great many drugs likely work for only a fraction of the patients who receive them. Indeed, drugs that may be bad if prescribed indiscriminately are sometimes useful for specific subpopulations (e.g., consider the story of thalidomide). Furthermore, after a drug is out in the world, it can be used off-label by doctors. If side effects appear in a subpopulation of patients, there isn't a forum to interactively analyze the properties of that subpopulation in a rapid way. Clinicians can publish the results of such observations in journals, but such observations often stand alone.

A complementary approach might be to continually accumulate data about a drug as it is used to treat different diseases, in different populations, over time. Each individual patient would be permanently associated with a data point, so that follow-up and further examination would become possible. As genomic information, brain imaging, and other information-dense measures become increasingly cheap to acquire, tracking multiple variables within a patient over long periods of time will become more and more valuable, allowing one to find better predictors of future outcomes in response to a specific treatment, and to derive conclusions that would be impossible from a limited snapshot of a person's life. This could speed up the process of testing out technologies, allowing evidence to be accumulated and analyzed in a distributed and open fashion, and enabling cures to be developed and tested faster. It could also simplify prospective studies, in which patients are tracked before and after disease onset, say for conditions such as autism or schizophrenia; right now it is very hard to do this because detailed studies of people before a disease occurs are difficult for all but the most common diseases. With integrated database design and accessibility, it would become possible to perform this analysis. Such a system would also need to have instantaneous peer review that would occur in a rewardable way; the system must track real identities and real reputations of people who comment on or synthesize insights from the database, to synthesize accountability, reputation, and trust, and to separate the experts from the nonexperts. Perhaps free access to the database's wealth of data would motivate people to contribute; people who contribute less, or who contribute lower-quality judgment, might instead pay to access it.

It's possible that this methodology could apply to other domains of life, exploring how to assist people to become better--for example, consider how to evaluate trajectories for the approximately two million inmates in U.S. prisons. Or consider mental health, in which many styles of therapy are continually being explored by a diverse set of psychologists, therapists, and psychiatrists. Or the economy: perhaps a way to help economies self-regulate is to build in self-analysis at every step of the way, continuously generating models and testing theories to catch disasters before they happen.

How many approaches to life ever get validated? When does a strategy or method need to be personalized to an individual, and when is an insight a general piece of wisdom? Systems that enable these questions to be answered by providing a continuously updated snapshot of the best practices of the world may change the way we live, and enable a new age of rational decision making. "Those who can't remember the past are doomed to repeat it." Well, currently that's just about all of us.

Cite as: Boyden, E. S. "Civilization as Experiment" Ed Boyden's Blog. Technology Review. 1/18/09. (http://www.technologyreview.com/blog/boyden/22512/).

When I was a new student at MIT, there were legends of a math class in which the professor would occasionally assign an unsolved (and possibly unsolvable) problem. And every now and then, a student would resoundingly nail it. Soon after arriving at MIT, I was successfully spending my leisure hours inventing control algorithms for underwater robots, writing physics-based computer animation engines, devising new pattern-recognition algorithms, and building new kinds of NMR spectrometers. Now, more than a decade later, and being a professor myself, it's clear that some of the most valuable learning I did at MIT occurred during the solving of real-world problems. Simply put, in the Internet age, once you learn the basic core material, perhaps the best way to direct the growth of learning is to chase down real-world problems and fix them. You learn how to wrestle with failure, and how to get the resources you need.

Every now and then, it's useful to see how seriously one takes one's ideas. So let's take the above observation to its logical end: what if we decided that all work that students do in service of their education--problem sets, homework, exams--should be aimed at having a direct impact on solving a major current real-world problem? Please note: this doesn't at all imply the abandonment of learning of core things (calculus, physics, basic chemistry and biology, signal processing); it's just that a particular piece of homework might involve, instead of proving a discovery by Einstein right for the thousandth time, the solving of a piece, however small, of something unknown and important.

Clearly, this requires a mapping process--professors and teachers must parse real-world problems into decoupled chunks that can be addressed by individuals, while still enabling learning of the core materials. There are certainly some good examples of classes like this already. Lab classes at many universities exist in which students build medical devices, create computers, design virtual worlds, write business plans for ventures in developing countries, and learn how to make autonomous robots. Here I am wondering if, in addition, it would be possible to map real-world problems into the problem sets, homework, and exams for all the other classes--perhaps even introductory core classes. It's interesting to think about whether this might help humanity solve some outstanding problems. A back-of-the-envelope calculation: if 4,000 undergraduates at a university spent 40 hours a week during the school year solving problems that map onto real-world problems, that's more than 3,000,000 extra hours a year of inventing, design work, and creation, aimed at the problems that face humanity today. Multiply that times the number of universities engaged in such fields, and the new ideas and contributions to the world could be staggering. At MIT, undergrads do a lot of research. In my group, undergrads are here nights and weekends, even on busy school weeks, innovating incredibly novel inventions and conducting complex experiments. It is interesting to think about how that passion could be harnessed during the rest of their schedules.

An open question, though, is how much work it would take to map real-world problems into the thousands of smaller pieces that would be appropriate for classwork. And then to render them in engaging, interesting ways so that students will learn their core materials while they solve them. The new field of human-based computation is beginning to explore related questions. I was particularly intrigued by a recently released game that people can play to help solve questions in the field of protein folding--but many problems are not as clearly understood, or modular enough, to be broken into many subparts in such a way. It's possible that a discipline will need to arise around the analysis of really tough problems, and the breaking down of them into smaller parts. We also need to devise new and effective strategies to engage humans (with the assistance of computers) in the solving of such problems. It'll be interesting to see how far these ideas will scale in the years to come.

Thanks to Joost Bonsen for suggesting the title of this blog post.

_{Cite as: Boyden, E. S. "Inverting the Core." Ed Boyden's Blog, Technology Review. 7/15/08. (http://www.technologyreview.com/blog/boyden/22096/).}

Neurological and psychiatric disorders such as traumatic brain injury, stroke, Parkinson's disease, autism, depression, post-traumatic stress disorder, and chronic pain affect well over a billion people worldwide. These disorders steal away not only life span, but also our selves and identities. More than $1,000,000,000,000 is spent yearly in the battle against these disorders, even in the absence of effective treatments for many of them. Compared with innovations in other fields, like cancer, neurotechnologies have trickled out of labs at a relatively slow pace, yielding a handful of good drugs, a couple of methods for brain stimulation, and a few ways to image and analyze brain structure and activity. Like many innovations in medicine and bioengineering, these triumphs often emerged in no small part by chance, which makes iterative improvement tricky. Clearly, something new is needed. That's why, over the past year, we've begun experimenting with a hands-on neuroengineering curriculum at MIT, in which undergraduate and graduate students actively engage in the process of becoming neuroengineers, learning to solve intractable problems of the brain by actually doing it.

Learning neuroengineering is a hands-on process. What do you need to learn to fix problems of the brain and nervous system? The answer is, in brief: whatever it takes. The brain is complex (with well over a hundred billion interconnected circuit elements), subtle (it mediates everything we sense, feel, decide, and do), and inaccessible (packed densely inside the skull). To be a neuroengineer, you must be able to take advantage of any idea or fact that you discover that lets you get a handle on a brain process or function. Teaching neuroengineering thus means empowering people to identify problems and create solutions, connecting often distant topics in logical and intuitive ways to arrive at elegant insights. In short, our students must learn neuroengineering by making it up as they go along.

With my colleagues at MIT, I've begun teaching students how to go through the neurotechnology life cycle, from concept to validation to revelation to the world. We've concocted a series of three hands-on classes, which are still in the beta-testing stage, to teach design, laboratory, and entrepreneurship skills. Students pick projects and are mentored to make them as high-impact, feasible, and novel as possible. These classes are aimed at helping students learn the principles of operation of the nervous system from an engineering standpoint, implement their best ideas in the lab, and learn the process of translating technologies out of the lab and into the world. In the first class, Principles of Neuroengineering, students learn the basic principles governing the reading of information out from, and getting information into, the nervous system. They also, alone or in interdisciplinary teams, design and model fundamentally new technologies that gain information about, or positively alter, the operation of the brain. In the second course, Applications of Neuroengineering, students do lab work, learning how to implement, debug, and validate technologies. They make plans, revise them when failure encroaches, and learn how to find collaborators, make contingency maps, and manage time and resources. Finally, in the last course, Neurotechnology Ventures, students explore how to get their technologies out of the lab and into the world, writing up business-plan executive summaries and defending their projects in class, and attending guest lectures by entrepreneurs who are paving the way in neuroengineering. Anyone can participate--even freshmen can get involved. The ideas that yield the best neuroengineering inventions are often absurdly simple.

The classes started a beta-testing run in February 2007. Last year, in the Principles of Neuroengineering class, students designed never-before-seen methods for reading out brain activity in a wearable device, delivering therapeutic genes to specific cell types in the nervous system, and precisely measuring blood flow in the brain. Some of the students even built prototypes of their devices. For the Applications of Neuroengineering class, we just received a pilot grant from the MIT Alumni Class Funds to supply students with consumables, so that they can implement and validate their very best ideas in the lab, learning from failure and iteration. Students enter this class with concrete ideas, and get to make them reality. (We don't yet have a dedicated laboratory at MIT for teaching neuroengineering, so students who are safety- and procedure-certified to work in my lab can do their projects there. I try to help the rest find other collaborating labs on campus in which to work.) And in the first round of Neurotechnology Ventures, up to 50 people (including some professors) came to hear speakers talk about their companies (with post-talk discussions often lasting late into the night). Twenty students completed the key project, the creation of a concise business plan for a technology.

Although it's still the early days, perhaps this is the beginning of a Synthetic Neurobiology curriculum. Like many endeavors, this current set of classes has had a long, evolutionary path. Joost Bonsen and Rutledge Ellis-Behnke, my co-instructors in the Neurotechnology Ventures class, envisioned such a class almost half a decade ago. When I arrived at MIT in 2006, I was deluged by e-mails from undergraduates and graduate students eager to enter the business of engineering the brain and mind. The time had come. But our work is only beginning. We are still revising our educational vision daily, as we define the abstraction layers for engineering the brain. In the long term, I will measure the success of this mission by the number of laboratories, companies, inventions, and, ultimately, cures that are accomplished by people who pass through this class. Someday, we will understand the brain and know how to fix its problems. But for now, we must focus on jump-starting this effort by encouraging direct action by the best minds in the world, at an intellectual scale that exceeds all that has come before.

Numerical data in the first paragraph is from a recent report by NeuroInsights, LLC.

_{Cite as: Boyden, E. S. "Teaching Neuroengineers." Ed Boyden's Blog, Technology Review. 4/21/08. (http://www.technologyreview.com/blog/boyden/22055/).}