"The opposite of a correct statement is a false statement. But the opposite of a profound truth may well be another profound truth." -- Niels Bohr

This series of articles explores the emergence of a new phase of artificial intelligence (AI). We will consider assumptions and discoveries of the R&D behind it and review some of the literature surrounding these new developments by juxtaposing two very different concepts: the ancient trade of artisan and the concept of a godlike artificial intellect (the "artilect"), which some believe will be the inevitable creation of the 21st century. Theory, opinion, and speculative developments are considered in this foray into the world of artificial intelligence and human-competitive machines.

Introduction

Artisans have been with us since the dawn of humanity. In the broadest sense, an artisan is a skilled craftsperson; one who imagines, plans, and builds something with their hands. Our cognitive emergence roots may be much, much older than we previously thought. Kate Wong, the editorial director of ScientificAmerican.com, reports in " The Morning of the Modern Mind" that evidence for symbolic thinking capabilities, a characteristic of modern man, dates back several hundred thousand years. Artisans were the first and most definitional avocation of humanity. Activities such as the making of jewelry by aboriginal artisans were the first expressions of symbolic thinking, which is the essence that separates us from our less semiotic cousins.

Our ancient ancestors used tools to create things of symbolic value prior to the advent of language. The artisan and tool together defined and enabled the emergence of modern man. The lucky accidental genetic modification that gave rise to big-brained primates who thought symbolically produced a race of eco-fit artisans, and we are their children. From ancient craftsman to Java developer, the unbroken line of semiotic workers has produced an ever-growing noosphere which seems to now be in something of an accelerated inflationary phase.

Defining Intelligence

Although theories of intelligence abound, there is no single standard by which we measure intelligence in human beings. Therefore, when it comes to the replication of human intelligence in machine form, there is no one specification or set of conditions that provides guidance. Even the Turing Test, which is probably the most cited cybernetic holy grail of human-competitive machine intelligence, is not universally accepted as an appropriate measure. In On Intelligence, Palm Computing founder Jeff Hawkins writes:

"To converse like a human on all matters (to pass the Turing Test) would require an intelligent machine to have most of the experiences and emotions of a real human, and to live a humanlike life. Intelligent machines will have the equivalent of a cortex and a set of senses, but the rest is optional. It might be entertaining to watch an intelligent machine shuffle around in a humanlike body, but it will not have a mind that is remotely humanlike unless we imbue it with humanlike emotional systems and humanlike experiences. That would be extremely difficult and, it seems to me, quite pointless." (On Intelligence, p. 208)

Intelligence may be slippery to define, but that hasn't slowed the slope of investments made attempting to create artificial versions of it. Turing's influence, coupled with the red herring that the Turing Test now seems to be, may have actually been obstacles to seminal achievements in AI. While the results may not have been Turing Test winners to date, change is in the wind. And with change will come concerns, as the changes we must now consider promise to dwarf the technological impact of previous epochs.

A Really Brief History of AI

If we were to chart the history of AI from the first known usage of the term, the really brief history of AI condensed to some of the most important developments would look like the following:

• 1956: John McCarthy coins the term "artificial intelligence" at the first AI conference. Lisp, semantic nets, and ELIZA follow.
• 1967: Richard Greenblatt at MIT builds a knowledge-based chess-playing program, MacHack, that was good enough to achieve a class-C rating in tournament play. Prolog, expert systems, and knowledge representation come next.
• 1976: Doug Lenat's AM program (Stanford PhD dissertation) demonstrates the discovery model (loosely guided search for interesting conjectures).
• 1987: Marvin Minsky publishes The Society of Mind, a theoretical description of the mind as a collection of cooperating agents.
• 1990s: Major advances occur in all areas of AI, with successes in machine learning, intelligent tutoring, case-based reasoning, multi-agent planning, scheduling, uncertainty reasoning, data mining, natural language understanding and translation, vision, virtual reality, games, and other areas.
• 1997: The Deep Blue chess program beats the current world chess champion, Garry Kasparov, in a widely followed match.
• Late 1990s: Web crawlers and other AI-based information extraction programs become essential in widespread use of the Web.

Full stop; end of story. We could leave it at that, and call it complete: all you ever really needed to know about AI, at least "classic AI," in two hundred words or less. This, of course, ignores the classic contributions of Aristotle, Decartes, Hobbes, Pascal, Leibnitz, Boole, Babbage, Whitehead, Russell, and even Turing.

Next, throw in the following four major application categories of AI systems, and we have enough for impressive cocktail party banter:

• Expert systems: Narrow and deep (domain-specific) knowledge systems.
  
• Natural language processing (NLP): Systems that analyze, understand, and generate languages that humans use naturally.
• Pattern recognition: Systems that classify data (patterns) based on either a priori knowledge or on statistical information extracted from the patterns.
• Intelligent device control: Systems that make time-sensitive adjustments to devices based on sensory input.
  

There is no such thing as a general-purpose artificial intelligence application; hence, the lack of Turing Test successes, which requires a much wider operating context and world view. AI applications to date can be very smart in their specific domains but about as bright as a rutabaga otherwise (akin to a few college professors I've known). Some would argue that it's just a question of time and horsepower--if you throw enough CPU cycles, fast memory and some evolutionary engineering (whatever that means) against a wall, enough will stick to give rise to a general-purpose, human-competitive entity.

A New Phase in AI

In the last few years, a new phase in AI started that is comprised of three major components: 1) the utilization of classic AI components in real-world, useful applications; 2) the emergence of a new phase in AI R&D; and 3) the stated objective by more than one organization to build the machine equivalent of a human brain.

The utilization of classic AI approaches for solving real-world problems has been steadily waxing for at least a decade. We actually use AI-manufactured products and AI applications on a regular basis. Do you drive a car that was manufactured after 1995? There was a robot somewhere on the assembly line, controlled in some way by an AI component. Do you buy or sell stocks? AI applications are widely available and used on a regular basis in the financial world. Do you use an auto-focus camera? Or a cell phone? Do you eat food? As it turns out, AI-based methods have been slowly creeping into the programmatic zeitgeist, finding quarter in a rainbow of market sectors, without the bother of significant media attention. As a matter of fact, the new phase in AI doesn't even use the word "intelligence." A synopsis of AI inroads into successful real-world applications includes:

• Rules and heuristics: Emergency response, diagnostics, and CRM.
• Neural networks: Fraud prevention, terrorist tracking, and consumer preference predictions.
  
• Bayesian networks: financial analysis, natural language processing, and software usability.
• Genetic and evolutionary algorithms: Molecular design applications, complexity analysis.
• Hybrid models: Machine vision and sensor applications.

All of these developments use "classic" AI; in other words, approaches to artificial intelligence and computing that are recognized and taught in a number of computer science programs around the world. While this is laudable, it is the new phase of AI that is of greater interest--the "Hawkins/de Garis" epoch of AI, which I have named for two esteemed gentlemen who have contributed influential thoughts in recent years, the sum of which will change the course of future AI work quite dramatically and forever.

What makes this phase "new?" Two characteristics epitomize the new phase in AI; the two terms in question also happen to serve as the juxtaposed entities in the title of this series: the artisan and the artilect. The artisan-like characteristics of this new phase in AI represent the unique semiotic skills of humanity, which form the basis for our claim to intelligence itself. It is the "engineering is the art of the possible" school of craftsmanship, which has taken raw learning and turned it into commerce. Indeed, all commercial instantiations of AI in the past decade have been enabled by the artisan component of the human complex. To now seed AI itself with the creative essence of homo sapiens clearly marks the beginning of a remarkable new phase. That, coupled with the artilect, the imaginary "god-like," hyper-super-duper-megalithic uber system of doom with more qbits and computational capability than has ever existed in this galactic quadrant, represents the end game in the Moore's Law saga. We are a few short years, or decades, or generations away from the emergence of the artilect, depending on who you believe.

The Age of the Artisan

This new phase in AI begins with Jeff Hawkins, the ultimate artisan. Hawkins is a founder of Palm and Handspring and almost single-handedly revived the handheld computing industry--the same Jeff Hawkins who worked with his father in a boat yard in Long Island, inventing crazy boat stuff while growing up. Hailed by Nobel laureates and venture capitalists alike, his gauntlet-laying 2004 opus On Intelligence breaks away from all of the classic approaches and posits anew the path to AI nirvana. Hawkins begins with a concept the entrenched AI community has failed to acknowledge since the beginning: human intelligence. In his book, Hawkins cites ingrained resistance to designing machine intelligence models based on the masterpiece, nature's finest cognitive creation: the human neocortex. Why? Probably because we simply do not understand how it works. How can we reverse engineer a device that is so inscrutable?

According to Hawkins, when he left Intel to study intelligence at MIT in 1981, the current thinking in the field was that we should not limit ourselves to the messiness of nature's flawed instruments when we can do so much better with our well-reasoned designs. Hawkins argued that the starting point must be real intelligence if we are to ever have hope of creating a machine version. Years later, together with Sandra Blakeslee, he wrote his account of his investigations into the workings of the human brain in his book. The book details much of Hawkins' research and his theories as to how the instrument actually works, and how that elegant design can imbue machines with something much more akin to human intelligence than anything yet seen.

An artisan-cum-entrepreneur, there was nothing else Hawkins could do after finally releasing the fruit of his 25-year investigation except to start yet another industry-disrupting device company. Go to Numenta.com and see for yourself the website of the firm that, if successful, will provide general purpose, truly human-competitive devices. If you think out-sourcing was bad, wait until what Dick Samson calls "off-peopling" moves from the early-adopter to the early majority phase--all jobs from truck drivers, burger flippers, and airline pilots to (yes) programmers, customer service representatives, and the bulk of the police force will be off-sourced--automated. No people needed. Even a marginally successful Numenta will force major structural changes the likes of which we cannot even anticipate. And that's the good news.

Enter the Artilect

The other side of the "new phase of AI" coin is even less cheerful. Hugo de Garis published The Artilect War with the term "war" in the title for a very good reason. Because as obtuse as his bloviated text tends to read, he at least understands what our species is really good at: the artisan apogee of accomplishment is war. The thesis of de Garis' book is that there's only room on this small globe for one dominant species--hence, inevitable conflict will erupt, the magnitude of which will be more devastating than all previous wars combined, because unfettered use of 21st-century weapons will be involved. It will be the "Cosmists versus the Terrans," in his nomenclature. The Cosmists being those highly enlightened human beings (like him) who are in favor of building artilects, despite the risk that these inscrutable entities may not like us very much. The Terrans, the less-evolved of us who simply cannot understand why we would ever risk building machines with the innate capacity to annihilate the entire artisan species, will not like the Cosmists very much, in the de Garis mythology. Therefore, war will erupt. Big war. The summation of all wars. And this will all come about due to Moore's Law, quantum computing, and the "evolutionary engineering" techniques which de Garis favors in his own brain-building research.

If my sarcasm has not yet successfully communicated my disdain for the de Garis hardcover, let me be clear: his book is tripe. As a sometime writer of tripe, I know it when I smell it, and his entire edition qualifies. The only possible exception is in the few pages he devotes to the merits of reversible logic vis-a-vis entropy and heat generation. From an artisan perspective, this is useful to know. But a single website like this one could have done just as well and left the rest of the nonsense where it belongs: in the "bad science fiction" bin at a Super Saver near you.

I cannot fairly judge de Garis' research, which one would hope is of higher quality than the prose, plot, and purpose of his writing (one gets the distinct impression that de Garis yearns for the celebrity-like status of Kurzweil, et al). What little he does have to say about his work in the book bears no resemblance to the innovative, well-considered investigation that Hawkins articulates. It seems as though the de Garis approach to machine intelligence is brute force--accept anything Turing had to say on the matter as gospel, wait for Moore's Law (which, like love, it would seem, covers a multitude of sins), and trust in "evolutionary engineering," which he doesn't explain in any detail in his book--and by the time his grandchildren are of the majority, magic will happen, artilects will appear, and the war is on. In all fairness, de Garis has published more in-depth material, which may prove to be of some use in the pursuit of AI solutions and even vindicate his evolutionary claims, although the work does appear to embrace a "kitchen sink" philosophy, which is probably not useful and certainly not innovative.

So why bother with de Garis? Why even consider "artilects" in a discussion of a new phase in AI? The answer is simple: because we simply do not know if general purpose, human-competitive intelligence can actually be engineered. We can imagine such machines. We often have: The Terminator, The Matrix, I Robot, the Borg of Star Trek. Our musings generally imbue these creations with evil, bone-munching intent. Before the Turing Test, Norbert Wiener co-opted the term "cybernetics" to christen a field of study and an intellectual movement that fundamentally equates human beings with machines. If the universe is a clock, we are the cogs. Can such machines be built? According to The Teleological Society (after WW2, it became the Cybernetics Group and the Macy Conferences), which Wiener helped to found with cohorts like fellow mathematician John von Neumann and anthropologist Margaret Mead, they already have. Look in a mirror to see an image of one. But that begs the question. Can we engineer them? Can we design, assemble, test and bless a machine with general purpose human-competitive intelligence? Can the artisan build the artilect? If so, it will probably be based on work more akin to a Hierarchical Temporal Memory system that Hawkins touts. The de Garis scenario is, therefore, something we must at least consider. Hawkins and de Garis are both out to build brains--mechanical brains with the right stuff to actually rival our own wetware.

The Teleology of Intelligence

The "anthropic principle," first proffered by physicist Brandon Carter in 1973 at a conference in Kraków celebrating the 500th birthday of Copernicus, hints as to the nature of being and purpose. His paper "Large Number Coincidences and the Anthropic Principle in Cosmology" stated: "Although our situation is not necessarily central [it was a Copernican celebration after all], it is inevitably privileged to some extent."

The designed-for-intelligence (-by-intelligence?) thesis presents circumstantial evidence for the specialness of the parameters of our universe insofar as it necessarily supports the emergence of universe-modeling intelligence: us artisans. As it would happen, the "Cogito ergo mundus talis est" mantra ("I think, therefore the world is as it is") that Carter first articulated has given rise to much debate in our church-versus-state, science-versus-religion, post-Enlightenment, post-modern era. The anthropic principle constrains the Big Bang; the initial conditions of the universe had to have been precisely tuned such as to give rise to the very specific set of laws that govern it, or else we wouldn't be here to reflect on the matter. There was such little room for deviation with so many physical variables that it is quite implausible to imagine coincidence at the helm. More on that later.

The term "teleology" refers to the study of design or purpose in natural phenomena. Darwin would argue that the purpose of intelligence is a natural outcome in the struggle for fitscape dominance--greater intelligence happens to do better at the propagation and survival game than does lesser intelligence--and no other explanation is needed. However, according to John Smart, a complexity theorist and editor of the advanced futurist publication Acceleration Watch:

"[A]s local computational complexity increased, our natural genetic parameters have became [sic] increasingly self-selected (e.g., rationally-directed, by human society, first through mating choices, and then through medical intervention). This is indeed the apparent teleology, or purpose, of intelligence, to move us from evolutionary (random, chaotic) to developmental (statistically predictable) contexts."

This implies that evolution itself as well as cultural developments, technology, and even the yearning for/fear of artilects is precisely the direction we are naturally goaded to take. The teleology of intelligence first gives rise to the artisan, and ultimately leads to the artilect. And now both of these poles of human endeavor, as epitomized by Hawkins and de Garis, are pursuing the ultimate engineering feat: building a general purpose, human-competitive brain.

So where exactly is AI headed? Are we heir to a teleology of intelligence, or simply stumbling forward in a drunken, random walk? How will we respond to the emergence of entities that have the potential of dramatically surpassing human intelligence, individually and collectively? Are such entities possible? If so, how long do we have before we come face-to-interface with them? And what might they think of us? Are we now at the dawn of that epoch? These, and other interesting questions are considered in the next installment in this series.

Max Goff is the Chief Marketing Officer for Digital Reasoning Systems, Inc., headquartered in Nashville, Tennessee.