by James Bailey |
John_Dewey 180x247

After last week's demoralizing encounter with Marcus Kello’s Virtual Weapon book, what a joy to get Alan Jacobs’ How To Think. He is like a 21st-century John Dewey, modest but trenchant, grounded in the real world but relentlessly well-read, aware of the magnitude of his task but committed to press ahead anyhow. And all in one slim volume instead of forty fat ones.

Self schooling is attracted to Prof. Jacobs because we share two convictions:

  • Habits of thought are sneaky important
  • Computers change the thinking game

Where Self Schooling focuses a lot on the ways computers enable new algorithms, Prof. Jacobs focuses on the social media that is also enabled by our computers. He is very clear about the ways that they can facilitate an us-versus-them way of thinking, at great cost both to us individually and to society:

"

The cold divisive logic of the [Repugnant Cultural Other] impoverishes us, all of us

If we look more closely at the argument-as-war metaphor, we’ll see that it depends on a habit of mind that is lodged very deep

"

Habits of thought also come packaged as myths and metaphors. The book quotes Mary Midgley:

"

Myths are not lies. Nor are they detached stories. They are imaginative patterns, networks of powerful symbols that suggest particular ways of interpreting the world

"

Like Dewey, who ran in elementary school, Prof. Jacobs is forced to test his ideas against the hard rock of reality, in his case the grading of 15,000 essays in the course of his college teaching career. This real-world experience has given him quiet confidence (hope?) that messed-up habits of thought can be addressed at the college level:

"

At every stage habits become more deeply ingrained in us, habits that inhibit our ability to think. We can only hope that there are strategies by which we might counteract the force of these habits—and develop new and better ones.

"

These strategies form the heart of the book and they need to be read in his own words. They certainly lower the boom on Self Schooling's own no-presence-on-social-media posture. While interacting this way can be a full-time job, Prof. Jacobs insists that there is no such thing as isolated thought: no interaction, no thinking.

Self Schooling, however, sides with Dewey on the timing of habits of thought. While Mr. Jacobs offers inspired strategies for correcting bad habits of thought among young adults, can we not use our computers to help install good ones in the early years so we don't have to fix them later on? Self Schooling’s current focus on Deep Learning offers one example of how our efforts might come together.

Prof. Jacobs, like Dewey, notes the constant tension between being open to new ideas and having some starch.

"

We don't want to be, and we don't want others to be, intractably stubborn, but we don’t want them to be pusillanimous and vacillating either.

"

Turns out that Deep Learning algorithms face this same conundrum. Algorithm designers call stubbornness "bias" and vacillation "variance." Basically, a biased algorithm doesn't pay enough attention to the data and wants to give the same answer to every question. ("The answer is yellow. Now what was the question?") The high variance algorithm contorts itself in order to suck up to every piece of data it sees.

Is a Deep Learning algorithm an accurate analog of the human mind? Unlikely, but imagine that a middle schooler could see a learning algorithm play out with high bias and then with high variance. They would then have a concrete hook on which to think about their own thinking. Am I intractably stubborn like that high bias thing? Or am I wishy-washy like that high variance one?

But wait. There's more. Deep Learning mavens detect high variance by the way the algorithm does a great job on the data it has been trained on, but then flunks on new examples. It cannot apply what it has learned to new situations. Graduate schools of education have a technical term for this situation. It is called teaching to the test and it does not send kids out into the world well prepared for novelty.

So what do deep learners do to fix high variance? The answer is right in How To Think. You turn down the damn volume knob. Deep Learning—and arguably our own—happens by netting out conflicting guesses as to what is going on. In our own brains, the neurons spike away furiously as they make their case. In computer algorithms, nodes raise their weightings to try to influence the result. The result is a mess. So Deep Learning algorithms penalize the shouters by a process called "regularization." Those who want to see the same quieting happen in our society, or in their own selves, will be nourished by Alan Jacobs’ How To Think.

tags:

Comments