In the second of the summer seminar series at work, I heard a lecture by Dr. James Albus, who is a NIST Fellow in the Intelligent Systems Division of the Manufacturing Engineering Laboratory. The lecture was entitled “Understanding the Mechanisms of Mind”.
Dr. Albus first mentioned that the mind is what distinguishes humans from the rest of creation, and that there are several ways to study the mind. Neuroscience is focused on an understanding the brain, including the chemistry, synaptic transmission, axonal connectivity, and functional MRI. Cognitive modeling is focused on the representation and use of knowledge in performing cognitive tasks, in cluding mathematics, logic, and language. Intelligent control is focused on making machines behave appropriately in an uncertain environment, such as in manufacturing, agriculture, mining, and autonomous vehicles, which is what he works on.
He then stated that the brain is a machine in which the processes of the mind occur. It is a control system that is divided into two parts — behavior generating and sensory processing. The different physical parts of the brain handle the various behaviors:
- Forebrain — long range plans
- Frontal cortex — reason, logic, abstract models
- Limbic — values, emotions
- Pre-motor cortex — complex behaviorial skills
- Primary motor cortex, basal ganglia — simple behaviorial skills
- Cerebellum, midbrain — coordination, balance, dynamics
- Spinal motor centers — position, force, velocity control
as well as the various senses:
- Forebrain — assessment of trends and social situations
- Frontal cortex — perception of abstract concepts
- Parietal cortex — perception of space, time, and motion
- Temporal cortex — perception of objects and words
- Primary sensory cortex — visual images, tactile map of the body
- Cerebellum, midbrain — coordination, balance, dynamics
- Spinal sensory centers — stretch, position, velocity
At the center of the brain are the modules that model and evaluate the world. The limbic system consists of emotional functions that compute good-bad, attractive-repulsive, important-irrelevant, hope-feat, love-hate, and confidence-uncertainty, among others. The hippocampus builds local maps and controls what is remembered. The thalamocortical loops perform recursive estimation. In general, there is a coupling of sensing and behavior.
He went on to say that each of the processes of the mind has a computational equivalent:
- Imagination — modeling, simulation, visualization based on an internal model of the world
- Thought — analysis of what might occur if certain actions were taken and conditions were to occur
- Feeling — experience of sensory input or emotional state
- Attention — focusing sensors and perception on what is important
- Knowledge — information organized so as to be useful
- Perception — transformation of sensation into knowledge
- Cognition — analysis, evaluation, and use of knowledge
- Meaning — relationships between forms of knowledge
- Belief — level of confidence assigned to knowledge
- Reason — logic applied to thinking
- Planning — thinking about possible future actions and goals
- Wisdom — ability to make decisions that achieve long term goals
- Intelligence — ability to achieve goals despite uncertainty
- Awareness — internal representation of the external world
- Understanding — correspondence between model and reality
- Consciousness — aware of self and one’s relationship to the world
- Emotion — value judgment, evaluation of good and bad
- Sense of good and bad — fundamental value judgment
- Sense of justice and duty — culturally derived value judgment
- Appreciation of beauty — perceptual value judgment
Curiously enough, he didn’t assign a computational equivalent to a sense of the religious. In general, there are internal models of the world that enable prediction and planning, and a characteristic range and resolution in time and space.
As an example of an intelligent system, he talked about the 4D/RCS Reference Model Architecture for Unmanned Vehicle Systems, which is used for automated driving. In particular he highlighed LADAR technology, which allows the imaging of geometric objects in real time. Therefore, locomotion is a fundamental capability of intelligent creatures that has almost been successfully duplicated by machines. It requires an understanding of space and time, awareness of the situation, and dynamic modeling of the world, plus the ability to plan for future actions, and react to unexpected events. In fact, the roadmap to fully autonomous driving looks roughly like:
- 2005 — robust autonomous road-following and off-road driving
- 2010 — LADAR cameras providing the range, resolution, and speed to cope with dense traffic
- 2015 — cognitive reasoning capabilities enabling competent tactical behaviors on the battlefield
- 2020 — cognitive reasoning and tactical behaviors approaching human levels of performance
- 2025 — autonomous vehicles surpassing human levels of performance in most, if not all, areas
So will driving become a skill of the past, much like weaving clothes or grinding wheat? Probably. This demonstrates that many fundamental processes can be modeled computationally. We now understand how to deal with complexity, acquire and use knowledge, and make decisions. Soon, computational power will exceed that of the human brain.
Then Dr. Albus started waxing philosophical. He predicted that intelligent weapons will revolutionize warfare, since they can outperform and cost less to train and maintain over manned systems; thus soldiers will be kept out of harm’s way. Well, maybe that’s true, but a bunch of battlebots fighting each other is not the goal. After all, if it’s just a bunch of battlebots, then there is no need to fight physically; one could just simulate the machines’ plan of attack and see who will win. In reality, there is no war without the threat of bloodshed. Humans will still need to plan and strategize and attack and defend after the battlebots have done their job. And, in the end, it is still the prerogative of a crazy human dictator to detonate a nuclear weapon.
Dr. Albus also predicted that intelligent systems have the potential to create wealth to pay for health care, education, housing, transportation, food, Social Security, and a clean environment, and in fact eliminate poverty. These machines have a much greater capacity to perform useful work. So yes, wealth, perhaps currently stored as ideas in the mind, can be infinitely created. However, how can this wealth be distributed in the world? His solution, which was borrowed from People’s Capitalism, was to build a real ownership society where everyone would receive a basic living income from ownership of intelligent machines. In essence, everyone, rich or poor, would own shares in a mutual fund that would pay dividends on the wealth generated by intelligent machines. A discussion with Clayton, who knows much more about economics than I do, confirmed that while socialism is a great idea in theory, it will never work in practice because human nature is not good and kind. Someone will inevitably use these intelligent machines for his/her own good, not for the good of the society, and what will prevent him/her from owning many more shares of stock than the others? Human nature in the form of corrupt government is what is preventing many poor countries from industrializing and creating wealth. So sure, it is a fine idea to develop intelligent machines, but to expect that all of the world’s economic problems will be eliminated is really unreasonable.
I also took issue with the idea that every one of our mind processes can be modeled computationally. In particular, the sense of right and wrong, and the fact that we don’t always choose what is right, cannot be put into a machine. It is a fundamental aspect of human nature. Perhaps with the right model a machine could make the right decisions, but it is essentially just mimicking human behavior, not making decisions on its own. Even within humans, there is a higher entity than the mind; I was trying to get Dr. Albus to admit that there is the concept of a soul, but he refused to say so, despite the fact that he received his bachelor’s degree from Wheaton College.
Can a created being or object eventually surpass its creator in intelligence? I don’t think so. Science fiction is replete with stories of machines taking over the world, but in reality there is always a way to stop the machines — to turn off the switch or cut the power. In that sense, I am not afraid of progress in artificial intelligence and understanding how the mind works. I do fear the consequences of evil human nature coupled with these machines, and to that end I hope that we humans will be able to behave responsibly in this new computer age.