The Transhuman Taste

Reviews of Extropian Interest

Theories of Everything: The Quest for Ultimate Explanation

by John D. Barrow

Ballantine Books, New York, 1991

Paperback $12.00, ISBN 0-449-90738-4

Reviewed by Peter McCluskey

Theories of Everything is a good exploration of the most fundamental questions of physics, mathematics, and why we exist. It contains a number of the fascinating ideas that were discussed in The Anthropic Cosmological Principle (ACP), but is simple enough that it can be understood by people with no mathematical background (it has about ten simple equations, versus enough tensor equations in ACP to scare off most non-physicists). It has enough new ideas so that people who have read ACP will still find significant parts of it interesting.

This is the book that Hawking’s Brief History of Time should have been. It describes attempts to unify forces, particles, space, time, and the initial conditions of the universe into a single coherent explanation, possibly even a single equation. The term “theory of everything” may be a somewhat exaggerated way to describe a unified theory encompassing the essentials of physics, since it would no more explain the complete configuration of the universe than a perfect theory of

thermodynamics would enable us to reliably forecast next year’s weather. All it would take is a little bit of chaos to make prediction intractable.

Barrow summarizes the history of comprehensive theories, such as ancient creation myths, and Roger Boscovitch’s Theory of Natural Philosophy (1758), which proposed a unification of the gravitational, electrical and magnetic forces.

Barrow recounts the historical separation between laws describing the evolution of the universe and the initial conditions that are treated as arbitrary inputs to the equations in those laws, and describes hints that a theory of quantum gravity would abolish that distinction, along with the distinction between time and space at the beginning. Since time is not an explicit component of quantum theories of cosmology, time can be regarded as a phenomenon which applies only to some parts of the universe. Closer to what we think of as the origin (English is not well

suited for handling directions when they are not readily divided into space and time), a model of 4 spatial dimensions could be a more useful way of describing things. In this case, the problem of explaining the “beginning” of the universe fades away, as there is no starting point to such a universe.

In ACP, Barrow and Tipler had appeared to prove that the universe had a finite past by showing that if intelligent life had ever existed in our infinite causal past, it would have taken over all that we can see. Barrow now describes a theory of an “eternal” inflationary universe in which regions of a universe which is infinite in both time and space can undergo an inflationary expansion of the kind that is postulated to have occurred shortly after the “big bang” in order to explain the large-scale uniformity of the visible part of the universe. Each such expanded region sees a dense big-bang-like past. As long as life cannot evolve in the superhot unexpanded

EXTROPY #11 Summer/Fall 1993

42

regions of the universe, it can exist in infinitely many inflated regions without being able to overrun newly created ones because the rapid expansion in the early phases of each region insures that the lightspeed barrier isolates each region.

If even the infinite spatial part of this idea is correct, then our search for universal theories is seriously hampered by the likelihood that the visible portion of the universe has been selected for its ability to evolve life, and is therefore probably very unrepresentative of the conditions that a truly general theory would say are typical. So what appear to be universal laws may be special cases of something much more general.

There are some numbers such as the ratio of the proton mass to the electron mass, which appear at first glance to be totally arbitrary, and therefore impossible to reduce to a principle derived from some more basic law. One of the ways they could have been determined, Barrow suggests, is by the effects of numerous small wormholes (little larger than the Planck length) connecting our region of the universe with distant parts of itself or with other regions which would otherwise be

independent universes. The effects of quasi-random connections of this nature would reportedly be to introduce random shifts in the values of the “constants” observed within each “normal” region. One example is the cosmological constant (a term Einstein added to General Relativity to allow the theory to predict a stationary rather than expanding or contracting universe), which is expected to be extremely close to zero for most possible wormhole interconnections.

Barrow expresses concern about the possibility that a Theory of Everything could be shown to be the only logically possible theory, and thus be an analytic truth, rather than a scientifically falsifiable truth. I find this to be disappointing, as deductively proven truths have withstood the test of time better than theories which are subject to falsifiability. While I doubt that such an analytic truth will be found, I consider falsifiability to be a crutch to use when nothing better is available, and want analytic truths to replace unprovable theories wherever possible.

In Our Own Image:

Building an Artificial Person

by Maureen Caudill

Oxford University Press, 1992

242 pp hardcover; $22.00

ISBN 0-19-507338-X

Reviewed by Derek Zahn

Books on Artificial Intelligence (AI) written for general audiences usually proceed either historically — tracing the development of the major AI ideas by focusing on the scientists responsible for them — or by presenting a case for the centrality of particular techniques or problems. In In Our Own Image: Building an Artificial Person, Maureen Caudill takes a different path; using the popular media image of The Android as a goal to be reached, she explores the problems that must be solved to achieve that goal. The exploration visits most of the research areas of AI, resulting in a satisfying and readable summary of the state of the art.

The book has two sections; the first concentrates on techniques and research; the second looks at more philosophical issues. Of these, the first is far superior. She focuses in turn on the different capabilities that an android must have—sight, movement, memory, reasoning, speech, etc, and evaluates the methods that AI has developed to give machines those capabilities.

Despite the broad scope of such a project, in less than 200 pages Caudill covers the territory skillfully, from the early insights into reasoning and logic, through the focus on knowledge representation, to the more recent emphasis on learning and parallel-processing alternatives to hard-wired computer programs. The explanations are necessarily only overviews, but they are very clear and easy to follow. For example, the neural network architectures developed by Carpenter and Grossberg, which embody their Adaptive Resonance Theory, are notoriously difficult to understand — even for AI researchers. Caudill does a superb job of making their operating principles clear. In fact, her explanations of various neural network techniques were for me the strongest parts of the book.

As the chapters unfold, one of the central tensions in AI is revealed: the contrast in approach between traditional “symbolic” AI — in which aspects of the world are represented as discrete abstract symbols which are then operated on by the computer (Expert Systems exemplify this approach) — and “connectionist” AI, which focuses on adaptable, “subsymbolic” processing mod-

elled loosely on the way the brain works. Caudill believes that these approaches can and will be merged in the quest for a functioning android.

I had a few negative reactions as well, though. First, her discussions of some of the newest directions that AI has taken — the use of evolutionary techniques, for example, are divorced from the main text and included in her philosophical speculations, without discussing how these emerging techniques will help achieve the project’s goals. Other hot topics are not even mentioned, for instance: modelling computation as a free market, the Rodney Brooks-led turn away from the functional decomposition of intelligence, Minsky’s Society of Mind model, and “distributed” AI, in which problem-solving takes place in the interaction between semi-autonomous agents who negotiate with one another. Also, the interesting engineering issues concerning the construction of an android’s musculature, skin, and other “hardware”, are not reported — which is appropriate for an

43

EXTROPY #11 Summer/Fall 1993

introduction to AI but is a sad oversight for a book putatively about androids.

My most severe disappointment was Caudill’s failure to address the integration issue. AI researchers have divided cognition up into parts along much the same lines as Caudill, and (as she points out) the progress has been promising. But a central puzzle that AI is trying (and failing, currently) to solve is how to put the pieces together. It seems that many of the hardest issues have somehow fallen between the cracks of the problem division. Caudill would have served her readers better by reporting this in a straightforward way, even though it might have tempered her optimistic prognosis for rapid progress along well-defined paths.

The second (much shorter) segment of the book takes a rather cursory look at the philosophical and social consequences of androids. She hits the main obvious points: What does it mean for something to be alive? How will society react to intelligent machines? Should they have rights? Would you let your daughter marry one? How will we feel when they surpass us? But the analysis is rather mechanical, and is completely divorced from the concurrent changes that will be sweeping society as other technologies are developed in the coming decades. Some readers who have been living in caves may not have seen these issues before, but most have — especially those who will actually buy her book. I’d suggest that the movie Blade Runner and a couple episodes of Star Trek do at least as good a job of making you think about the ramifications of this fundamentally important component of our future. Much written science fiction does even better.

But that is a minor part of her book. For readers who are largely mystified by the methods and issues of AI, I firmly recommend In Our Own Image, on the strength of Caudill’s considerable talent for clear explanation. In fact, were I teaching an introductory AI course, I would consider assigning this book as a first week’s introductory reading. However, if you have taken such a course or have independently followed AI, much of this book will probably be rehashes of things you already know (though the neural net material still may make it worthwhile).

These particular strengths and weaknesses aside, Caudill has done a service to Extropians with this book, by bringing the challenge more into public view. The creation of Androids will increase the intelligence that can be applied to progress in all endeavors, but additionally — as the book’s title suggests — it will help us understand the human mind and so empower our own self-transformations in ways as yet unforeseen.

Mirror Worlds — Or the day software puts the Universe in a shoebox… How it will happen and what it will mean

by David Gelernter.

Oxford University Press, 1992. New York.

ISBN: 0-19-506812-2, 0-19-507906-X

Reviewed by Harry Shapiro Hawk

Extropians will find Gelernter’s views appealing in many respects. He writes about how groups of uncoupled entities can communicate and work together by choice. That is certainly a metaphor for an idealized Extropian community.

Mirror Worlds is Gelernter’s prediction of a software revolution that will allow us to enter into a real-time, dynamic model of our world. We can enter it for pleasure and profit. We can send our software agents to do our bidding. It is like the box Gibson’s character ‘Bobby’ is plugged into; a simulation of our Universe running forward and in real-time. Gelernter contends ‘[it] will allow us to explore the world in unprecedented depth and detail without leaving the comfort of home.’

Mirror Worlds is also a technical primer for creating highly modular software for parallel and massively parallel computers. It primes us with Gelernter’s view of computer architecture and how that can represent a ‘fine art form’ in its own right. Gelernter’s views have historically been unorthodox, especially his strong believe in parallelism. You can expect to find his views refreshing, even if you don’t completely agree with them. Like the clearly more Extropian-minded, Hans Moravec, Gelernter presents an interesting thought experiment, based on his actual research, of where he feels the future of computer science and our society will go.

The prologue starts the book appealingly, Gelernter poetically writing, ‘technology is the ocean on a bright cool Spring day. Sparkling in the far distance; breathtakingly cold; exhilarating once you’ve plunged in… The cold ocean is coming to meet you,’ and extols, ‘Why not give it a try? Hold your breath. Let’s plunge.’

The first chapter is devoted to explaining, in a non-technical way, what a simulation like his proposed Mirror Worlds would be like. He compares the view of software in 1991 to the view of technology in 1791:

‘People were pretty sure, in 1791, that the industrial revolution had ‘happened’. It was history… In retrospect, little had changed… In 1791, the industrial revolution was merely building up a head of steam… Glancing backwards from a vantage-point two centuries hence, 1991 will look a lot like 1791. The real software revolution won’t have much to do with fancy robots, computers in education, … or the other hot topics that dominate this month’s hit parade. It will have to do with… Mirror Worlds… Today, software as a building material resembles mosaic tile. In the future, software will metamorphose into a something more like stone or steel or concrete. The metamorphosis has in fact (just) begun.’

He then spends the remainder of the book writing about creating and using software tools. These tools are: Tuples which are passive data, Infomachines which are tuples actively running ‘their’ programs, Ensembles which are groups of Infomachines, Trellises, which are layers of specialized Ensembles, and finally FGP machines (F=Fetch, G=Generalize, P=Project) which are collections of multiple Trellises being manipulated to extract inference and conclusions from a vast sea of tuples which he calls Tuplespace. When he says vast sea of tuples he really means vast. He imagines every sensor like I.C.U medical equipment and every datum like a bid or an ask on some trading floor being available.

I first heard about David Gelernter when reading an article he had written for the August, 1989 issue of Scientific American [Vol. 261; No. 2; Pg. 66]. He was explaining a new method for writing programs that relied upon ‘anonymous uncoupled communication,’ where each ‘component that produces data need not know who will use it or when’ and ‘components that require data need not know who produces it.’ Gelernter had developed such a system while a graduate

EXTROPY #11 Summer/Fall 1993

44

student; he called it Linda. Linda allows multiple processes to run and share data across multiple CPUs.

In the SciAm article Gelernter wrote:

Linda programs inhabit it what we call ‘tuple space.’ (A tuple is a chunk of data; the term is a generalization of terms such as quadruple and quintuple.) Passive tuples are just data available for reading or processing. Active tuples are subprograms, all executing simultaneously, that consume and produce other tuples. Active tuples turn into passive ones, available for reading or processing, once they have finished executing.

He depicted a tuple space structured into a ‘trellis: a row of modules at the bottom connected to sensors in the real world, a second, higher row to refine the data and make connections between different items, a third row for further refining and so on. Two-way communication between rows permits the lower-level modules to alter their actions in response to queries or comments passed down from upper levels…’

With many trellises connected into different parts of the real world: hospitals, traffic systems, your home’s heating system, the FAA flight control database, and many more, at the highest level of abstraction you would have a software model of the ‘universe.’ That is a Mirror World.

Gelernter has clearly refined his thinking since the 1989 SciAM publication. He also presents work by various graduate students of his, who have been actually building the types of systems he describes. He presents the work of Researcher Scott Fertig who created a FGP environment for radiology diagnosis. This system with a very limited database of 70 cases was able to correctly determine that 1) a new case was unlike anything it had seen before and 2) the breast tumor in question was malignant.

In this radiology example, there was a human feeding in data to the computer. Gelernter sees a different type of human interaction in future systems. He sees a human plugged into the top level of a trellis. He even shows a illustration with a bunch of floating heads above and seemingly connected to a trellis. For any protoposthuman, this is a example of how a human consciousness could retain its identity, do profitable work and be connected to millions of other minds though a vast data-scape (Gelernter’s Tuplespace). Gelernter writes:

A Trellis, it turns out is like a crystal… When you turn it on it vibrates at a certain frequency… In concept each

Trellis element is an Infomachine. All these Infomachines run separately and simultaneously… In practice, we do things somewhat differently… Workers collaborate to make the whole thing work predictably. Predictability is crucial. We run the Trellis in a series of sweeps… we instruct each worker to run through its list… sticking with each one just long enough… When it is done, it waits until all the rest have finished. Then all the workers proceed into the next sweep. Hence the ‘frequency’ of a Trellis… In a fast Trellis, sweeps are short; the frequency is high… In a slow Trellis, the opposite… It’s easy to imagine a Trellis that includes human elements alongside the software ones. In a… Trellis, lower rungs act ‘instinctively.’ Higher levels look for the big picture… In these areas, we could use people to realize some of the higher-rung elements… At some level of the hierarchy, human elements start to intermingle with software ones.

Gelernter calls such combinations, Turingware.

What is largely missing from the book is an in-depth focus on the agoric aspects of the computer models Gelernter presents. He does briefly mention Bernardo Huberman’s work at Xerox. I would like to see more about how bits of tuples are bought, sold and traded. How FGP machines will evolve a ‘correct’ ratio between buying expensive data versus the cost of building complex inference engines. However, despite the lack of such agoric concepts I don’t see Gelernter’s overall architecture precluding them.

In one other general sense is the overall work non-Extropian. For example, early in the book Gelernter spends time writing how a Mirror World connected into the heart of a democratic government will allow citizens of that democracy actually seek out and find how it is actually working and thus make it work better. I see such systems working in the opposite direction; namely showing why a strong central government can never work. Of course in a idealized Extropian PPL$^{1}$ society such a system would allow each community member to monitor, to what ever degree desired, how well each citizen was meeting their contractual obligations.

While not mentioning memes, Gelernter does write about people’s fear of the unknown. He, most likely, unknowingly tells how Mirror Worlds will fill the receptor site for that fear, ‘I’ve claimed that Mirror Worlds are a development of a large potential importance. This is why… They will make the world

SMART DRUGS

Piracetam, Vinpocetine, DHEA, Oxiracetam, Centrophenoxine, Minaprine & Deprenyl are used to enhance mental functions such as memory, attentiveness and intelligence. These and other nootropics are available to Americans by mail order, for personal use. Send $12 for the Nootropic News: medical reviews, newsletter, book reviews, and the Directory of Mail Order Suppliers. This updated directory includes the addresses, and if available, the phone and fax numbers of fifteen+ overseas mail order suppliers who make shipments to Americans. Current prices for the nootropics are listed. As we are not affiliated with any suppliers, our information is objective and non-promotional.

Nootropic News

P.O. Box 177-ex
Camarillo, CA
93011

run better and smoother… My guess is that, by offering topsight to the millions (not merely to the visionaries who have monopolized it in the past), they speak directly to the large, perpetually unsatisfied human craving to understand ‘what’s going on,’ to see things whole. For ‘reasons’ that transcend the rational, they will be hard to resist.’ I think that memetics adds enough reason so that we can rationally understand why humans want to understand the ‘whole.’ I fear Gelernter means that such a better understanding of the whole will lead to better governments rather than simply better societies.

Nevertheless, Gelernter sees that Mirror Worlds will force people to model and interact with accurate rational models of the world. This is something that too many people don’t do. If Mirror Worlds can do this, then I think we will be working our way towards a better society.

If these ideas intrigue you or you want more technical details, I strongly recommend Mirror Worlds as a good starting point. Since Linda is commercially (and competitively) available from several sources, for those inclined to tinker, these ideas can soon become very real.

$^{1}$Privately Produced Law. See the article of this title by Tom Morrow in Extropy #7.

45

EXTROPY #11 Summer/Fall 1993