There’s a very nice post at IEET by Marcelo Rinesi entitled Education and Learning: Still in the Middle Ages. He points out that we’re pretty damn bad at education compared to the improvements we’ve seen in most other endeavors:

Our lecture halls are better than those of the Middle Ages, our textbooks friendlier than those of early 20th century, and the Internet often a good replacement for an afternoon in a library. We even use, if we are good or lucky in our choice of educational system, exercises, simulations, and concrete projects. It could even be said that, with all these resources and technology at our disposal, we are somewhat better learners, in average, than we used to be. Quite a bit better, probably, than when the Middle Ages finished and Western civilization rediscovered the scientific method.

Therein lies the problem. Because in almost every other economically important activity, our performance isn’t somewhat better or quite a bit better, but much better, if not astoundingly so. … They waged war, but nothing at all like we do. They built (and beautifully, it must be said), but their engineering was nowhere close to ours. They would consider our doctors wizards, and would shake their heads when we told them about our telescopes orbiting Earth, peering into the residual radiation from literally the first microsecond of time.

But they would understand our schools and universities. They would be surprised at their size and the abundance of books, but they would nod at the teacher lecturing, the students at the library, the concept, if not the precise form, of the exam. It would be to them an inspiring sight, perhaps, but not an astounding one. … We have gone from Leonardo’s drawings of leather wings to routine flights between Rome and Tokyo, and from the faithful students with their books, their teachers, and their notes… to the same place.

It seems clear to me that we haven’t drastically improved learning because we don’t have a clue how learning works. This, as I have said on many occasions, is really the big roadblock between us and true AI — we know how to build machines that have skills, but we don’t know how to build machines that learn skills.

What are the figures of merit that would characterize an educational technology as advanced over medieval as the jetplane is over the oxcart?  Three orders of magnitude faster just about captures it.  In other words, our entire standard course, primary through a Master’s degree, ought to take a week.

In the story Profession (found in Nine Tomorrows), Isaac Asimov gives what is probably the classic science-fiction account of the teaching machine:  At age 8 you have Reading Day, when you go into an institution of some kind, they put a helmet on your head with lots of wires attached, it buzzes for 15 minutes, and you can read.  You then enjoy a carefree (in particular, school-free) childhood for ten more years until Education Day, wherein the process is repeated and you’re programmed to be an expert at whatever experts are most needed at, and you proceed to your adult career.

Accepted theory says that the bandwidth into human long-term memory is on the order of one bit per second.  Learning rates in schools fit — there’s about a megabit in a book, and about half a million seconds in a 3-hour-per-week one semester course plus homework time. It’s become understood that long-term learning makes significant physical changes in your brain; dendrites grow, synapses are formed.  In the absence of some really major invasive reconstruction by nanobots or the like, it’s going to be hard to improve that natural bandwidth bu more than a small factor.

What’s the alternative?  One good possibility is that we learn to learn more efficiently.  Example: never learn a fact or skill a computer can do for you.  Have instead a built-in appliance — actually an AI smarter than you are and connected to all the world’s knowledge — that can remember things and show them to you by direct neural connection.  Your brain goes from being the whole company to being just the upper management.  What you do learn is now management techniques for a much more powerful labor force than your wetware neurons.

This won’t extend infinitely, but I’d bet it’s good for three orders of magnitude.

By | 2017-06-01T14:05:25+00:00 July 30th, 2009|Complexity, Machine Intelligence, Nanodot|2 Comments

About the Author:


  1. Al Neunzoller July 30, 2009 at 4:00 am - Reply

    Very interesting and inspiring post.
    The 1bps bandwidth of long-term memory is an idea that I have always disagreed with. Consider that the average human child needs an average of 12 hours sleep per day until 5 years of age. We know that new information is not absorbed during sleep, so this means that even if a child were to learn for a full 12 hours per day, only 5.2 kilobytes of information would be absorbed. This is equal to 9.5 megabytes for 5 whole years, assuming that conditions were favorable and that the child starts learning immediately after birth. According to our experience with AI, the idea that the entire range of social skills, language (an average vocabulary of over 2500 words!) and relationships between concepts that a 5 year old has learned could be translated into just 9.5 megabytes is nothing short of insane.

    The key is that the child has many things to help it learn – much of the subcortical circuitry is devoted just to this task – and can absorb information by many channels simultaneously. The typical family setting (mother, father, optional siblings, each interacting with each other and the child in ways similar across different families and cultures) is, as it turns out, the optimal learning environment for a young child. Indeed, we have evolved for this purpose. However, soon after childhood, this system loses steam, simply because at 13 or 14 humans are ‘fully-grown’ as far as living in the natural world is concerned. Indeed, in many cultures it is not uncommon for a 10-year-old girl to start living independently and start raising a family. Our modern culture, however, has evolved needs that require highly specialized knowledge that we simply don’t know how to teach effectively, resulting in (sometimes) decades-long education systems. Evolution has simply not had enough time to evolve optimal learning environments for this kind of knowledge. Yet the earlier learning mechanisms are still there.

    I think that by in-depth study of neural systems we could engineer teaching systems that could teach at far higher rates than many would think possible, while at the same time putting little pressure on the learning individual. In the future, an advanced thermodynamics course might be as simple and carefree as childhood.

  2. Susan August 2, 2009 at 9:22 am - Reply

    The famous American philosopher and educator, John Dewey, wrote:
    “It (the technique of Mr. Alexander – known today as the Alexander Technique) bears the same relation to education that education itself bears to all other human activity.” My own experience with Alexander lessons certainly bears this out.

    You can learn more about the Dewey-Alexander onnection at and about the Alexander Technique itself at

Leave A Comment