The Singularity Is Near_ When Humans Transcend Biology Part 24

There are limits to the power of information technology, but these limits are vast. I estimated the capacity of the matter and energy in our solar system to support computation to be at least 1070 cps (see chapter 6). Given that there are at least 10 cps (see chapter 6). Given that there are at least 1020 stars in the universe, we get about 10 stars in the universe, we get about 1090 cps for it, which matches Seth Lloyd's independent a.n.a.lysis. So yes, there are limits, but they're not very limiting. cps for it, which matches Seth Lloyd's independent a.n.a.lysis. So yes, there are limits, but they're not very limiting.

The Criticism from Software

A common challenge to the feasibility of strong AI, and therefore the Singularity, begins by distinguis.h.i.+ng between quant.i.tative and qualitative trends. This argument acknowledges, in essence, that certain brute-force capabilities such as memory capacity, processor speed, and communications bandwidths are expanding exponentially but maintains that the software (that is, the methods and algorithms) are not.

This is the hardware-versus-software challenge, and it is a significant one. Virtual-reality pioneer Jaron Lanier, for example, characterizes my position and that of other so-called cybernetic totalists as, we'll just figure out the software in some unspecified way-a position he refers to as a software "deus ex machina,"2 This ignores, however, the specific and detailed scenario that I've described by which the software of intelligence will be achieved. The reverse engineering of the human brain, an undertaking that is much further along than Lanier and many other observers realize, will expand our AI toolkit to include the self-organizing methods underlying human intelligence. I'll return to this topic in a moment, but first let's address some other basic misconceptions about the so-called lack of progress in software. This ignores, however, the specific and detailed scenario that I've described by which the software of intelligence will be achieved. The reverse engineering of the human brain, an undertaking that is much further along than Lanier and many other observers realize, will expand our AI toolkit to include the self-organizing methods underlying human intelligence. I'll return to this topic in a moment, but first let's address some other basic misconceptions about the so-called lack of progress in software.

Software Stability. Lanier calls software inherently "unwieldy" and "brittle" and has described at great length a variety of frustrations that he has encountered in using it. He writes that "getting computers to perform specific tasks of significant complexity in a reliable but modifiable way, without crashes or security breaches, is essentially impossible." Lanier calls software inherently "unwieldy" and "brittle" and has described at great length a variety of frustrations that he has encountered in using it. He writes that "getting computers to perform specific tasks of significant complexity in a reliable but modifiable way, without crashes or security breaches, is essentially impossible."3 It is not my intention to defend all software, but it's not true that complex software is necessarily brittle and p.r.o.ne to catastrophic breakdown. Many examples of complex mission-critical software operate with very few, if any, breakdowns: for example, the sophisticated software programs that control an increasing percentage of airplane landings, monitor patients in critical-care facilities, guide intelligent weapons, control the investment of billions of dollars in automated pattern recognition-based hedge funds, and serve many other functions. It is not my intention to defend all software, but it's not true that complex software is necessarily brittle and p.r.o.ne to catastrophic breakdown. Many examples of complex mission-critical software operate with very few, if any, breakdowns: for example, the sophisticated software programs that control an increasing percentage of airplane landings, monitor patients in critical-care facilities, guide intelligent weapons, control the investment of billions of dollars in automated pattern recognition-based hedge funds, and serve many other functions.4 I am not aware of any airplane crashes that have been caused by failures of automated landing software; the same, however, cannot be said for human reliability. I am not aware of any airplane crashes that have been caused by failures of automated landing software; the same, however, cannot be said for human reliability.

Software Responsiveness. Lanier complains that "computer user interfaces tend to respond more slowly to user interface events, such as a key press, than they did fifteen years earlier....What's gone wrong?" Lanier complains that "computer user interfaces tend to respond more slowly to user interface events, such as a key press, than they did fifteen years earlier....What's gone wrong?"5 I would invite Lanier to attempt using an old computer today. Even if we put aside the difficulty of setting one up (which is a different issue), he has forgotten just how unresponsive, unwieldy, and limited they were. Try getting some real work done to today's standards with twenty-year-old personal-computer software. It's simply not true to say that the old software was better in any qualitative or quant.i.tative sense. I would invite Lanier to attempt using an old computer today. Even if we put aside the difficulty of setting one up (which is a different issue), he has forgotten just how unresponsive, unwieldy, and limited they were. Try getting some real work done to today's standards with twenty-year-old personal-computer software. It's simply not true to say that the old software was better in any qualitative or quant.i.tative sense.

Although it's always possible to find poor-quality design, response delays, when they occur, are generally the result of new features and functions. If users were willing to freeze the functionality of their software, the ongoing exponential growth of computing speed and memory would quickly eliminate software-response delays. But the market demands ever-expanded capability. Twenty years ago there were no search engines or any other integration with the World Wide Web (indeed, there was no Web), only primitive language, formatting, and multimedia tools, and so on. So functionality always stays on the edge of what's feasible.

This romancing of software from years or decades ago is comparable to people's idyllic view of life hundreds of years ago, when people were "unenc.u.mbered" by the frustrations of working with machines. Life was unfettered, perhaps, but it was also short, labor-intensive, poverty filled, and disease and disaster p.r.o.ne.

Software Price-Performance. With regard to the price-performance of software, the comparisons in every area are dramatic. Consider the table on p. 10 With regard to the price-performance of software, the comparisons in every area are dramatic. Consider the table on p. 103 on speech-recognition software. In 1985 five thousand dollars bought you a software package that provided a thousand-word vocabulary, did not offer continuous-speech capability, required three hours of training on your voice, and had relatively poor accuracy. In 2000 for only fifty dollars, you could purchase a software package with a hundred-thousand-word vocabulary that provided continuous-speech capability, required only five minutes of training on your voice, had dramatically improved accuracy, offered natural-language understanding (for editing commands and other purposes), and included many other features. on speech-recognition software. In 1985 five thousand dollars bought you a software package that provided a thousand-word vocabulary, did not offer continuous-speech capability, required three hours of training on your voice, and had relatively poor accuracy. In 2000 for only fifty dollars, you could purchase a software package with a hundred-thousand-word vocabulary that provided continuous-speech capability, required only five minutes of training on your voice, had dramatically improved accuracy, offered natural-language understanding (for editing commands and other purposes), and included many other features.6

Software Development Productivity. How about software development itself? I've been developing software myself for forty years, so I have some perspective on the topic. I estimate the doubling time of software development productivity to be approximately six years, which is slower than the doubling time for processor price-performance, which is approximately one year today. However, software productivity is nonetheless growing exponentially. The development tools, cla.s.s libraries, and support systems available today are dramatically more effective than those of decades ago. In my current projects teams of just three or four people achieve in a few months objectives that are comparable to what twenty-five years ago required a team of a dozen or more people working for a year or more. How about software development itself? I've been developing software myself for forty years, so I have some perspective on the topic. I estimate the doubling time of software development productivity to be approximately six years, which is slower than the doubling time for processor price-performance, which is approximately one year today. However, software productivity is nonetheless growing exponentially. The development tools, cla.s.s libraries, and support systems available today are dramatically more effective than those of decades ago. In my current projects teams of just three or four people achieve in a few months objectives that are comparable to what twenty-five years ago required a team of a dozen or more people working for a year or more.

Software Complexity. Twenty years ago software programs typically consisted of thousands to tens of thousands of lines. Today, mainstream programs (for example, supply-channel control, factory automation, reservation systems, biochemical simulation) are measured in millions of lines or more. Software for major defense systems such as the Joint Strike Fighter contains tens of millions of lines. Twenty years ago software programs typically consisted of thousands to tens of thousands of lines. Today, mainstream programs (for example, supply-channel control, factory automation, reservation systems, biochemical simulation) are measured in millions of lines or more. Software for major defense systems such as the Joint Strike Fighter contains tens of millions of lines.

Software to control software is itself rapidly increasing in complexity. IBM is pioneering the concept of autonomic computing, in which routine information-technology support functions will be automated.7 These systems will be programmed with models of their own behavior and will be capable, according to IBM, of being "self-configuring, self-healing, self-optimizing, and self-protecting." The software to support autonomic computing will be measured in tens of millions of lines of code (with each line containing tens of bytes of information). So in terms of information complexity, software already exceeds the tens of millions of bytes of usable information in the human genome and its supporting molecules. These systems will be programmed with models of their own behavior and will be capable, according to IBM, of being "self-configuring, self-healing, self-optimizing, and self-protecting." The software to support autonomic computing will be measured in tens of millions of lines of code (with each line containing tens of bytes of information). So in terms of information complexity, software already exceeds the tens of millions of bytes of usable information in the human genome and its supporting molecules.

The amount of information contained in a program, however, is not the best measure of complexity. A software program may be long but may be bloated with useless information. Of course, the same can be said for the genome, which appears to be very inefficiently coded. Attempts have been made to formulate measures of software complexity-for example, the Cyclomatic Complexity Metric, developed by computer scientists Arthur Watson and Thomas McCabe at the National Inst.i.tute of Standards and Technology.8 This metric measures the complexity of program logic and takes into account the structure of branching and decision points. The anecdotal evidence strongly suggests rapidly increasing complexity if measured by these indexes, although there is insufficient data to track doubling times. However, the key point is that the most complex software systems in use in industry today have higher levels of complexity than software programs that are performing neuromorphic-based simulations of brain regions, as well as biochemical simulations of individual neurons. We can already handle levels of software complexity that exceed what is needed to model and simulate the parallel, self-organizing, fractal algorithms that we are discovering in the human brain. This metric measures the complexity of program logic and takes into account the structure of branching and decision points. The anecdotal evidence strongly suggests rapidly increasing complexity if measured by these indexes, although there is insufficient data to track doubling times. However, the key point is that the most complex software systems in use in industry today have higher levels of complexity than software programs that are performing neuromorphic-based simulations of brain regions, as well as biochemical simulations of individual neurons. We can already handle levels of software complexity that exceed what is needed to model and simulate the parallel, self-organizing, fractal algorithms that we are discovering in the human brain.

Accelerating Algorithms. Dramatic improvements have taken place in the speed and efficiency of software algorithms (on constant hardware). Thus the price-performance of implementing a broad variety of methods to solve the basic mathematical functions that underlie programs like those used in signal processing, pattern recognition, and artificial intelligence has benefited from the acceleration of both hardware and software. These improvements vary depending on the problem, but are nonetheless pervasive. Dramatic improvements have taken place in the speed and efficiency of software algorithms (on constant hardware). Thus the price-performance of implementing a broad variety of methods to solve the basic mathematical functions that underlie programs like those used in signal processing, pattern recognition, and artificial intelligence has benefited from the acceleration of both hardware and software. These improvements vary depending on the problem, but are nonetheless pervasive.

For example, consider the processing of signals, which is a widespread and computationally intensive task for computers as well as for the human brain. Georgia Inst.i.tute of Technology's Mark A. Richards and MIT's Gary A. Shaw have doc.u.mented a broad trend toward greater signal-processing algorithm efficiency.9 For example, to find patterns in signals it is often necessary to solve what are called partial differential equations. Algorithms expert Jon Bentley has shown a continual reduction in the number of computing operations required to solve this cla.s.s of problem. For example, to find patterns in signals it is often necessary to solve what are called partial differential equations. Algorithms expert Jon Bentley has shown a continual reduction in the number of computing operations required to solve this cla.s.s of problem.10 For example, from 1945 to 1985, for a representative application (finding an elliptic partial differential solution for a three-dimensional grid with sixty-four elements on each side), the number of operation counts has been reduced by a factor of three hundred thousand. This is a 38 percent increase in efficiency each year (not including hardware improvements). For example, from 1945 to 1985, for a representative application (finding an elliptic partial differential solution for a three-dimensional grid with sixty-four elements on each side), the number of operation counts has been reduced by a factor of three hundred thousand. This is a 38 percent increase in efficiency each year (not including hardware improvements).

Another example is the ability to send information on unconditioned phone lines, which has improved from 300 bits per second to 56,000 bps in twelve years, a 55 percent annual increase.11 Some of this improvement was the result of improvements in hardware design, but most of it is a function of algorithmic innovation. Some of this improvement was the result of improvements in hardware design, but most of it is a function of algorithmic innovation.

One of the key processing problems is converting a signal into its frequency components using Fourier transforms, which express signals as sums of sine waves. This method is used in the front end of computerized speech recognition and in many other applications. Human auditory perception also starts by breaking the speech signal into frequency components in the cochlea. The 1965 "radix-2 Cooley-Tukey algorithm" for a "fast Fourier transform" reduced the number of operations required for a 1,024-point Fourier transform by about two hundred.12 An improved "radix-a" method further boosted the improvement to eight hundred. Recently "wavelet" transforms have been introduced, which are able to express arbitrary signals as sums of waveforms more complex than sine waves. These methods provide further dramatic increases in the efficiency of breaking down a signal into its key components. An improved "radix-a" method further boosted the improvement to eight hundred. Recently "wavelet" transforms have been introduced, which are able to express arbitrary signals as sums of waveforms more complex than sine waves. These methods provide further dramatic increases in the efficiency of breaking down a signal into its key components.

The examples above are not anomalies; most computationally intensive "core" algorithms have undergone significant reductions in the number of operations required. Other examples include sorting, searching, autocorrelation (and other statistical methods), and information compression and decompression. Progress has also been made in parallelizing algorithms-that is, breaking a single method into multiple methods that can be performed simultaneously. As I discussed earlier, parallel processing inherently runs at a lower temperature. The brain uses ma.s.sive parallel processing as one strategy to achieve more complex functions and faster reaction times, and we will need to utilize this approach in our machines to achieve optimal computational densities.

There is an inherent difference between the improvements in hardware price-performance and improvements in software efficiencies. Hardware improvements have been remarkably consistent and predictable. As we master each new level of speed and efficiency in hardware we gain powerful tools to continue to the next level of exponential improvement. Software improvements, on the other hand, are less predictable. Richards and Shaw call them "worm-holes in development time," because we can often achieve the equivalent of years of hardware improvement through a single algorithmic improvement. Note that we do not rely on ongoing progress in software efficiency, since we can count on the ongoing acceleration of hardware. Nonetheless, the benefits from algorithmic breakthroughs contribute significantly to achieving the overall computational power to emulate human intelligence, and they are likely to continue to accrue.

The Ultimate Source of Intelligent Algorithms. The most important point here is that there is a specific game plan for achieving human-level intelligence in a machine: reverse engineer the parallel, chaotic, self-organizing, and fractal methods used in the human brain and apply these methods to modern computational hardware. Having tracked the exponentially increasing knowledge about the human brain and its methods (see chapter 4), we can expect that within twenty years we will have detailed models and simulations of the several hundred information-processing organs we collectively call the human brain. The most important point here is that there is a specific game plan for achieving human-level intelligence in a machine: reverse engineer the parallel, chaotic, self-organizing, and fractal methods used in the human brain and apply these methods to modern computational hardware. Having tracked the exponentially increasing knowledge about the human brain and its methods (see chapter 4), we can expect that within twenty years we will have detailed models and simulations of the several hundred information-processing organs we collectively call the human brain.

Understanding the principles of operation of human intelligence will add to our toolkit of AI algorithms. Many of these methods used extensively in our machine pattern-recognition systems exhibit subtle and complex behaviors that are not predictable by the designer. Self-organizing methods are not an easy shortcut to the creation of complex and intelligent behavior, but they are one important way the complexity of a system can be increased without incurring the brittleness of explicitly programmed logical systems.

As I discussed earlier, the human brain itself is created from a genome with only thirty to one hundred million bytes of useful, compressed information. How is it, then, that an organ with one hundred trillion connections can result from a genome that is so small? (I estimate that just the interconnection data alone needed to characterize the human brain is one million times greater than the information in the genome.)13 The answer is that the genome specifies a set of processes, each of which utilizes chaotic methods (that is, initial randomness, then self-organization) to increase the amount of information represented. It is known, for example, that the wiring of the interconnections follows a plan that includes a great deal of randomness. As an individual encounters his environment the connections and the neurotransmitter-level patterns self-organize to better represent the world, but the initial design is specified by a program that is not extreme in its complexity. The answer is that the genome specifies a set of processes, each of which utilizes chaotic methods (that is, initial randomness, then self-organization) to increase the amount of information represented. It is known, for example, that the wiring of the interconnections follows a plan that includes a great deal of randomness. As an individual encounters his environment the connections and the neurotransmitter-level patterns self-organize to better represent the world, but the initial design is specified by a program that is not extreme in its complexity.

It is not my position that we will program human intelligence link by link in a ma.s.sive rule-based expert system. Nor do we expect the broad set of skills represented by human intelligence to emerge from a ma.s.sive genetic algorithm. Lanier worries correctly that any such approach would inevitably get stuck in some local minima (a design that is better than designs that are very similar to it but that is not actually optimal). Lanier also interestingly points out, as does Richard Dawkins, that biological evolution "missed the wheel" (in that no organism evolved to have one). Actually, that's not entirely accurate-there are small wheel-like structures at the protein level, for example the ionic motor in the bacterial flagellum, which is used for transportation in a three-dimensional environment.14 With larger organisms, wheels are not very useful, of course, without roads, which is why there are no biologically evolved wheels for two-dimensional surface transportation. With larger organisms, wheels are not very useful, of course, without roads, which is why there are no biologically evolved wheels for two-dimensional surface transportation.15 However, evolution did generate a species that created both wheels and roads, so it did succeed in creating a lot of wheels, albeit indirectly. There is nothing wrong with indirect methods; we use them in engineering all the time. Indeed, indirection is how evolution works (that is, the products of each stage create the next stage). However, evolution did generate a species that created both wheels and roads, so it did succeed in creating a lot of wheels, albeit indirectly. There is nothing wrong with indirect methods; we use them in engineering all the time. Indeed, indirection is how evolution works (that is, the products of each stage create the next stage).

Brain reverse engineering is not limited to replicating each neuron. In chapter 5 we saw how substantial brain regions containing millions or billions of neurons could be modeled by implementing parallel algorithms that are functionally equivalent. The feasibility of such neuromorphic approaches has been demonstrated with models and simulations of a couple dozen regions. As I discussed, this often results in substantially reduced computational requirements, as shown by Lloyd Watts, Carver Mead, and others.

Lanier writes that "if there ever was a complex, chaotic phenomenon, we are it." I agree with that but don't see this as an obstacle. My own area of interest is chaotic computing, which is how we do pattern recognition, which in turn is the heart of human intelligence. Chaos is part of the process of pattern recognition-it drives the process-and there is no reason that we cannot harness these methods in our machines just as they are utilized in our brains.

Lanier writes that "evolution has evolved, introducing s.e.x, for instance, but evolution has never found a way to be any speed but very slow." But Lanier's comment is only applicable to biological evolution, not technological evolution. That's precisely why we've moved beyond biological evolution. Lanier is ignoring the essential nature of an evolutionary process: it accelerates because each stage introduces more powerful methods for creating the next stage. We've gone from billions of years for the first steps of biological evolution (RNA) to the fast pace of technological evolution today. The World Wide Web emerged in only a few years, distinctly faster than, say, the Cambrian explosion. These phenomena are all part of the same evolutionary process, which started out slow, is now going relatively quickly, and within a few decades will go astonis.h.i.+ngly fast.

Lanier writes that "the whole enterprise of Artificial Intelligence is based on an intellectual mistake." Until such time that computers at least match human intelligence in every dimension, it will always remain possible for skeptics to say the gla.s.s is half empty. Every new achievement of AI can be dismissed by pointing out other goals that have not yet been accomplished. Indeed, this is the frustration of the AI pract.i.tioner: once an AI goal is achieved, it is no longer considered as falling within the realm of AI and becomes instead just a useful general technique. AI is thus often regarded as the set of problems that have not yet been solved.

But machines are indeed growing in intelligence, and the range of tasks that they can accomplish-tasks that previously required intelligent human attention-is rapidly increasing. As we discussed in chapters 5 and 6 there are hundreds of examples of operational narrow AI today.

As one example of many, I pointed out in the sidebar "Deep Fritz Draws" on pp. 27478 that computer chess software no longer relies just on computational brute force. In 2002 Deep Fritz, running on just eight personal computers, performed as well as IBM's Deep Blue in 1997 based on improvements in its pattern-recognition algorithms. We see many examples of this kind of qualitative improvement in software intelligence. However, until such time as the entire range of human intellectual capability is emulated, it will always be possible to minimize what machines are capable of doing.

Once we have achieved complete models of human intelligence, machines will be capable of combining the flexible, subtle human levels of pattern recognition with the natural advantages of machine intelligence, in speed, memory capacity, and, most important, the ability to quickly share knowledge and skills.

The Criticism from a.n.a.log Processing

Many critics, such as the zoologist and evolutionary-algorithm scientist Thomas Ray, charge theorists like me who postulate intelligent computers with an alleged "failure to consider the unique nature of the digital medium."16 First of all, my thesis includes the idea of combining a.n.a.log and digital methods in the same way that the human brain does. For example, more advanced neural nets are already using highly detailed models of human neurons, including detailed nonlinear, a.n.a.log activation functions. There's a significant efficiency advantage to emulating the brain's a.n.a.log methods. a.n.a.log methods are also not the exclusive province of biological systems. We used to refer to "digital computers" to distinguish them from the more ubiquitous a.n.a.log computers widely used during World War II. The work of Carver Mead has shown the ability of silicon circuits to implement digital-controlled a.n.a.log circuits entirely a.n.a.logous to, and indeed derived from, mammalian neuronal circuits. a.n.a.log methods are readily re-created by conventional transistors, which are essentially a.n.a.log devices. It is only by adding the mechanism of comparing the transistor's output to athreshold that it is made into a digital device.

More important, there is nothing that a.n.a.log methods can accomplish that digital methods are unable to accomplish just as well. a.n.a.log processes can be emulated with digital methods (by using floating point representations), whereas the reverse is not necessarily the case.

The Criticism from the Complexity of Neural Processing

Another common criticism is that the fine detail of the brain's biological design is simply too complex to be modeled and simulated using nonbiological technology. For example, Thomas Ray writes:

The structure and function of the brain or its components cannot be separated. The circulatory system provides life support for the brain, but it also delivers hormones that are an integral part of the chemical information processing function of the brain. The membrane of a neuron is a structural feature defining the limits and integrity of a neuron, but it is also the surface along which depolarization propagates signals. The structural and life-support functions cannot be separated from the handling of information.17

Ray goes on to describe several of the "broad spectrum of chemical communication mechanisms" that the brain exhibits.

In fact, all of these features can readily be modeled, and a great deal of progress has already been made in this endeavor. The intermediate language is mathematics, and translating the mathematical models into equivalent nonbiological mechanisms (examples include computer simulations and circuits using transistors in their native a.n.a.log mode) is a relatively straightforward process. The delivery of hormones by the circulatory system, for example, is an extremely low-bandwidth phenomenon, which is not difficult to model and replicate. The blood levels of specific hormones and other chemicals influence parameter levels that affect a great many synapses simultaneously.

Thomas Ray concludes that "a metallic computation system operates on fundamentally different dynamic properties and could never precisely and exactly 'copy' the function of a brain." Following closely the progress in the related fields of neurobiology, brain scanning, neuron and neural-region modeling, neuron-electronic communication, neural implants, and related endeavors, we find that our ability to replicate the salient functionality of biological information processing can meet any desired level of precision. In other words the copied functionality can be "close enough" for any conceivable purpose or goal, including satisfying a Turing-test judge. Moreover, we find that efficient implementations of the mathematical models require substantially less computational capacity than the theoretical potential of the biological neuron cl.u.s.ters being modeled. In chapter 4, I reviewed a number of brain-region models (Watts's auditory regions, the cerebellum, and others) that demonstrate this.

Brain Complexity. Thomas Ray also makes the point that we might have difficulty creating a system equivalent to "billions of lines of code," which is the level of complexity he attributes to the human brain. This figure, however, is highly inflated, for as we have seen our brains are created from a genome of only about thirty to one hundred million bytes of unique information (eight hundred million bytes without compression, but compression is clearly feasible given the ma.s.sive redundancy), of which perhaps two thirds describe the principles of operation of the brain. It is self-organizing processes that incorporate significant elements of randomness (as well as exposure to the real world) that enable so relatively small an amount of design information to be expanded to the thousands of trillions of bytes of information represented in a mature human brain. Similarly, the task of creating human-level intelligence in a nonbiological ent.i.ty will involve creating not a ma.s.sive expert system comprising billions of rules or lines of code but rather a learning, chaotic, self-organizing system, one that is ultimately biologically inspired. Thomas Ray also makes the point that we might have difficulty creating a system equivalent to "billions of lines of code," which is the level of complexity he attributes to the human brain. This figure, however, is highly inflated, for as we have seen our brains are created from a genome of only about thirty to one hundred million bytes of unique information (eight hundred million bytes without compression, but compression is clearly feasible given the ma.s.sive redundancy), of which perhaps two thirds describe the principles of operation of the brain. It is self-organizing processes that incorporate significant elements of randomness (as well as exposure to the real world) that enable so relatively small an amount of design information to be expanded to the thousands of trillions of bytes of information represented in a mature human brain. Similarly, the task of creating human-level intelligence in a nonbiological ent.i.ty will involve creating not a ma.s.sive expert system comprising billions of rules or lines of code but rather a learning, chaotic, self-organizing system, one that is ultimately biologically inspired.

Ray goes on to write, "The engineers among us might propose nanomolecular devices with fullerene switches, or even DNA-like computers. But I am sure they would never think of neurons. Neurons are astronomically large structures compared to the molecules we are starting with."

This is exactly my own point. The purpose of reverse engineering the human brain is not to copy the digestive or other unwieldy processes of biological neurons but rather to understand their key information-processing methods. The feasibility of doing this has already been demonstrated in dozens of contemporary projects. The complexity of the neuron cl.u.s.ters being emulated is scaling up by orders of magnitude, along with all of our other technological capabilities.

A Computer's Inherent Dualism. Neuroscientist Anthony Bell of Redwood Neuroscience Inst.i.tute articulates two challenges to our ability to model and simulate the brain with computation. In the first he maintains that Neuroscientist Anthony Bell of Redwood Neuroscience Inst.i.tute articulates two challenges to our ability to model and simulate the brain with computation. In the first he maintains that

a computer is an intrinsically dualistic ent.i.ty, with its physical set-up designed not to interfere with its logical set-up, which executes the computation. In empirical investigation, we find that the brain is not a dualistic ent.i.ty. Computer and program may be two, but mind and brain are one. The brain is thus not a machine, meaning it is not a finite model (or computer) instantiated physically in such a way that the physical instantiation does not interfere with the execution of the model (or program).18

This argument is easily dispensed with. The ability to separate in a computer the program from the physical instantiation that performs the computation is an advantage, not a limitation. First of all, we do have electronic devices with dedicated circuitry in which the "computer and program" are not two, but one. Such devices are not programmable but are hardwired for one specific set of algorithms. Note that I am not just referring to computers with software (called "firmware") in read-only memory, as may be found in a cell phone or pocket computer. In such a system, the electronics and the software may still be considered dualistic even if the program cannot easily be modified.

I am referring instead to systems with dedicated logic that cannot be programmed at all-such as application-specific integrated circuits (used, for example, for image and signal processing). There is a cost efficiency in implementing algorithms in this way, and many electronic consumer products use such circuitry. Programmable computers cost more but provide the flexibility of allowing the software to be changed and upgraded. Programmable computers can emulate the functionality of any dedicated system, including the algorithms that we are discovering (through the efforts to reverse engineer the brain) for neural components, neurons, and brain regions.

There is no validity to calling a system in which the logical algorithm is inherently tied to its physical design "not a machine." If its principles of operation can be understood, modeled in mathematical terms, and then instantiated on another system (whether that other system is a machine with unchangeable dedicated logic or software on a programmable computer), then we can consider it to be a machine and certainly an ent.i.ty whose capabilities can be re-created in a machine. As I discussed extensively in chapter 4, there are no barriers to our discovering the brain's principles of operation and successfully modeling and simulating them, from its molecular interactions upward.

Bell refers to a computer's "physical set-up [that is] designed not to interfere with its logical set-up," implying that the brain does not have this "limitation." He is correct that our thoughts do help create our brains, and as I pointed out earlier we can observe this phenomenon in dynamic brain scans. But we can readily model and simulate both the physical and logical aspects of the brain's plasticity in software. The fact that software in a computer is separate from its physical instantiation is an architectural advantage in that it allows the same software to be applied to ever-improving hardware. Computer software, like the brain's changing circuits, can also modify itself, as well as be upgraded.

Computer hardware can likewise be upgraded without requiring a change in software. It is the brain's relatively fixed architecture that is severely limited. Although the brain is able to create new connections and neurotransmitter patterns, it is restricted to chemical signaling more than one million times slower than electronics, to the limited number of interneuronal connections that can fit inside our skulls, and to having no ability to be upgraded, other than through the merger with nonbiological intelligence that I've been discussing.

Levels and Loops. Bell also comments on the apparent complexity of the brain: Bell also comments on the apparent complexity of the brain:

Molecular and biophysical processes control the sensitivity of neurons to incoming spikes (both synaptic efficiency and post-synaptic responsivity), the excitability of the neuron to produce spikes, the patterns of spikes it can produce and the likelihood of new synapses forming (dynamic rewiring), to list only four of the most obvious interferences from the subneural level. Furthermore, transneural volume effects such as local electric fields and the transmembrane diffusion of nitric oxide have been seen to influence, respectively, coherent neural firing, and the delivery of energy (blood flow) to cells, the latter of which directly correlates with neural activity.The list could go on. I believe that anyone who seriously studies neuromodulators, ion channels or synaptic mechanism and is honest, would have to reject the neuron level as a separate computing level, even while finding it to be a useful descriptive level.19

Although Bell makes the point here that the neuron is not the appropriate level at which to simulate the brain, his primary argument here is similar to that of Thomas Ray above: the brain is more complicated than simple logic gates.

He makes this explicit:

To argue that one piece of structured water or one quantum coherence is a necessary detail in the functional description of the brain would clearly be ludicrous. But if, in every cell, molecules derive systematic functionality from these submolecular processes, if these processes are used all the time, allover the brain, to reflect, record and propagate spatiotemporal correlations of molecular fluctuations, to enhance or diminish the probabilities and specificities of reactions, then we have a situation qualitatively different from the logic gate.

At one level he is disputing the simplistic models of neurons and interneuronal connections used in many neural-net projects. Brain-region simulations don't use these simplified models, however, but rather apply realistic mathematical models based on the results from brain reverse engineering.

The real point that Bell is making is that the brain is immensely complicated, with the consequent implication that it will therefore be very difficult to understand, model, and simulate its functionality. The primary problem with Bell's perspective is that he fails to account for the self-organizing, chaotic, and fractal nature of the brain's design. It's certainly true that the brain is complex, but a lot of the complication is more apparent than real. In other words, the principles of the design of the brain are simpler than they appear.

To understand this, let's first consider the fractal nature of the brain's organization, which I discussed in chapter 2. A fractal is a rule that is iteratively applied to create a pattern or design. The rule is often quite simple, but because of the iteration the resulting design can be remarkably complex. A famous example of this is the Mandelbrot set devised by mathematician Benoit Mandelbrot.20 Visual images of the Mandelbrot set are remarkably complex, with endlessly complicated designs within designs. As we look at finer and finer detail in an image of the Mandelbrot set, the complexity never goes away, and we continue to see ever finer complication. Yet the formula underlying all of this complexity is amazingly simple: the Mandelbrot set is characterized by a single formula Z = Z Visual images of the Mandelbrot set are remarkably complex, with endlessly complicated designs within designs. As we look at finer and finer detail in an image of the Mandelbrot set, the complexity never goes away, and we continue to see ever finer complication. Yet the formula underlying all of this complexity is amazingly simple: the Mandelbrot set is characterized by a single formula Z = Z2 + C, in which Z is a "complex" (meaning two-dimensional) number and C is a constant. The formula is iteratively applied, and the resulting two-dimensional points are graphed to create the pattern. + C, in which Z is a "complex" (meaning two-dimensional) number and C is a constant. The formula is iteratively applied, and the resulting two-dimensional points are graphed to create the pattern.

The point here is that a simple design rule can create a lot of apparent complexity. Stephen Wolfram makes a similar point using simple rules on cellular automata (see chapter 2). This insight holds true for the brain's design. As I've discussed, the compressed genome is a relatively compact design, smaller than some contemporary software programs. As Bell points out, the actual implementation of the brain appears far more complex than this. Just as with the Mandelbrot set, as we look at finer and finer features of the brain, we continue to see apparent complexity at each level. At a macro level the pattern of connections looks complicated, and at a micro level so does the design of a single portion of a neuron such as a dendrite. I've mentioned that it would take at least thousands of trillions of bytes to characterize the state of a human brain, but the design is only tens of millions of bytes. So the ratio of the apparent complexity of the brain to the design information is at least one hundred million to one. The brain's information starts out as largely random information, but as the brain interacts with a complex environment (that is, as the person learns and matures), that information becomes meaningful.

The actual design complexity is governed by the compressed information in the design (that is, the genome and supporting molecules), not by the patterns created through the iterative application of the design rules. I would agree that the roughly thirty to one hundred million bytes of information in the genome do not represent a simple design (certainly far more complex than the six characters in the definition of the Mandelbrot set), but it is a level of complexity that we can already manage with our technology. Many observers are confused by the apparent complexity in the brain's physical instantiation, failing to recognize that the fractal nature of the design means that the actual design information is far simpler than what we see in the brain.

I also mentioned in chapter 2 that the design information in the genome is a probabilistic fractal, meaning that the rules are applied with a certain amount of randomness each time a rule is iterated. There is, for example, very little information in the genome describing the wiring pattern for the cerebellum, which comprises more than half the neurons in the brain. A small number of genes describe the basic pattern of the four cell types in the cerebellum and then say in essence, "Repeat this pattern several billion times with some random variation in each repet.i.tion." The result may look very complicated, but the design information is relatively compact.