Jeff Hawkins developed the Palm Pilot, the Treo smart phone, and other mobile computerized devices. It’s not surprising that he’s also very interested in our brain—another pretty good mobile information processor. He’s now teamed with Sandra Blakeslee (the renowned New York Times science writer) to produce a marvelous thoughtprovoking book, On Intelligence: How a New Understanding of the Brain Will Lead to the Creation of Truly Intelligent Machines (2004, Henry Holt).
Educators have obviously always been interested in the nature and development of intelligence, and this interest escalated during the past twenty-five years. The multiple intelligences theories of Howard Gardner (1983), Robert Sternberg (1985), and David Perkins (1995) were especially influential in shaping contemporary educational thought and practice. Intelligence had previously been viewed as a combined cognitive property that could be quantified and placed somewhere along a general intelligence scale that uses 100 as an average score for a given age (IQ, intelligence quotient).
Multiple intelligence theorists argued that intelligence encompasses several separate but interactive cognitive abilities, and that the nature of the current challenge determines the combination that will be used to resolve it. The robustness of a person’s individual intelligences may vary, and this could affect the response. For example, a person might be above average in linguistic ability and below average in mathematical ability.
Hawkins On Human Intelligence
Hawkins is convinced that contemporary computerized machines can’t have the intelligence of a brain because of fundamental differences in the organization and function of brains and computers. He believes however that a better understanding of intelligence can lead to the development of truly intelligent machines (perhaps within ten years), and that these machines may differ considerably from current information processing technologies.
Hawkins believes that the hierarchical organization and conscious operation of our brain’s cerebral cortex provides the key to understanding intelligence and how it might be incorporated into machines. The cerebral cortex is our brain’s deeply folded outside surface. It comprises 77% of our brain and it’s composed of six distinct layers. When unfolded, the cortex is about the area and thickness of a stack of six sheets of 12X18 inch construction paper.
The cortex is organized horizontally and vertically. Each horizontal layer processes important general functions, such as to receive incoming information (layer 4). The perhaps 30 billion cortical neurons are also organized vertically into several hundred million hair-thin columns of about 100 connected neurons that extend through the six layers and interconnect with other cortical columns and brain areas. Each column (and aggregate of related columns) has a distinct function, such as to respond to a specific tone or line segment, or to move a specific muscle—but the columns are highly interconnected.
Although our various senses seem distinct, Hawkins argues that our brain processes them all as related spatial/temporal patterns. Spatial patterns occur when multiple receptors in a sensory system are simultaneously stimulated (the notes of a chord or the elements of a scene). Temporal patterns occur when such patterns change (chord sequences form a melody, movement occurs within the scene).
Initially fragmented sensory information becomes integrated as it moves hierarchically through our highly interconnected cortex—shapes, textures, and colors combine to form a face; sounds become melodic sequences that become a song—and the face is singing the song.
Our cortex can understand the world, because it and the world are both organized hierarchically. Everything in the world is composed of parts that predictably combine to create more complex forms—molecules into cells into organs into bodies, letters into words into sentences into stories.
Such integrated perceptual information is constantly compared to previous related memorized information, and recognition occurs at some point in the processing sequence. It’s not necessary for our brain to have all the information before it recognizes something familiar. We can recognize a good friend from the back, from an introductory “Hi” on the phone, or even from a number displayed on caller ID.
Most memories involve sequential information, such as a song’s melody and lyrics, or a recipe’s directions. Our memory represents such complex sequences with a simple identifying name—America the Beautiful, angel food cake—that can activate the entire sequence. Language thus materially enhances cognitive processing.
More remarkable, our brain stores and recognizes information in invariant (or conceptual) forms. For example, although chairs occur in a wide variety of shapes, we can identify an object we’ve never seen before as a chair (and add it to our memory’s chair repertoire). This comparison process sparks analogy and metaphoric thought. As our highly interconnected brain continually seeks matches between what it’s currently experiencing and what it has experienced, it activates concepts that aren’t a perfect match, but rather close enough to be useful. Such metaphoric matches form the base of much of literature and the arts, and of common discourse (“Your room looks like a pig sty!”).
Prediction and Intelligence
We’ve tended to think of cognition as a feed-forward phenomenon—moving from the fragmentary/unknown to the integrated/known to a conscious decision and behavioral response. Hawkins argues however that intelligent thought requires the backward flow of information to be as robust as the forward flow. Analysis of cortical organization and the direction of neuronal fiber projections suggests that he’s correct.
Hawkins defines intelligence as the ability to correctly predict what will occur, and argues that prediction requires a continual comparison between what is occurring and what we expect to occur. Feedback pathways thus intelligently insert memories of previous related events into cognitive processing before sensory input records the actual event. For example, we expect to see our car in the garage before we actually see it.
We then especially attend to whatever doesn’t match our predictions—the car isn’t in the garage. Such unexpected events activate critical thinking and problem solving behaviors. An intelligent person with broad experience thus moves confidently through a mostly predictable life, imagining plausible explanations and developing successful alternate strategies when the unexpected occurs. Creative thought occurs when predictions are based on analogy because the current challenge doesn’t precisely match our prior experience.
Hawkins on Machine Intelligence
The intelligent machines Hawkins envisions will be neither human clones nor industrial robots. Their set of senses may differ considerably from human senses, but they’ll extract appropriate patterns from the environment they confront. Their hierarchical memory system and modeling capabilities will work on the same principles as the cortex, and will require training analogous to human training. What would make such machines intelligent is the nature and level of predictive ability that emerges out of a hierarchical memory tuned to the specific challenges the machine confronts.
Intelligent computerized machines could thus range from simple single-application systems to very powerful superhuman intelligent systems, but they won’t be humanlike. For example, computers aren’t limited to human sensory/motor limitations. Computers can function much more rapidly and process much more information than a brain. Computers don’t have to be encased within a single processing unit. Functional computers don’t require decades to mature like a brain but rather can be endlessly replicated.
When I begin to type in the address of a website I frequently use, my computer inserts the rest of the address after I’ve only typed a few letters—and most of the time, its prediction is correct. The predictive technology that Hawkins envisions is thus beginning to emerge.
Hawkins and Blakeslee have developed a fascinating, imaginative, and informative book that will intrigue anyone who is interested in our intelligent brain and the currently smart technologies it develops.