Is it possible computers and their software will become so robust in their capabilities we will think of them as being conscious? Might they one day pursue their own lines of thought without being directed by humans? Will their interactions with us be so sophisticated that, without visually seeing them, we wouldn’t be able to distinguish them from humans?
These questions relate to the age-old question of how a mind emerges as a property of a physical brain. Ray Kurzwell examines these questions in his recent book, “How to Create a Mind: The Secret of Human Thought Revealed.” Kurzweil is a leader in artificial intelligence, with particular expertise in language.
In addition to explaining the many recent rapid advances in technology that enable computer architecture to more closely mimic the operation of the human brain, the author discusses the implications of these developments on what is meant by a mind.
He asks how our physical brain gives us a sense of ourselves, directs thoughts and embraces both everyday problem solving and the deeper profound questions we ask about the physical universe. He makes a strong case that, in the not too distant future, computers will display these same attributes.
Before entertaining questions about possessing a mind, an artificial machine would first have to clear at least three physical hurdles. First, there would have to be basic units replicating the function of the neuron.
Neurons are the basic cellular building blocks of the brain. They act as switches or nodes in an extremely complex interconnected web. Nodes are points where multiple input signals are processed and, if certain criteria are met, output signals are generated to neurons elsewhere in the web.
For a neuron to send out a signal there must be a sufficient number and strength of incoming signals from other neurons that push it past a threshold.
How readily a neuron triggers — exceeds its threshold — is a matter of how often it fires. That is, the sensitivity of a neuron for firing depends on its history. It is, at a basic level, a component of memory.
Patterns of heavily used neurons become entrenched. This means frequent use causes increases in the concentration of neurotransmitter molecules at the points of contact (synapses) between neurons. As a consequence, there is an increased likelihood those patterns will recur given certain stimuli. In a sense, the pattern is recalled.
For patterns that are not heavily used, this potentiation decreases. At some point, the brain abandons these connections. This self-editing gives rise to the brain’s plasticity, its ability to adapt and modify what was previously learned.
It is also evident that having a conscious mind requires an astronomical number of neurons. There are about 100 billion neurons packed into the 3 pound brains in humans.
Finally, there needs to be massive connectivity. Each neuron has about 10,000 connections with other neurons. This makes around one quadrillion synaptic connections.
Improved understanding of how individual neurons work has led to fabrication of ever-smaller and faster artificial neurons that mimic the functions of the real thing. But, weaving these units into ever larger networks with improved connectivity isn’t enough.
What is now turning the tide are advances in understanding the hierarchical architecture of the neocortex. This includes how connectivity develops between the various layers of the architecture, how the networks self-edit (see above) as it develops, and what overarching constraints are imposed by the general organization of the various modules of the brain.
The neocortex preforms the highest level associations in the brain. It is particularly interesting that it only evolved in the brains of mammals. Humans have, by far, the largest quantity of this tissue, followed by our closest ape relatives, who outstrip all other mammals.
Within the taxonomic class Mammalia, those with a larger neocortex have enhanced ability to perform complex associations, better anticipate outcomes from broader sources of sensory information and develop more nuanced relationships.
The neocortex overlies the other structures that are present in the brains of all animals. It can be said that it compliments, shares and competes with those more “primitive” parts of the brain. More accurately, those parts are more primitive in the sense they are common to the ancestors of each.
Materials scientists are developing methods to generate huge numbers of ever-smaller and faster neuron analogs with massive connectivity. Soon some of these performance parameters will outstrip the human brain.
Eventually, developers will be able to mimic the hierarchical structure of the neocortex. Software will be able to enhance or prune out neural pathways in much the same way humans learn and adapt.
Computers will process signals faster than humans, have a larger information capacity and have nearly instantaneous access, if we allow it, to more information than any one human can possess.
If you saw the computer named Watson beat the two all-time champions of the quiz show “Jeopardy” a few years ago, you haven’t seen anything yet.
Steve Luckstead is a medical physicist in the radiation oncology department at St. Mary Medical Center. He can be reached at firstname.lastname@example.org.