Thursday, May 13, 2010

Decentralized Learning

As an undergraduate, one of the most thought-provoking classes I ever took was a class on the philosophy of Paul Churchland. His eliminative materialist philosophy challenged my traditional thinking, especially as I come from a Christian background, yet his whole approach to philosophy has captured my imagination and continues to shape my thinking.

What has always stuck with me is Churchland's goal of merging scientific inquiry about the brain with philosophical inquiry about the mind. Modern science has made that possible. Whereas before neuroscience came into its own, philosophers generally engaged in the philosophy of mind merely by, well, thinking.

Perhaps Churchland's most challenging idea about the mind is that it really doesn't "contain" the beliefs and ideas that we talk as if we "have." This is a direct consequence of looking at the mind as equivalent to the brain, and asking, how does the brain work? The brain, as it turns out, is a complex network of cells, each with a set of connections to other cells, with each of those connections adjusting its "weight" over time due to a series of interactions between cells. The upshot is that what we know and believe is really embedded in a set of weights that describe this neural network at a given time.

Think of it as a market. In a market, people freely exchange goods with one another. You can think of people in the market corresponding to cells in the brain, and economic exchanges as the connections between cells. The state of the market is described by all the exchanges that are taking place. The efficiency of the market is directly analogous to the efficiency or health of the brain.

In both pictures, all we can really say is that a complex network is in a certain "state" in a multi-dimensional "state space." We can never tell the whole story by boiling it all down to a collection of thoughts or ideas (in the case of the brain) or a bottom line GDP per capita or some other economic measure (in the case of the market).

How can a complex network that behaves according to an unpredictable series of exchanges be said to "learn" anything? This is where our intuition fails us, but if we challenge ourselves to think a little more abstractly about ourselves, we find some startling insights. Here is what Churchland says about learning and conceptual change in Chapter 11 of A Neurocomputational Perspective:
To cognitive creatures, the world is a highly ambiguous place. not just in the ambiguity it presents to our sensory systems, where the initial coding is typically consistent with a diversity of external circumstances, but more profoundly in the ambiguity it presents to our conceptual systems. Any conceptual framework, no matter how robust or natural its categories may seem to us, is but a single point in a practically infinite space of alternative possible frameworks, each with a comparable a priori claim on our commitment. ...

This talk of a vast space of alternatives is not merely romantic. Each of us has a history of conceptual diversity already. For you were not born with your adult framework. You came to it slowly, through a long period of development. There is indeed a space, through which each of us has a complex journey already completed. ...

Talk of conceptual "space" may seem metaphorical still, but as outlined in the two preceding chapters, recent research has shown us how to make literal and very useful sense of it. If we assume that the human brain is a multilayered network of interconnected units, we can uniquely specify its current position in conceptual space by specifying the individual strengths or weights of its myriad synaptic connections.... That configuration of weights can be directly represented by a specific point in a multidimensional space, a space with a distinct axis for each of the brain's 10^14 dimensions with at least 10 possible positions along each. Its volume is almost unimaginably vast--at least 10^(10^14) functionally distinct positions--as our guiding metaphor suggested.

That number, 10^(10^14), is far too large to fathom. 10^14, by itself, is 1 followed by 14 zeros, as in 100000000000000. So the number we're talking about is 10^100000000000000, or 1 followed by 100000000000000 zeros. I'm not going to type that out for you.

We find ourselves hopelessly incapable of contemplating the vast realm of possible states which our own brain may explore. Yet it is precisely by exploring that incomprehensible space that our brains learn. We develop categories by which to understand the world around us, we develop ideas that help us predict events in our lives, and we develop moral beliefs and attitudes that help us decide what to do next.

This has a very troubling yet delightful consequence: I am not really an "I." Although intuitively I would like to believe that there is a singular force or will driving my mind, such is not the case (if smart men like Churchland are to be believed). What we see instead is that I am the product of a series of unpredictable exchanges in a complex network. There is no central authority in this network; in essence, there is no "I" except what has arisen out of this series of exchanges.

Intuitively, I find this troubling. Surely there is an "I" lurking about somewhere, driving the machine. The image of a reasonable soul driving the body is a long-standing one that will not quickly go away. However, scientific study of the brain has provided me with an opportunity to rethink my understanding of myself in an infinitely more interesting way. I am not singular, but plural.

When I say plural I emphatically do not mean democratic. In no way do the cells of the brain cast ballots and vote on what the brain as a whole will believe. To think this would be missing the point. There is no belief contained somewhere in the brain. Beliefs are mere "cross sections" of a state in a vast state space. When I say I am plural, I mean rather who I am, what I think, and how I approach life is the sum of a series of exchanges that I neither planned from the beginning nor fully understand or control now. It is as if my mind is a marketplace, and I am merely the result that some "invisible hand" has created out of a complicated series of economic exchanges.

But if I am troubled at the thought of thinking of myself in this decentralized way, it does not surprise me in the least if we as a society tremble at the thought of thinking of our economy in a decentralized way. The desire to have centralized planning of our complex economy is probably related to the desire to believe that I have a centralized will that drives the complex network that constitutes my own mind.

To be very concrete, consider the desire for universal health care. In our day, this is commonly called a very "collectivist" goal, with the implication being that we seek the good of the whole rather than of individuals. We put our trust in politicians to carry out our collective desire to give everyone health care, because we believe that the whole society ought to be driven in part by this single purpose. In other words, there has to be some singular will driving the whole thing.

The fact that we come up with this democratically leads us to believe that we are meeting the needs of the whole society. But this is not how the brain works, and neither is it how society works most effectively. In the end, it is not democratic decisions that are most effective, but rather decentralized decisions. The latter are not as readily accepted by society, because we can't see them as directly or concretely define them as easily as the former. Yet a society that grows in a decentralized way harnesses infinitely more learning power than it ever could if driven by some central (albeit democratic) authority. (This idea is not original to me; Friedrich Hayek said much the same thing. Incidentally, Hayek was also trained as a neuroscientist.)

For many people, the thought of a truly decentralized society, in which people enjoy the freedom to make private economic decisions rather than be guided by public opinion, is scary, because it seems like a purposeless society. It has no central goal; how can such a society have a meaningful collective life?

Yet that is precisely the life of the human mind. It is not given some central goal to pursue from birth. It is given an incomprehensibly vast state space to explore as it attempts to interact with the world around it. The result is a human person--not a static, well-defined concept, but a dynamic personality which is never fully predictable.

The same can be said of a free society. While it does not appear to be given some overarching goal, yet it can be truly said to have some collective life. I reject the conservative notion that the individual is more important than the collective. I think the collective is vitally important, but I also believe that the way in which we truly grow as a collective is through individual freedom, allowing our society to grow through complex interaction between its constituent members.

In essence, a free society can be said to "learn" in a way that a socialized society cannot. Socialism depends on central planning, which leaves no room for real learning. The reason today that the average person cannot imagine a computer learning is that most of the computers we use are based on central planning: an engineer designs the machine to work according to a certain set of instructions, and the machine must faithfully carry out those instructions. However, the reality is that many computers out there can and do learn, at least in a primitive sense. This is because we have begun to design parallel processing systems which mimic the complex neural networks found in the brain.

There are many people today who wish that some centralized governing agency would engineer society in a way that we could follow the instructions exactly and achieve a desired outcome. Thus they clamor for more regulations of the market, more government spending, universal health care, and so on.

Then there are those of us who desire a free society, because we understand that real societal learning takes place without central planning. Not all of us may be able to accept the idea of a decentralized self; but at least we are able to accept the idea of a decentralized market, in which human beings acting in their own self interest actually work to the benefit of the collective.

I have not even begun to treat these ideas with sufficient care, but maybe this blog post is a start.

No comments:

Post a Comment

I love to hear feedback!