Darwin Among the Machines
Page 24
According to a 1995 estimate by the International Telecommunications Union, 2.3 trillion dollars circulates electronically every twenty-four hours—equivalent to 180,000 tons of gold, or a 1,500-mile stack of hundred-dollar bills. Electronic currency has diffused outward from the central banking networks to penetrate the street corner, the desktop, the telephone system, and a host of card-based payment systems, smart and dumb. Banks are becoming networks, and networks are becoming banks. Having seen corporate mainframes replaced with desktop computers, some analysts believe the powers of large banking institutions will be similarly overturned. But the banks are here to stay. “Commercial banking has been around over 600 years,” consultant Eric Hughes has explained. “Computers are less than 60 years old. Microcomputer software companies are 20 years old and still reinvent the wheel. Assuming a convergence, who do you think will learn the other’s business first?”35
The result will be more money, faster money, and money more tightly coupled to things, network architecture, people, and ideas. The scales are shifting both in distance and in time; the intelligence a large corporation once gathered for its annual report is now available to any small business using a personal computer to manage its day-to-day accounts. “We felt that the distinction between micro- and macro-economics, while appropriate in a non-computer age, was no longer necessary,”36 remarked economist Gerald Thompson, recalling his final collaboration with Oskar Morgenstern in 1975, two years before Morgenstern’s death.
Money is a recursive function, defined, layer upon layer, in terms of itself. The era when you could peel away the layers to reveal a basis in precious metals ended long ago. There’s nothing wrong with recursive definitions. (Definition of recursive: see recursive; or, Gregory Bateson’s definition of information as “any difference that makes a difference”—the point being that information and meaning are self-referential, not absolute.) But formal systems based on recursive functions, whether in finance or mathematical logic, have certain peculiar properties. Gödel’s incompleteness theorems have analogies in the financial universe, where liquidity and value are subject to varying degrees of definability, provability, and truth. Within a given financial system (i.e., a consistent system of values) it is possible to construct financial instruments whose value can be defined and trusted but cannot be proved without assuming new axioms that extend the system’s reach. As Gödel demonstrated for logic and arithmetic, there are two sides to this. No financial system can ever be completely secure and closed. On the other hand, like mathematics or any other sufficiently powerful system of languages, there is no limit to the level of concepts that an economy is able to comprehend.
All free-market economies show signs of intelligence, to varying degrees. Conversely, close inspection of many mechanisms we regard as intelligent reveals fundamentally economic systems underneath. When Oskar Morgenstern was asked to explain the power of game theory to a popular audience in 1949, he used the example of a simplified form of poker for two players, using a three-card deck and a one-card, no-draw hand.37 To determine all possible strategies for this game by brute-force computation requires two billion arithmetic operations. Simple economic systems are able to arrive at practical solutions to problems that are computationally difficult to solve. That brains in nature operate more as economies than as digital computers should come as no surprise. Indeed, economic principles are the only known way to evolve intelligent systems from primitive components that are not intelligent themselves. As Marvin Minsky explained in his Society of Mind: “You can build a mind from many little parts, each mindless by itself. . . . Any brain, machine, or other thing that has a mind must be composed of smaller things that cannot think at all.”38 Or, as Samuel Butler put it in 1887: “Man is only a great many amoebas, most of them dreadfully narrow-minded, going up and down the country with their goods and chattels.”39 The archetypal invisible hand of Adam Smith (“He intends only his own gain, and he is in this, as in many other cases, led by an invisible hand to promote an end which was no part of his intention.”)40 appears to be capable of building not only an economy, or a damage-resistant communications network, but a brainlike structure, perhaps a mind. “Probably the closest parallel structure to the Internet is the free market economy,” observed Paul Baran.41
The incubation of intelligence within a network requires an exceptionally fluid, arborescent structure and the infiltration of this architecture by a statistical language analogous to the primary statistical language that von Neumann identified as the machine language of the brain. At one level, this language may appear to us to be money, especially the new, polymorphous E-money that circulates without reserve at the speed of light. E-money is, after all, simply a consensual definition of “electrons with meaning,” allowing other levels of meaning to freely evolve. Composed of discrete yet divisible and liquid units, digital currency resembles the pulse-frequency coding that has proved to be such a rugged and fault-tolerant characteristic of the nervous systems evolved by biology. Frequency-modulated signals that travel through the nerves are associated with chemical messages that are broadcast by diffusion through the fluid that bathes the brain. Money has a twofold nature that encompasses both kinds of behavior: it can be transmitted, like an electrical signal, from one place (or time) to another; or it can be diffused in any number of more chemical, hormonelike ways.
Money has the self-reinforcing tendencies and semantic transparency that allow neural networks to work. The flow of money permeates all components of the network, strengthens frequently used connections, propagates backward, transforms local processing mechanisms, and encourages new connection pathways in response. This architectural plasticity allows neural networks to adapt, remember, and learn to predict events. Freely reversible financial gradients direct when and where new connections are formed and which connections die out. The flow of currency transports, integrates, and accumulates signals; a myriad of financial instruments function as neurotransmitters and bridge synaptic gaps.
“Neural processes are insulated from the extra-cellular fluid by a membrane only approximately 50 angstroms thick,” wrote semiconductor pioneer Carver Mead, explaining how to build integrated circuits modeled after the neural circuits found in our brains. “The capacitance of this nerve membrane serves to integrate charge injected into the dendritic tree by synaptic units. Much of the real-time nature of neural computation is vastly simplified because this integrating capability is used as a way of storing information for short periods—from less than 1 millisecond to more than 1 second. There is an important lesson to be learned here, an insight that would not follow naturally from the standard lore of either computer science or electrical engineering. Like the spatial smoothing performed by resistive networks . . . temporal smoothing is an essential and generally useful form of computation.”42 Whether conveyed by bullion or binary numbers, accounts accumulate incoming currency over various periods and release outgoing currency at intervals more or less closely related to patterns generated by the currency coming in. In an age when nanoseconds count, it is easy to forget that the components of a neural net must have some temporal delay, however small, to allow the network to compute.
In drawing these analogies, what of the data that now flood the telecommunications net: pictures, sound, video, interactive data communications, encyclopedias of text? All this traffic means something to somebody, and some of it advances our sciences, our culture, and our arts, but is it the stuff of meaning (or the measure of a utility function) across the system as a whole? Maybe or maybe not. What counts is not so much the data that flow in any given direction, but the money that flows the other way. In the coalescence of the software, banking, and telecommunications industries, we are spawning the precursors of collective digital organisms that will roam the network like social insects, sending packets of digital currency back to their nests. The push toward interactive communications over the Web is aimed not at delivering content to the consumer (this can be done already), but at delivering mon
ey, in real time, the other way. Electronic money allows organizations to do things and immediately sense the results.
This was the original premise of purposive systems as expounded by Norbert Wiener and Julian Bigelow in 1943: intelligent behavior evolves as a consequence of the ability to measure and keep account of the effects of a given signal through feedback loops that return a message signifying the magnitude of the result. These principles are common to automatic anti-aircraft guns firing at a moving target, neurons seeking to make the right connections inside a brain, laboratory animals facing a maze, corporations facing a free-market economy, or any other situation where it is possible to place a value on an objective at which to aim.
The goal of life and intelligence, if there is one, is awkward to define. A general aim can be detected in the tendency toward a local decrease in the entropy of that fragment of the universe considered to be intelligent or alive. This is a measurable way of saying that life and intelligence tend to organize themselves. Order, however, is only available in limited quantities, at a certain price. Organization can be increased or created only by absorbing existing sources of order (by eating other creatures as food, joining them in symbiosis, or by photosynthesis exploiting the ordered energy of the sun) or by shedding disorder (by excreting waste, radiating heat, or learning from experience through the attrition of less-meaningful connections in the developing infant brain). In human society, money serves to measure and mediate local markets for decreasing entropy, whether it measures the refinement of an ounce of gold, the energy available in a ton of coal, the price of a share in a multinational organization, or the value of the information accumulated in a book. We invented the science of economics, but economy came first.
In 1965, twenty years after the disbanding of Alan Turing’s crew at Bletchley Park, Irving J. Good published his speculations on the development of an ultraintelligent machine, later described as “a machine that believes that people cannot think.”43 Central to the development of an indisputable mechanical intelligence is the question of what meaning is and how meaning is evolved. In Good’s analysis, meaning and economy are deeply intertwined; where there is meaning, there is an economy of things representing information (or information representing things) by which the meaning of things can be evaluated and from which meaningful information structures can be built. “The production of meaning can be regarded as the last regeneration stage in the hierarchy,” noted Good, “and it performs a function of economy just as all the other stages do. It is possible that this has been frequently overlooked because meaning is associated with the metaphysical nature of consciousness, and one does not readily associate metaphysics with questions of economy. Perhaps there is nothing more important than metaphysics, but, for the construction of an artificial intelligence, it will be necessary to represent meaning in some physical form.”44
In 1677, William Petty, in a letter to his cousin Robert Southwell on “The Scale of Creatures,” wrote that “between God and man, there are holy Angells, Created Intelligences, and subtile materiall beings; as there are between man and the lowest animall a multitude of intermediate natures.”45 Whether he saw economic systems as among these created intelligences is left unsaid. By the time of Alfred Smee, the forest was obscured by the trees. Proposing, in 1851, that mechanical processing of ideas would require a relational and differential machine the size of the City of London, Smee failed to notice from his quarters on Threadneedle Street that the Bank of England’s network of linked transactions, mediated by a hive of accountants, already constituted such a machine. “The average daily transactions in the London Bankers’ Clearing House amount to about twenty millions of pounds sterling, which if paid in gold coin would weigh about 157 tons,” reported Stanley Jevons in 1896.46
John von Neumann, although halted in midstream, was working toward a theory of the economy of mind. In the universe according to von Neumann, life and nature are playing a zero-sum game. Physics is the rules. Economics—which von Neumann perceived as closely related to thermodynamics—is the study of how organisms and organizations develop strategies that increase their chances for reward. Von Neumann and Morgenstern showed that the formation of coalitions holds the key, a conclusion to which all observed evidence, including Nils Barricelli’s experiments with numerical symbioorganisms, lends support. These coalitions are forged on many levels—between molecules, between cells, between groups of neurons, between individual organisms, between languages, and between ideas. The badge of success is worn most visibly by the members of a species, who constitute an enduring coalition over distance and over time. Species may in turn form coalitions, and, perhaps, biology may form coalitions with geological and atmospheric processes otherwise viewed as being on the side of nature, not on the side of life.
Coalitions, once established, can be maintained across widening gaps, such as the levels of abstraction that separate the metaphysics of a language from the metabolism of its host. Fortunes shift, and if a symbiont develops a strategy that dominates the behavior of its host, the roles may be reversed. Our own species is doing its best to adjust to a three-way coalition of self-reproducing human beings, self-reproducing numbers, and self-reproducing machines. Signs of intelligence are evident at every turn, but because this intelligence envelopes us in all directions the whole picture lies beyond our grasp. We have made only limited progress in the three hundred years since Robert Hooke explained how the soul is somehow “apprehensive” of “a continued Chain of Ideas coyled up in the Repository of the Brain.”47 What mind, if any, will become apprehensive of the great coiling of ideas now under way is not a meaningless question, but it is still too early in the game to expect an answer that is meaningful to us.
10
THERE’S PLENTY OF ROOM AT THE TOP
We’re doing this the way you’d plan walkways in a park: Plant grass, then put sidewalks where the paths form.
—JOE VAN LONE1
“There’s Plenty of Room at the Bottom” was the title of an after-dinner talk given by physicist Richard Feynman at the California Institute of Technology on 29 December 1959. Feynman’s timing was perfect. He kept his audience awake with a series of outlandish speculations that soon turned out to be spectacularly right. “In the year 2000, when they look back at this age,” announced Feynman, “they will wonder why it was not until the year 1960 that anybody began seriously to move in this direction.” Imagining small machines being instructed to build successively smaller and smaller machines, Feynman estimated the orders of magnitude by which such devices could become cheaper, faster, more numerous, and collectively more powerful. Molecules, and eventually atoms, would supply mass-produced low-cost parts.
“Computing machines are very large; they fill rooms,” said Feynman. “Why can’t we make them very small, make them of little wires, little elements—and by little, I mean little. For instance, the wires should be 10 or 100 atoms in diameter, and the circuits should be a few thousand angstroms across.” Besides all the other good reasons to avoid building computers the size (and cost) of the Pentagon, Feynman pointed out that “information cannot go any faster than the speed of light—so, ultimately, when our computers get faster and faster and more and more elaborate, we will have to make them smaller and smaller.
“How can we make such a device? What kind of manufacturing processes would we use?” Feynman asked. “One possibility we might consider, since we have talked about writing by putting atoms down in a certain arrangement, would be to evaporate the material, then evaporate the insulator next to it. Then, for the next layer, evaporate another position of a wire, another insulator, and so on. So, you simply evaporate until you have a block of stuff which has the elements—coils and condensers, transistors and so on—of exceedingly fine dimensions.”2
Feynman did not limit his speculations to electronic microprocessors, however intriguing or lucrative these prospects might be, but continued on down to atom-by-atom manufacturing, “something, in principle, that can be done; but
, in practice, has not been done because we are too big.” He greeted the implications with an enthusiasm “inspired by the biological phenomena in which chemical forces are used in a repetitious fashion to produce all kinds of weird effects (one of which is the author).”3 He left other even weirder effects unsaid. Many of Feynman’s techniques are now in routine use, the convergence between microbiology and microtechnology steadily eroding the underpinnings of distinction between living organisms and machines. No new laws of physics have turned up to render his predictions less probable than they were in 1959.
Yes, there is plenty of room at the bottom—but nature got there first. Life began at the bottom. Microorganisms have had time to settle in; most available ecological niches have long been filled. Many steps higher on the scale, insects have been exploring millimeter-scale engineering and socially distributed intelligence for so long that it would take a concerted effort to catch up. Insects might be reinvented from the top down by the miniaturization of machines, but we are more likely to reinvent them from the bottom up, by recombinant entomology, for the same reasons we are reengineering existing one-celled organisms rather than developing new ones from scratch.
Things are cheaper and faster at the bottom, but it is much less crowded at the top. The size of living organisms has been limited by gravity, chemistry, and the inability to keep anything much larger than a dinosaur under central-nervous-system control. Life on earth made it as far as the blue whale, the giant sequoia, the termite colony, the coral reef—and then we came along. Large systems, in biology as in bureaucracy, are relatively slow. “I find it no easier to picture a completely socialized British Empire or United States,” wrote J. B. S. Haldane, “than an elephant turning somersaults or a hippopotamus jumping a hedge.”4