Table of Contents:

i. Introduction/Synopsis

PART I: HUMAN INTELLIGENCE AND THE NEOCORTEX

Section 1: An Introduction to Intelligence, the Neocortex and Hierarchical Thinking

1. Intelligence, the Neocortex and Hierarchical Thinking

Section 2. The Human Neocortex: The Structure of the Neocortex and the Pattern Recognition Theory of Mind

2. The Structure of the Neocortex: Uniform and Hierarchical

3. The Pattern Recognition Theory of Mind (PRTM)

• a. Pattern Recognizers
• b. Learning, Recognition and Thinking
• c. Top to Bottom Communication in Pattern Recognizers
• d. Positive and Negative Signals in Pattern Recognizers
• e. Establishing New Pattern Recognizers, Laying Down Redundant Ones and Replacing Unused Ones
• f. The Ubiquity and Reach of the Pattern Recognition Scheme

4. From Human Intelligence to Artificial Intelligence

PART II: THE EVOLUTION OF ARTIFICIAL INTELLIGENCE

5. Solving the Speech Recognition Problem

• a. Vector Quantization
• b. Hidden Markov Models (and Hierarchical Hidden Markov Models)
• c. Genetic Algorithms
• d. Training a Computer to Recognize Speech (and More)

6. The Latest in AI: Combining Self-Organizing Systems with Pre-Programmed Systems

7. Our Most Advanced AI Machine: Watson

• a. An Introduction to Watson
• b. Watson and the Turing Test

PART III: THE FUTURE OF ARTIFICIAL INTELLIGENCE

8. Modeling and Simulating the Human Brain

• a. The Human Connectome Project
• b. The Blue Brain Project
• c. Educating a Simulated Human Brain
• d. The Exponential Growth of Information-Based Technologies

9. The Existential Capabilities of a Simulated Human Brain: Consciousness, Free Will and Identity

10. Beyond Human Intelligence & Merging with Our Machines

• a. Beyond Human Intelligence
• b. Merging with Our Machines

11. Conclusion: Intelligence Goes Cosmic

 i. Introduction/Synopsis

When IBM’s Deep Blue defeated humanity’s greatest chess player Gary Kasparov in 1997 it marked a major turning point in the progress of artificial intelligence (AI).  A still more impressive turning point in AI was achieved in 2011 when another creation of IBM named Watson defeated Jeopardy! phenoms Ken Jennings and Brad Rutter at their own game. As time marches on and technology advances we can easily envision still more impressive feats coming out of AI. And yet when it comes to the prospect of a computer ever actually matching human intelligence in all of its complexity and intricacy, we may find ourselves skeptical that this could ever be fully achieved. There seems to be a fundamental difference between the way a human mind works and the way even the most sophisticated machine works—a qualitative difference that could never be breached. Famous inventor and futurist Ray Kurzweil begs to differ

To begin with—despite the richness and complexity of human thought—Kurzweil argues that the underlying principles and neuro-networks that are responsible for higher-order thinking are actually relatively simple, and in fact fully replicable. Indeed, for Kurzweil, our most sophisticated AI machines are already beginning to employ the sample principles and are mimicking the same neuro-structures that are present in the human brain.

Beginning with the brain, Kurzweil argues that recent advances in neuroscience indicate that the neocortex (whence our higher-level thinking comes) operates according to a sophisticated (though relatively straightforward) pattern recognition scheme. This pattern recognition scheme is hierarchical in nature, such that lower-level patterns representing discrete bits of input (coming in from the surrounding environment) combine to trigger higher-level patterns that represent more general categories that are more abstract in nature. The hierarchical structure is innate, but the specific categories and meta-categories are filled in by way of learning. Also, the direction of information travel is not only from the bottom up, but also from the top down, such that the activation of higher-order patterns can trigger lower-order ones, and there is feedback between the various levels. (The theory that sees the brain operating in this way is referred to as the Pattern Recognition Theory of the Mind or PRTM).

As Kurzweil points out, this pattern recognition scheme is actually remarkably similar to the technology that our most sophisticated AI machines are already using. Indeed, not only are these machines designed to process information in a hierarchical way (just as our brain is), but machines such as Watson (and even Siri, the voice recognition software available on the iPhone), are structured in such a way that they are capable of learning from the environment. For example, Watson was able to modify its software based on the information it gathered from reading the entire Wikipedia file. (The technology that these machines are using is known as the hierarchical hidden Markov model or HHMM, and Kurzweil was himself a part of developing this technology in the 1980’s and 1990’s.)

Given that our AI machines are now running according to the same principles as our brains, and given the exponential rate at which all information-based technologies advance, Kurzweil predicts a time when computers will in fact be capable of matching human thought—right down to having such features as consciousness, identity and free will (Kurzweil’s specific prediction here is that this will occur by the year 2029).

What’s more, because computer technology does not have some of the limitations inherent in biological systems, Kurzweil predicts a time when computers will even vastly outstrip human capabilities. Of course, since we use our tools as a natural extension of ourselves (sometimes figuratively, but also literally), this will also be a time when our own capabilities will vastly outstrip our capabilities of today.  Ultimately, Kurzweil thinks, we will simply use the markedly superior computer technology to replace our outdated neurochemistry (as we now replace a limb with a prosthetic), and thus fully merge with our machines (a state that Kurzweil refers to as the singularity). In the end, the author maintains, the singularity will stretch to all corners of the universe. This is the argument that Kurzweil makes in his new book How to Create a Mind: The Secret of Human Thought Revealed.

Here is Ray Kurzweil introducing and discussing his new book:

*To check out this book at Amazon.com, or purchase it, please click here: How to Create a Mind: The Secret of Human Thought Revealed

What follows is a full executive summary of Ray Kurzweil’s How to Create a Mind: The Secret of Human Thought Revealed.

PART I: HUMAN INTELLIGENCE AND THE NEOCORTEX

Section 1: An Introduction to Intelligence, the Neocortex and Hierarchical Thinking

1. Intelligence, the Neocortex and Hierarchical Thinking

Cognitive intelligence includes many aspects (from object recognition, to memory, to abstract thinking, to imagination and creativity among others), but when we strip it down to its most basic level  we may think of intelligence quite simply as the ability to process, store and manipulate information (loc. 117). In functional terms, it is what allows an individual to learn from their environment, to predict what will come next, and to adjust accordingly (loc. 1152). The human species is far above all others when it comes to intelligence, of course; but it is not the case that we enjoy a total monopoly here. Indeed, many other species show some ability to store and manipulate information, and other mammals in particular even demonstrate a fair bit of sophistication in this regard—to the point where they too are able to learn from and adjust to their environment in complex ways.

The reason why mammals (including humans) excel in terms of intelligence has to do with a very special brain structure that is unique to our class and which is called the neocortex (loc. 121). The neocortex is a relatively recent evolutionary add-on that sits atop the older, more primitive structures of the brain (loc. 559) (As shown below).

The neocortex is responsible for virtually all of our higher-level thinking, and, as Kurzweil explains, it is particularly important in intelligence because it allows mammals to think hierarchically (loc. 121, 558). Essentially, hierarchical thinking allows us to understand individual things as part of a larger superstructure, and to understand these larger superstructures as part of still larger superstructures, all of the way up to higher and higher levels of abstraction. Here is Kurzweil to explain: “the mammalian brain has a distinct aptitude not found in any other class of animal. We are capable of hierarchical thinking, of understanding a structure composed of diverse elements arranged in a pattern, representing that arrangement with a symbol, and then using that symbol as an element in a yet more elaborate configuration. This capability takes place in a brain structure called the neocortex” (loc. 124). Elsewhere, Kurzweil writes that “we do know the neocortex is responsible for our ability to deal with patterns of information and to do so in a hierarchical fashion. Animals without a neocortex (basically nonmammals) are incapable of understanding hierarchies” (loc. 557).

Hierarchical thinking is particularly important and powerful for two reasons. To begin with, as Kurzweil points out, nature itself is in many ways arranged hierarchically (loc. 147, 558). For example, “trees contain branches; branches contain leaves; leaves contain veins [etc.]” (loc. 147). Thinking hierarchically therefore allows us an especially rich and accurate representation of the world around us—which we can then exploit to advance our ends (loc. 557). Second, as we go further up the levels of the hierarchies, our thoughts become more and more abstract and complex. And as our thinking becomes increasingly complex it affords us an increasing range of mental capabilities with which to understand and react to the environment (loc. 125) (more on these capabilities in a moment).

In this sense, then, we can well see why the neocortex and its hierarchical thinking evolved in the first place; for it allows an individual to be very flexible and to adapt to its environment very quickly, which ultimately helps it to survive and reproduce. As Kurzweil explains, “once biological evolution stumbled on a neural mechanism capable of hierarchical learning, it found it to be immensely useful for evolution’s one objective, which is survival… The salient survival advantage of the neocortex was that it could learn in a matter of days” (loc. 1150).

Being able to learn quickly is a distinct advantage no matter what the circumstances are, of course, but it is especially valuable in times when the environment is changing rapidly. It is for this reason, scientists believe, that mammals were able to survive and flourish during the Cretaceous period, while the dinosaurs (who did not enjoy this advantage) were reduced to extinction (loc. 1156).

Now, while all mammals have a neocortex, they differ to a significant degree in just how large and developed it is. For example, “in rodents it is the size of a postage stamp and is smooth” (loc. 563). In primates, on the other hand, the neocortex is a fair bit larger, and is also “folded over the top of the rest of the brain with deep ridges, grooves, and wrinkles to increase its surface area” (loc. 563). In our species the neocortex is particularly large, as it “constitutes the bulk of the human brain, accounting for 80 percent of its weight” (loc. 563).

As the neocortex increases in size and complexity it becomes capable of achieving higher and higher levels of abstraction. These higher levels of abstraction lead not just to quantitative improvements in intelligence, but to qualitative differences that translate into new capabilities. For example, the size of the human neocortex affords us two important capabilities not found anywhere else in the animal kingdom. The first is the ability to build new ideas on top of old ones (thereby ever-increasing their complexity); and the second is the ability to use our scaffolded ideas to build increasingly complex tools (loc. 129). As Kurzweil explains, “through an unending recursive process we are capable of building ideas that are ever more complex. We call this vast array of recursively linked ideas knowledge. Only Homo sapiens have a knowledge base that itself evolves, grows exponentially, and is passed down from one generation to another. Our brains gave rise to yet another level of abstraction, in that we have used the intelligence of our brains plus one other enabling factor, an opposable appendage—the thumb—to manipulate the environment to build tools. These tools represented a new form of evolution, as neurology gave rise to technology” (loc. 129).

In addition to these impressive capabilities, the neocortex is (as mentioned above) also responsible for virtually all of our higher-level thinking. As the author explains, the human neocortex “is responsible for sensory perception, recognition of everything from visual objects to abstract concepts, controlling movement, reasoning from spatial orientation to rational thought, and language—basically, what we regard as ‘thinking’” (loc. 559). These capabilities are very diverse and wide ranging; but again, Kurzweil’s contention is that they are all built off of the same capacity, and that is the capacity to process information hierarchically (loc. 173, 204-208, 231). Essentially, then, while human intelligence may appear to be extraordinarily complex, it can actually be reduced to a very simple and straightforward ability adapted to many different types of phenomenon, and in many different ways (loc. 208, 231).

Section 2. The Human Neocortex: The Structure of the Neocortex and the Pattern Recognition Theory of Mind

2. The Structure of the Neocortex: Uniform and Hierarchical

The idea that all of the many wonders of the neocortex can be reduced to a single type of thought process (hierarchical thinking) is lent credence by the structure of the neocortex itself. This proves to be the case because the neocortex is remarkably uniform in structure throughout, and this structure is itself hierarchical in nature.

To begin with, Kurzweil explains how “a critically important observation about the neocotrex is the extraordinary uniformity of its fundamental structure. This was first noticed by American neuroscientist Vernon Mountcastle (born in 1918). In 1957 Mountcastle discovered the columnar organization of the neocortex… That year he described the remarkably unvarying organization of the neocortex, hypothesizing that it was composed of a single mechanism that was repeated over and over again” (loc. 578).

The fundamental uniformity of the neocortex was reconfirmed in a recent study using the latest in brain scanning technology (loc. 1199). The lead scientist in this study, Harvard neuroscientist and physicist Van J. Weeden, explains the findings thus: “‘using magnetic resonance imaging… what we found was that rather than being haphazardly arranged or independent pathways, we find that all of the pathways of the brain taken together fit together in a single exceedingly simple structure. They basically look like a cube. They basically run in three perpendicular directions, and in each one of those three directions the pathways are highly parallel to each other and arranged in arrays. So, instead of independent spaghettis, we see that the connectivity of the brain is, in a sense, a single coherent structure” (loc. 1212).

Here are some images of this structure from the study (more about which will be said below):

In addition to this, there is also evidence that this single coherent structure is itself organized hierarchically. For example, the Swiss neuroscientist Henry Markram performed a study wherein he scanned mammalian neocortexes in search of neural assemblies (collections of neurons that are arranged together). As Markram explains, the “‘study found evidence [of] innate Lego-like assemblies of a few dozen neurons…. Connections between assemblies may combine them into super-assemblies within a neocortical layer, then in higher-order assemblies in a cortical column, even higher-order assemblies in a brain region, and finally in the highest possible order assembly represented by the whole brain’” (loc. 1188).

3. The Pattern Recognition Theory of Mind (PRTM)

a. Pattern Recognizers

Kurzweil’s contention is that each of the layers of neural assemblies described here represents a pattern recognizer that picks up on a discrete level of hierarchically organized information in the environment (loc. 243, 582, 1194). Depending on the type of information being processed, this might be a level of auditory information, or visual information, or mnemonic information, or linguistic information etc. (loc. 701). For Kurzweil, the neural assemblies are themselves innate and pre-organized, but it is learning that fills in the precise information contained at each level of the neural assembly (that is, at each level of the conceptual hierarchy). As Kurzweil explains, “the brain starts out with a very large number of ‘connections-in-waiting’ to which the pattern recognizing modules can hook up… as we learn and have experiences, the pattern recognizing modules of the neocortex are connecting to preestablished connections that were created when we were embryos” (loc. 1219).

Kurzweil further contends that each pattern recognizer is made up of the most basic assembly of neurons—which consists of a network of about 100 neurons (loc. 580, 608).  It makes intuitive sense that the most basic function of the neocortex would be carried out by its most basic structure, but this correspondence makes sense for one other reason: there are about 300 billion neurons in the neocortex, and by Kurzweil’s estimates, about 30 million pattern recognizers are needed in order to account for human higher-order thinking (loc. 627-639).

b. Learning, Recognition and Thinking

As we make our way in the world and are exposed to new things, the information that we take in is recorded into the hierarchies of neural assemblies in our brains. For example, when we are exposed to a new face the eyes, nose, mouth, ears etc. are recorded on a separate hierarchical level of visual information than the pattern of the face taken as a whole. When we are subsequently exposed to the same face, the pattern recognizer in which each of the bits of hierarchically organized information was recorded will be triggered, and the face will be recognized. As Kuzweil explains, “learning and recognition take place simultaneously. We start learning immediately, and as soon as we’ve learned a pattern, we immediately start recognizing it” (loc. 937).

Actually, it is not the case that all of the aspects of a face need be present in order for the face to be recognized, for the levels of neural assemblies in the brain communicate with one another, and when enough of the pattern recognizers at one level of a hierarchy are triggered, it will automatically trigger a pattern recognizer at the next highest level (loc. 673). So, in the face example, if a person’s nose and mouth are detected, each of the pattern recognizers that are responsible for recognizing these features will be triggered, and—should the accumulated effect of this triggering be strong enough—the pattern recognizer at the next highest level of the hierarchy will be triggered, and the face will be recognized (loc. 670-73).

It should also be noted that each of the pattern recognizers at one level of a hierarchy may come to be weighted differently based on the likelihood that their presence indicates the presence of a pattern one level above it (loc. 714). So, for instance, if someone has a very distinctive nose, the pattern recognizer that is responsible for recognizing the nose may come to have a greater weight than the pattern recognizer that picks out their mouth. As Kurzweil explains, “not every input pattern has to be present for a recognizer to fire. The recognizer may still fire if an input with a low weight is missing, but it is less likely to fire if a high-importance input is missing” (loc. 714). The weights that are given to each pattern recognizer in each pattern hierarchy are themselves determined by experience and learning (loc. 740).

In any event, once a pattern recognizer is established in the brain, it is now also ready to be used in our regular course of thinking.  This includes both directed thought and undirected thought (loc. 811). Directed thought is consciously driven, while undirected thought is the free-flow of ideas that accompanies a wandering mind (such as when we daydream—and also when we dream at night) (loc. 811-15, 1042-54). In either case, though, our flow of thought is subject to the rules that guide pattern recognizers and their interactions (loc. 813), so let us learn more about these rules now.

c. Top to Bottom Communication in Pattern Recognizers

Even before the pattern recognizers at one level of a hierarchy trigger one higher up, they begin to prime it. This pattern recognizer in turn sends signals back down to the pattern recognizers at the next-lowest level, priming them and rendering them extra-sensitive for firing. So, if someone’s mouth is detected, the pattern recognizer representing that person’s face will be primed, and this pattern recognizer will in turn send signals down to the pattern recognizers representing other aspects of the person’s face and putting them on the look-out for detecting these features (loc. 768). As Kurzweil explains, “the neocortex is, therefore, predicting what it expects to encounter. Envisioning the future is one of the primary reasons we have a neocortex. At the highest conceptual level, we are continually making predictions—who is going to walk through the door next, what someone is likely to say next, what we expect to see when we turn the corner, the likely results of our own actions, and so on. These predictions are constantly occurring at every level of the neocortex hierarchy” (loc. 773).

d. Positive and Negative Signals in Pattern Recognizers

Pattern recognizers not only communicate to one another with green-light, or positive signals, but also with red light, or negative signals (loc. 777). Whereas the positive signals encourage an adjacent pattern recognizer to fire, the negative signals inhibit this firing, and “indicate that a certain pattern is less likely to exist” (loc. 777). As Kurzweil explains, these negative signals “can come from lower conceptual levels (for example, the recognition of a mustache will inhibit the likelihood that a person I see in the checkout line is my wife), or from a higher level (for example, I know that my wife is on a trip, so the person in the checkout line can’t be she)” (loc. 777). Inhibitory signals essentially work by way of raising the threshold needed to trigger a given pattern recognizer (loc. 777).

e. Establishing New Pattern Recognizers, Laying Down Redundant Ones and Replacing Unused Ones

Now, the brain is constantly trying to match-up the stimuli that it is exposed to with the pattern recognizers that it has already established (loc. 936). However, should it come across some stimuli that does not match any of these pattern recognizers, it will be identified as a new pattern, and a new pattern recognizer will be set down for it in the brain (loc 936-949).

In fact, even if a pattern is not entirely new, but just represents some variation on an existing one, a new pattern recognizer may still be laid down next to the original (loc 951). For example, if a familiar face is seen from a new angle, or in a different light, a new pattern recognizer may be established to cover it. As Kurzweil explains, “the face of a loved one, for example, is not stored once but on the order of thousands of times. Some of these repetitions are largely the same image of the face, whereas most show different lighting, different expressions, and so on” (loc. 628). This redundancy ensures that you have a very high chance of recognizing the pattern even in unusual circumstances (loc. 703, 841).

Still, there is a limit to how many redundant pattern recognizers the brain will lay down. As the author explains, “while it makes sense to allow for a certain amount of redundancy, it would not be practical to fill up the entire available storage area (that is, the entire neocortex) with repeated patterns, as that would not allow for a sufficient diversity of patterns” (loc. 955). Kurzweil estimates that the redundancy factor of pattern recognizers to actual patterns is about 100 to 1, though it differs depending on just how important or new each given pattern is (loc. 633-37).

Aside from limiting redundancy, the brain also has another storage-space-saving strategy, and that is that pattern recognizers that are not used for long periods of time are replaced with other patterns (loc. 879). According to Kurzweil, “that is why memories grow dimmer with time: the amount of redundancy becomes reduced until certain memories become extinct” (loc. 879).

f. The Ubiquity and Reach of the Pattern Recognition Scheme

The example used above pertains specifically to facial recognition, but it is important to remember that, for Kurzweil, the exact same processes occur in the neocortex with all of the higher-level thinking that we do with this part of our brain—from our ability to understand and use language (loc. 677, 831) to our ability to process visual, auditory and sensory information (loc. 701), to our motor control (loc. 534, 1291), to our memory (loc. 793), to our creativity (loc. 1613-21) (loc. 235). As Kurzweil puts it, “just as an astonishing diversity of organisms arises from the different combinations of the values of the genetic code found in nuclear and mitochondrial DNA, so too does an astounding array of ideas, thoughts, and skills form based on the values of the patterns (of connection and synaptic strengths) found in and between our neocortical pattern recognizers” (loc. 235). One of the main pieces of evidence that Kurzweil marshals in support of this contention is that each of our higher-level thinking capabilities may be understood both conceptually and experientially as being arranged hierarchically (loc. 525-33, 668-706, 793, 830).

Now, the various mental capabilities that arise out of the neocortex are each located in a separate area of this brain structure (loc. 1263) (as pictured below). However, it has also been shown that if a certain part of the neocortex is damaged, or is missing from birth, other parts of the neocortex will be recruited to perform the function normally carried out by the damaged or missing part (loc. 1267). This neural plasticity is well documented.  As Kurzweil explains, “plasticity has been widely noted by neurologists, who observed that patients with brain damage from an injury or a stroke can relearn the same skills in another area of the neocortex” (loc. 1267). As mentioned, this phenomenon has also been observed in people with congenital defects. For example, a team of researchers led by the American neuroscientist Marina Bedny found that “congenitally blind individuals also activate the visual cortex in some verbal tasks. We provide evidence that this visual cortex activity in fact reflects language processing. We find that in congenitally blind individuals, the left visual cortex behaves similarly to classic language regions…. We conclude that brain regions that are thought to have evolved for vision can take on language processing as a result of early experience” (loc. 1275). This neural plasticity of ours is perhaps the biggest piece of evidence in support of the view that the neocortex is uniform in structure, and operates according to the same principles no matter what capability it is performing (loc. 1258).

It should also be noted that the facial recognition example used above is fairly limited when it comes to the levels of hierarchies at play, and does not do justice to the very high levels of abstraction that we regularly achieve with our neocortex. When it comes to language, for example, “at the highest level we recognize patterns such as that’s funny, or she’s pretty, or that’s ironic” (loc. 867). “Our memories,” Kurzweil continues, “include these abstract recognition patterns as well. For example, we might recall that we were taking a walk with someone and that she said something funny, and we laughed, though we may not remember the actual joke itself” (loc. 867). The point is that even our most abstract thoughts are explicable in terms of the hierarchies of pattern recognizers in our brains (loc. 1233-49).

Now, one objection that we may bring against the pattern recognition theory of mind is that, as Kurzweil puts it, “experience seems much richer than just an orderly trip up a hierarchy of features” (loc. 853). For Kurzweil, though, what we must keep in mind here is the speed at which the signals are sent and received, and the sheer number of pattern recognizers that are firing at any given moment—as well as the fact that pattern recognizers are firing across many different faculties all at once (loc. 856). To give an example, “if we were to imagine examining your neocortex when you were looking at a particular loved one, we would see a great many firings of the axons of the pattern recognizers at every level, from the basic level of primitive sensory patterns up to many different patterns representing that loved one’s image. We would also see massive numbers of firings representing other aspects of the situation, such as that person’s movements, what she is saying and so on. So if the experience seems much richer than just an orderly trip up a hierarchy, it is” (loc. 850).

Having said that, it is also the case that there is much more going on in the brain than just what takes place in the neocortex. This includes everything from such basic skills as directing one’s attention (performed by the thalamus [loc. 1428]), to the massively important features of emotion and motivation (covered by such brain structures as the amygdala and insula [loc. 1522, 1563]). For the purposes of understanding higher intelligence, though—which is our particular concern when it comes to artificial intelligence—the workings of the neocortex are our principle interest (loc. 1537).

4. From Human Intelligence to Artificial Intelligence

It should be clear at this point that the human neocortex is capable of a very wide range of very complex abilities. At the same, it should be equally clear that the underlying structures and principles that are responsible for these abilities are very simple and straightforward. Still, the processes at play are nonetheless quite sophisticated, and we may legitimately wonder whether a machine could ever be designed to operate in the same way.

As Kurzweil points out, though, our most advanced AI machines and software programs are in fact already beginning to employ the principles and processes at play in the neocortex. In order to get a good idea of how these machines are operating (and in what way they mimic what is going on in the brain) it is helpful to start from the beginning, and to explore how they were designed in the first place, which is where we will turn to now.

PART II: THE EVOLUTION OF ARTIFICIAL INTELLIGENCE

5. Solving the Speech Recognition Problem

When computer scientists were first dabbling in artificial intelligence, the general approach that they took was to first come up with intelligent solutions to problems, and then program a computer to apply these solutions to the problems that they were met with (loc. 2153). This is the approach that Kurzweil himself took with his initial attempts at programming computers to recognize speech (and convert it to text) in the 1980’s. As the author explains, “at first, we used traditional AI approaches by directly programming expert knowledge about the fundamental units of speech—phonemes—and rules from linguists on how people string phonemes together to form words and phrases” (loc. 1894).

However, computer scientists—including Kurzweil—found that this approach quickly ran into difficulties. In Kurzweil’s case, it was found that his early speech recognition system was tripped up by the fact that there is a massive amount of variability in enunciation and pronunciation between different people (and even in the same person across time) (loc. 1909). As Kurzweil explains, “the linguistic rules we had programmed were breaking down and could not keep up with the extreme variability of language” (loc. 1913).

a. Vector Quantization

In order to solve this problem, Kurzweil and his team needed to employ a few different techniques. To begin with, the team needed to come up with a way to simplify the sonic information that they were dealing with. They did this by way of making use of a technique called vector quantization (loc. 1928).  Adapted to the speech recognition problem, the technique involved breaking down speech into a limited number of phonemic iterations or points (1,024 in this case) (loc. 1938). Once these were established, each phoneme in a speech sample would be drawn to the most similar pre-established point (loc. 1951). As Kurzweil explains, “the result of this technique is that instead of having the millions of points that we started with (and an even larger number of possible points), we have now reduced the data to just 1,024 points that use the space of possibilities optimally” (loc. 1947). Interestingly, it was only after the fact that Kurzweil discovered that the brain does in fact do something very similar when it processes auditory information (loc. 1924).

b. Hidden Markov Models (and Hierarchical Hidden Markov Models)

The sonic information that Kurzweil and his team were dealing with was now simplified, but this still left the much taller task of finding a way to allow a computer to identify new utterances (loc. 1966). Kurweil believed that the best approach here would be to somehow find a way to recreate what was going on in the speaker’s brain as they were speaking, and then use this simulation to help identify new utterances. “So [he] wondered: Was there a mathematical technique that would enable us to infer the patterns in the speaker’s brain based on her spoken words? One utterance would obviously not be sufficient, but if we had a large number of samples, could we use that information to essentially read the patterns in the speaker’s neocortex (or at least formulate something mathematically equivalent that would enable us to recognize new utterances)?” (loc. 1978).

A mathematical technique to infer unknown data points from known ones (called the hidden Markov model) was in fact available (loc. 1993), and Kurzweil believed that this model held the hope of solving the speech recognition problem (loc. 1997). Now, Kurzweil already had a strong hunch at this point that speech and language are processed hierarchically in the brain (long before the aforementioned discoveries regarding the neocortex were made, and long before Kurzweil became an adherent of the PRTM) (loc. 1916, 1971). So Kurzweil incorporated this hunch into the hidden Markov model that he used, and ensured that the unknown data points would be organized hierarchically (loc. 1997). As Kurzweil explains, “I envisioned a system in which we would take samples of human speech, apply the hidden Markov model technique to infer a hierarchy of states with connections and probabilities (essentially a simulated neocortex for producing speech), and then use this inferred hierarchical network of states to recognize new utterances” (loc. 1997). Since the unknown data points of the hidden Markov  model would be organized into hierarchies, Kurzweil refers to this as a hierarchical hidden Markov model (HHMM) (loc. 2002).

c. Genetic Algorithms

In order to set the parameters of the unkown data points and their organizational hierarchies, Kurzweil borrowed a trick from the biological world: evolution. Specifically, for each of the parameters that needed to be set (which included “the initial topology of hierarchical states…, the recognition threshold at each level of the hierarchy, the parameters that control the handling of the size parameters, and many others” [loc. 2037]), Kurzweil first determined all of the possible values of its component parts (loc. 2045). So, for example, “if the problem is optimizing the design parameters for a circuit, then we define a list of all of the parameters (with a specific number of bits assigned to each parameter) that characterize the circuit” (loc. 2044).

Each of the possible values for each of the component parts of each of the specific parameters were then randomly combined with one another to produce an enormous number of set parameters (loc. 2045). Each of the set parameters is called a genetic code, and is considered to be a ‘simulated solution organism’ (loc. 2045). The solution organisms were then tested to see which ones worked best. As Kurzweil explains, “we evaluate each simulated organism in a simulated environment by using a defined method to assess each set of parameters… we would run each program generated by the parameters and judge it on appropriate criteria (did it complete the task, how long did it take, and so on)” (loc. 2049).

Once the test is complete the very best solution organisms are kept, while all the rest are disposed of (loc. 2052). The surviving solution organisms are then cross-bred with one another to produce a whole new generation of solution organisms: “in other words, we create new offspring where each new creature draws one part of its genetic code from one parent and another part from a second parent” (loc. 2053). This process mimics biological evolution right down to the level of including random mutations in the parameter values. As Kurzweil explains, “as these simulated organisms multiply, we allow some mutation (random change) in the chromosomes to occur” (loc. 2057).

The new generation of solution organisms is now run through the same set of tests to determine which ones stand out (loc. 2057). So long as some improvement is made from one generation to the next, the breeding process continues (loc. 2057). Eventually, though, the improvement tails off. At this point, the breeding process is stopped and the researchers pick out the most successful solution organisms: “when the degree of improvement in the evaluation of the design creatures from one generation to the next becomes very small, we stop this iterative cycle and use the best design(s) in the last generation” (loc. 2061). The best design(s) are then used to set the parameters of the hierarchical hidden Markov model (loc. 2061).

d. Training a Computer to Recognize Speech (and More)

This is only the initial stage in the procedure, though. For the HHMM still needs to be trained by way of being exposed to human speech (loc. 2000-04). In order to ensure that the system would be able to recognize anyone’s speech, it needed to be trained using speech samples from many different people (loc. 1998). In the course of this training, the HHMM learns “the likelihood that specific patterns of sound are found in each phoneme, how the phonemes influence one another, and the likely orders of phonemes. The system can also include probability networks on higher levels of language structure, such as the order of words, the inclusion of phrases, and so on up the hierarchy of language” (loc. 2020). The result of this learning process was that many of the rules that the system learned differed in subtle but important ways from the hand-coded rules that had originally been used (loc. 2025)—thus giving the system a great advantage in terms of efficiency.

As we can see, then, the system was designed to be self-organizing, in that it was designed to learn as it proceeds (loc. 2157). In fact, as Kurzweil points out, the system is self-organizing in 2 ways: “we… used two self-organizing methods: a genetic algorithm to simulate the biological evolution that gave rise to a particular cortical design, and HHMMs to simulate the cortical organization that accompanies human learning” (loc. 2068).

So, did the system work? Yes! Much to the surprise of Kurzweil’s colleagues, the system worked admirably (loc. 2007). As the author explains, “our effort turned out to be very successful, having succeeded in recognizing speech comprising a large vocabulary with high accuracy” (loc. 2010).

The system worked so well, in fact, that all of Kurzweil’s “subsequent speech recognition efforts have been based on hierarchical hidden Markov models” (loc. 2010). This includes Kurzweil’s latest speech recognition system, “a product called Dragon Naturally Speaking (Version 11.5) for the PC from Nuance (formerly Kurzweil Computer Products)” (loc. 2124).  And others in the industry have also been won over by the HHMM technique: “other speech recognition companies appeared to discover the value of this method independently, and since the mid-1980’s most work in automated speech recognition has been based on this approach” (loc. 2010).

Here is a (promotional) demonstration of Dragon Naturally Speaking:

What’s more, the hierarchical hidden Markov model approach (as well as other, closely related self-organizing systems [loc. 2267]) has also been extended to other areas of artificial intelligence (loc. 2154), such as speech simulation (loc. 2010) and natural-language recognition (loc. 2013, 2267). For example, “Siri, the personal assistant on contemporary Apple iPhones, uses the same speech recognition technology with extensions to handle natural-language understanding” (loc. 2126). Even Watson, the IBM machine that impressively beat Jeapordy! prodigies Ken Jennings and Brad Rutter operates with the same technology (loc. 189, 2314, 2331).

6. The Latest in AI: Combining Self-Organizing Systems with Pre-Programmed Systems

Now, all of this may make it sound like self-organizing systems have come to completely replace pre-programmed systems in the field of AI, but this is not actually the case. Indeed, nowadays it is more common to find both types of systems in an AI machine than either in isolation. The reason for this is that it was found that though pre-programmed systems have significant limitations, they also have some important strengths.

To begin with, it was found that while pre-programmed systems may lack the subtlety to deal with very novel information, they often do better with very familiar information. As Kurzweil explains, “hand-built rules work well for a core of common basic knowledge” (loc. 2300). Self-organizing systems are able to master common knowledge as well, but they must learn to do this, and this can take a good deal of time (loc. 2304). By incorporating pre-programmed knowledge in an AI system, then, the system can achieve an adequate level of performance right away, which is a great benefit (loc. 2306). The self-organizing system can then gradually take over, sharpening the machine’s efficiency as it progresses (loc. 2307).

Over and above this, pre-programmed knowledge was also found to be valuable because it acts as sound foundation for a self-organizing system to learn off of (loc. 2311). That is, pre-programmed systems “provide a solid basis for the lower levels of the conceptual hierarchy so that the automated learning can begin to learn higher conceptual levels” (loc. 2311). Combining both pre-programmed and self-organizing systems in an AI machine, then, not only allows the machine to be ready for use quicker, but it also allows it to learn quicker.

Given the benefits of combining both types of systems, most AI machines now make use of both. For example, “this is… how Siri and Dragon Go! Work—using rules for the most common and reliable phenomenon and then learning the ‘tail’ of the language in the hands of real users” (loc. 2307). Watson also combines the two types of systems in its operations (loc. 2311, 2339-52).

7. Our Most Advanced AI Machine: Watson

a. An Introduction to Watson

Actually, in Watson’s case, the machine uses multiple systems of both varieties. As Kurzweil explains, “Watson deploys literally hundreds of different systems—many of the individual language components in Watson are the same ones that are used in publicly available natural-language-understanding systems—all of which are attempting to either directly come up with a response to the Jeopardy! query or else at least provide some disambiguation of the query” (loc. 2344).

To coordinate Watson’s many systems, the machine uses an ‘expert manager’ that picks out the appropriate system(s) to be used in any given situation (loc. 2344). The expert manager is called UIMA (Unstructured Information Management Architecture). As the author explains, “UIMA is basically acting as the expert manager to intelligently combine the results of the independent systems. UIMA goes beyond earlier systems, such as the one we developed in the predecessor company to Nuance, in that its individual systems can contribute to a result without necessarily coming up with a final answer. It is sufficient if a subsystem helps narrow down the solution” (loc. 2374).

It is also UIMA that is responsible for determining Watson’s relative confidence in the answers that it comes up with. This relative confidence is expressed in terms of a probability percentage, and these probability percentages were displayed during the Jeopardy! matches.

Here is Watson in action in the first episode of the 3-part Jeopardy! challenge (links to parts 2 and 3 are below):

Part 2

Part 3

Some have contended that the fact that Watson comes up with its answers based on probability factors means that it is doing nothing more than performing statistical analysis—and so does not in fact have any true understanding of the language that it is using (loc. 186, 2394). However, Kurzweil counters that “hierarchical statistical analysis is exactly what the human brain is doing when it is resolving multiple hypotheses based on statistical inference” (loc. 2395). Therefore, if we do not consider what Watson is doing to express true understanding, then neither can we say that what humans do involves true understanding (loc. 190).

b. Watson and the Turing Test

While Kurzweil argues that Watson operates in a way that is essentially the same as the biological neocortex, he is the first to admit that Watson is far from passing the famous Turing test (which would mean that it could fool someone into thinking that it is in fact human [loc. 2226]). There are two reasons for this. The first is that Watson was never designed to pass the Turing test, but to perform a very particular task: compete at Jeopardy! So, for example, Watson was not designed to engage in a conversation (loc. 2356), or to expound on things like the themes of a great story (loc. 2373), but to answer very specific and brief questions (loc. 2365). For Kurzweil, though, the 2 former capabilities are not at all far off from what Watson is already doing, and if its systems were tweaked in certain ways, Kurzweil believes that it could probably perform reasonably well in these tasks (loc. 2369, 2373).

Still, Kuzweil concedes that even if a system were designed to pass the Turring test, at this point the technology is simply not quite advanced enough to pull it off (loc. 2377, 2385). Nevertheless, Kurzweil has no doubt that this will one day be accomplished. The fact is that there are already significant parallels between the way that our most sophisticated AI technology works and the way the most sophisticated part of our brain (the neocortex) works. Incredibly, many of the advances in AI occurred before the discoveries regarding the neocortex were made (loc. 3864-68). Nevertheless, now that we understand that our AI machines have in fact been mimicking the human brain in important ways all along, it points the way toward future progress. Specifically, we can now look to the human brain as a model for future advances in AI (loc. 3864).

When it comes to our understanding of the human brain, it is encouraging to know that we have already achieved a good degree of progress. Specifically, our ability to scan the human brain to learn its inner workings has already achieved impressive heights, as has our ability to simulate what is going on here. Nevertheless, much progress remains to be made on both counts. Thankfully, this progress is being made very quickly, as both technologies are advancing in leaps and bounds (indeed exponentially). Projects that aim to make use of these technologies to model and simulate the entire human brain are already well underway. We will now take a look at a couple of these projects and the technology behind them.

PART III: THE FUTURE OF ARTIFICIAL INTELLIGENCE

8. Modeling and Simulating the Human Brain

a. The Human Connectome Project

The basis for all attempts to simulate the human brain is the technology needed to model it in a very precise way. And in this great advances have already been made. For example, a very precise form of scanning technology was used to reveal the grid-like structure of the connections in the neocortex that was mentioned above (loc. 1837). In this case, the technology that was used was “a variety of noninvasive scanning technologies, including new forms of MRI, magnetoencephalography (measuring the magnetic fields produced by the electrical activity in the brain), and diffusion tractography (a method to trace the pathways of fiber bundles in the brain)” (loc. 1837).

The discoveries that have been made using this technology (and that we have mentioned) is part of a larger project called the Human Connectome Project (loc. 1833). This project, which is being undertaken by the National Institutes of Health, aspires to create a full 3-D map of all of the connections in the human brain (loc. 1833). The project aims to be complete by 2014 (here is the project’s website: http://www.humanconnectomeproject.org/).

The following is a short clip about the Human Connectome Project:

b. The Blue Brain Project

A different and still more impressive technology is being used in the Blue Brain Project, which is a project that aims to both model and “simulate the human brain, including the entire neocortex as well as the old-brain regions such as the hippocampus, amygdala, and cerebellum” (loc. 1773). The Blue Brain Project is using a scanning technology called a patch-clamp robot, with which the researchers are “measuring the specific ion channels, neurotransmitters, and enzymes that are responsible for the electrochemical activity within each neuron” (loc. 1784). As Kurzweil explains, this is “an automated system with one-micrometer precision that can perform scanning of neural tissue at very close range without damaging the delicate membranes of the neurons” (loc. 1790).

The Blue Brain Project, led by Swiss neuroscientist  Henry Markram, has already used their scanning technology to help them simulate a single neuron (in 2005), a neocortical column consisting of 10,000 neurons (in 2011), and a neural mesocircuit consisting of 100 neocortical columns (in 2011) (loc. 1776). Markram expects to be able to simulate an entire rat brain consisting of “100 mesocircuits, totaling 100 million neurons and about a trillion synapses by 2014” (loc.1780). As for the human brain, Markram’s aim is to have a full simulation by the year 2023 (loc. 1784). Here is the project’s web site: Blue Brain Project.

Here is Henry Markram speaking about the Blue Brain Project in a TED talk:

c. Educating a Simulated Human Brain

Simulating the entire human brain would be a major feat, but, as Kurzweil points out, a simulated brain would not in itself be capable of human-level thinking as we know it. This is because the simulated brain would still need to be filled with the appropriate human content. In other words, it would still need to do all of the learning that a human does in order to mimic human thought. As Kurzweil puts it, “if the Blue Brain Project brain is to ‘speak and have an intelligence and behave very much as a human does’… it will need to have sufficient content in its simulated neocortex to perform those tasks. As anyone who has tried to hold a conversation with a newborn can attest, there is a lot of learning that must be achieved before this is feasible” (loc. 1799).

Kurzweil identifies a few different ways that this learning might be achieved (loc. 1800-10), but the most promising one, he believes, is one that involves the self-organizing method that he describes in the book (and that we have seen above), adapted to operate at a much finer (indeed molecular) level (loc. 1815). As Kurzweil explains, “one can simplify molecular models by creating functional equivalents at different levels of specificity, ranging from my own functional algorithmic method (as described in this book) to simulations that are closer to full molecular simulations. The speed of learning can thereby be increased by a factor of hundreds or thousands depending on the degree of simplification used” (loc. 1815). Kurzweil believes that all of this will ultimately be achievable in the 2020’s, and that a machine capable of human-level intelligence will have been created by the year 2029 (loc. 2378, 2386, 2998).

d. The Exponential Growth of Information-Based Technologies

These timelines may seem like a bit of a stretch, but, as Kurzweil reminds us, all of the technologies needed to model and simulate a human brain, and educate it in a human way are information-based technologies, and these technologies have a well-established habit of advancing at an exponential pace (loc. 1845, 3600, 3620-3740, 3788-3804). As the author explains, “there is a dramatic difference between linear and exponential progressions (forty steps linearly is forty, but exponentially is a trillion), which account for why my predictions stemming from the law of accelerating returns seem surprising to many observers at first. We have to train ourselves to think exponentially. When it comes to information technologies, it is the right way to think” (loc. 3611).

Exponential advancements appear in all systems where previous advances are used to build still more complex ones; which, as the author points out, “pertains to both biological and technological evolution” (loc. 142). As evidenced in the quote above, Kurzweil refers to the exponential advancement of complex systems as the law of accelerating returns (LOAR) (loc. 142). The only difference between the LOAR in biological evolution and that in technological evolution is that the rate of exponential growth in the latter is much quicker than the former (loc. 3681, 3981).

9. The Existential Capabilities of a Simulated Human Brain: Consciousness, Free Will and Identity

The prospect of a machine that is capable of human-level thinking brings up all sorts of interesting questions about just how human such a machine would be. That is, would such a machine be conscious? Would it have free will? An identity? These questions are impossible to answer beforehand, but Kurzweil is willing to speculate here.

With regards to consciousness (in the sense of having a subjective experience of one’s life [loc. 2835, 2867]), Kurzweil’s hunch is that this is an emergent property that develops gradually and incrementally as thinking climbs higher and higher up each successive level of abstractness. As the author explains, “my own view… is that consciousness is an emergent property of a complex physical system. In this view a dog is also conscious but somewhat less than a human. An ant has some level of consciousness, too, but much less that of a dog” (loc. 2885). The logical end of this view is that if an AI machine manifests the same level of complexity as human thought, then it too would have the same level of consciousness, and this is indeed what Kurzweil believes: “a computer that is successfully emulating the complexity of a human brain would also have the same emergent consciousness as a human” (loc. 2888).

As for free will, this opens up a whole new can of worms of philosophical issues. To begin with, Kurzweil points out that it is not even entirely certain that we humans have free will. Indeed, there is a good argument to say that we do not (loc. 3290-99). Nevertheless, it has been suggested that free will might also be an emergent property that crops up at sufficiently high levels of complexity (loc. 3359), and Kurzweil holds out the hope that this may in fact be the case (loc. 3363, 3447). If this is true, then it stands to reason that a machine that is capable of human-level thinking would also have free will (loc. 3363). Regardless of whether or not we actually have free will, though, it is certainly true that we feel as though we do (loc. 3391), and in this a machine that is capable of human-level thinking is likely to feel the same way (loc. 3446).

When it comes to identity, as Kurzweil points out, it seems clear that this emerges out of our sense of free will and our subjective experience of our own lives (loc. 3480). Given that this is the case—and given that machines that are capable of human-level thinking will, according to Kurzweil, have both of these—then it stands to reason that (if he is correct) that these AI machines will also have their own identities. Whatever turns out to be the case, though, it will certainly be interesting to see how it all unfolds. For the process will not only reveal much about our machines, but also about ourselves (loc. 3743).

10. Beyond Human Intelligence & Merging with Our Machines

a. Beyond Human Intelligence

Creating a machine with human-level intelligence (and other existential capabilities) would surely be an astounding achievement; but for Kurzweil, this is really only the beginning. Indeed, the author maintains that while the human brain is indeed a wondrous thing, that many improvements could afford to be made on it.

For starters, while we humans are confined to our 300 million pattern processors, a synthetic brain could be made with vastly more. As Kurzweil explains, “ultimately our brains, combined with the technologies they have fostered, will permit us to create a synthetic neocortex that will contain well beyond a mere 300 million pattern processors. Why not a billion? Or a trillion?” (loc. 653, 1756). The added pattern processors will not only allow for more raw intelligence of the kind we already see in humans, but also higher orders of abstract thought (and complexity) than those that we can achieve (loc. 1652-62, 3973).

In addition, while our own neurons communicate with physical connections (axons and dendrites), digital neurons can be made to communicate wirelessly, which is a far more efficient process. As Kurzweil explains, “all the ‘wires’ in a software implementation operate via virtual links (which, like Web links, are basically memory pointers) and not actual wires. This system is actually much more flexible than that in the biological brain” (loc. 2445). The added flexibility and range of wireless connections could be exploited to generate still more impressive levels of intelligence (loc. 2445-53).

AI machines could not only be made to be vastly more intelligent than human brains, but could also be equipped with added features that clean up some of the bugs in our thinking. For example, we humans quite easily hold multiple ideas that are inconsistent with one another (loc. 2480). But an AI machine could be designed with a critical thinking module that ensures that this does not occur (loc. 2480). As Kurzweil explains, such a module “would perform a continual background scan of all of the existing patterns, reviewing their compatibility with the other patterns (ideas) in this software neocortex. We have no such facility in our biological brains, which is why people can hold completely inconsistent thoughts with equanimity. Upon identifying an inconsistent idea, the digital module would begin a search for a resolution, including its own cortical structures as well as all of the vast literature available to it” (loc. 2484).

Finally, while we humans are confined to the aims and motivations supplied to us by evolution, we could create an AI machine with a whole new set of goals (loc. 1539). For example, “future systems can have goals such as actually curing disease and alleviating poverty” (loc. 1544). Interestingly, some machines could be given the task of creating still more intelligent machines (loc. 3975).

b. Merging with Our Machines

The technologies mentioned here might not only be used to create separate AI machines, but also, eventually, to enhance our own brains. This can be done noninvasively, by creating an AI cloud accessible to all (loc. 1663, 1754-60), or more directly by way of designing and installing cortical implants straight into our brains (loc. 3557). These implants could not only be used as add-ons, but to replace existing structures, in order to beef-up and improve their functioning (loc. 3519). Ultimately, every piece of our biological brain could be replaced with new and improved computer parts (loc. 3556).

Some may worry that this process could cause a loss of identity, but Kurzweil argues that our identity is independent of the precise make-up of our neural substructures; and that, therefore, our identities should remain intact: “fundamentally we are not the stuff that makes up our bodies and brains. These particles essentially flow through us in the same way that water molecules flow through a river. We are a pattern that changes slowly but has stability and continuity, even though the stuff constituting the pattern changes quickly. The gradual introduction of nonbiological systems into our bodies and brains will be just another example of the continual turnover of parts that compose us. It will not alter the continuity of our identity any more than the natural replacement of our biological cells does” (loc. 3556).

11. Conclusion: Intelligence Goes Cosmic

As this process unfolds, the speed and power of computation will continue to advance exponentially. Now, computation does appear to have a physical limit (“based on our current understanding of the physics of computation” (loc. 3673), but, as Kurzweil points out, “those limits… are not very limiting. Ultimately we can expand our intelligence trillions-fold based on molecular computing” (loc. 3673).

As computation-capable matter is much more powerful than normal matter, Kurzweil predicts that we will continue to convert the latter into the former. Matter and energy organized as smart-matter is called ‘computronium’ (loc. 3980), and Kurzweil sees the creation of computronium as a runaway process: “over time we will convert much of the mass and energy in our tiny corner of the galaxy that is suitable for this purpose to computronium. Then, to keep the law of accelerating returns going, we will need to spread out to the rest of the galaxy and universe” (loc. 3984).

The limits of travelling at the speed of light will initially slow this process down, but Kurzweil envisions a time when even this limit will be overcome (loc. 3988), or bypassed altogether (by way of such things as shortcuts through spatial dimensions [loc. 3988]). I will let this quote from Kurzweil finish us off: “How long will it take for us to spread our intelligence in its nonbiological form throughout the universe? If we can transcend the speed of light—admittedly a big if—for example, by using wormholes through space (which are consistent with our current understanding of physics), it could be achieved within a few centuries. Otherwise, it will take much longer. In either scenario, waking up the universe, and then intelligently deciding its fate by infusing it with our human intelligence in its nonbiological form, is our destiny” (loc. 4000).

*To purchase the book at Amazon.com, please click here: How to Create a Mind: The Secret of Human Thought Revealed

*Thank you for taking the time to read this article. If you have enjoyed this summary of Ray Kurzweil’s How to Create a Mind: The Secret of Human Thought Revealed or just have a thought, please feel free to leave a comment below. Also, if you feel others may benefit from this article, please feel free to click on the g+1 symbol below, or share it on one of the umpteen social networking sites hidden beneath the ‘share’ button. Finally, if you feel inclined to tip me for my efforts, then I second that emotion! I have given you the opportunity to do so below.

Cheers,

Aaron,

The Book Reporter