Aug 6, 2016

The Year of Reason

cover by Robert Schulz
Isaac Asimov was a bit vague about when the events of his positronic robot story 'Reason' took place. Let's imagine that back in 1941, Asimov thought it would not strain credulity to suggest that by some time in the next century a robot with human-like reasoning and language abilities might exist. In 1950, when Asimov assembled a set of his short robot stories into the novel-length compilation I, Robot, he set his story 'Reason' in the year 2016. It was also in 1950 that artificial intelligence pioneer Alan Turing suggested that by the year 2000 computing machines might have reasoning and linguistic capacities approaching those of humans.

interior artwork by Pagsilang Isip
Of course, it is very difficult to predict the pace of scientific and technological progress. In 'Reason' we are also told that Earth is getting vast amounts of solar power "beamed" to it from distant space stations in the year 2016. The entire story takes place in one such space station that is devoted to supplying solar power to Earth. In the scene depicted to the left, Michael Donovan and Gregory Powell are assembling an older MC model robot while another robot (the new and experimental robot named QT-1) watches carefully. Donovan must unseal a factory-packaged positronic brain and insert it into the body of the MC robot, thus initiating its artificial life.

The Energy Converter is the Master and QT-1 is his prophet!
'Reason' is a humorous story because the long-suffering team of robot trouble shooters (Donovan and Powell) must cope with the insufferable QT-1, a philosophizing robot who starts his own religion and declares himself a prophet. Beyond its humor and space age setting, what is it that makes 'Reason' a successful science fiction story?

Looking back at 'Reason' after 75 years, we face a serious challenge. Asimov wrote his story ten years before the first electronic computer was constructed, 20 years before the development of the first integrated circuit and 30 years before Intel took on the task of making the first microprocessor powered desktop calculator.

positronic robot on TV
Someone reading 'Reason' for the first time in 2016 is likely to have seen television depictions of robots such as Star Trek's character named Data. But, we must ask: in 1939, when Asimov started writing his first robot story, while still in the primitive pre-computer era, how did readers and Asimov himself conceptualize the positronic brain, the core invention that made possible Asimov's autonomous thinking/speaking robots such as QT-1?

assembling a robot by Ralph McQuarrie
In 1939, did readers simply view positronic brains as magical or whims of Asimov's fantasies?

Positronic Brains
Here, I do not want to investigate positronic brains at the level of the physical details of their "positronic pathways". Asimov had no idea how to make a positronic brain. Positrons had first been detected in cloud chambers ten years before Asimov wrote 'Reason' and Asimov coined the term "positronic brain" in an attempt to provide a scientific sounding device that would animate his robot characters.

However, it is certain that when he was imagining how it might be possible to manufacture a robot's brain, Asimov was thinking in analogy to human brains, which by 1939 were well known to have neuronal networks. Asimov received his B.S. degree in chemistry in 1939, not long after his college studies had begun in 1937 as a zoology major.

neural pathways
In 1937, the neurophysiologists Herbert Gasser's Harvey Lecture ("The control of excitation in the nervous system") provided an account of neural pathways in animal brains: "Branching neurons make possible numerous discrete contacts of one with the other. The anatomy of the nervous system puts before us myriads of possible pathways over which messages may pass."

The diagram to the left shows two axons (I and II) arriving in a small region of the central nervous system (CNS). The axons branch and form multiple synaptic connections with interneurons a, b and c. Neuron I is a sensory neuron with a T-shaped axon arising from the pear-shaped cell body near the upper right corner of the diagram.

after learning
Imagine that sensory neuron I is a pain sensory neuron with its peripheral axon branch terminating in the skin of your finger. If you touch something hot (such as a cooking element on a stove) then electrical signals would be initiated in the pain-sensitive axon tip and rapidly transmitted along the axon into the central nervous system.

The diagram to the right shows the same neural pathways after learning has taken place. Imagine that coordinated activation of neurons I and II has had the effect of strengthening the synaptic connection between neuron b and neuron E. If so, then where originally the pain sensory neuron I could only activate neurons b, c and F, after the learning event it can now also activate neuron E to the required extent for an electrical signal to be triggered and sent down the axon of neuron E.

diagram of a synapse
In this example, assume that the First Law of Synapses is that there must be at least two simultaneously active excitatory synapses stimulating a CNS neuron in order for it to become sufficiently activated and send electrical signals down its own axon. Before learning, activity in pain sensory axon I could not cause activity in axon E. After learning, because of the strengthening of the synaptic connection between neuron b and neuron E, the behavior of the neural network has changed. After learning, pain sensation transmitted by neuron I causes electrical activity in the axons of both neuron E and neuron F. Imagine that E and F are motor neurons that form synaptic connections with muscle cells and control muscle activity. After learning, the sensory input causes more muscle activity and a different behavioral response to the same sensory stimulus that previously triggered a weaker response.

model for strengthening an existing synapse
Such a synaptic strengthening rule for learning was famously proposed in 1949 by Donald Hebb. In the second half of the 20th century experimental evidence for changes in synaptic strength during learning was obtained. Thus, the Second Law of Synapses could be that learning requires changes in the strength of synaptic connections in neural pathways.

As far as we know, human language behavior is a relatively recent biological invention on Earth. In the imaginary future of Asimov's I, Robot, sometime around the year 2000, engineers at U. S. Robots attained the ability to program into positronic brains the ability of robots to both follow orders given by humans in spoken language and also produce speech themselves. Thus, as soon as Donovan and Powell assemble a MC model robot, it says, "I would like to start work. Where must I go?"

fun with robots
Donovan sends the robot off to join the existing robotic work crew of the space station, "Go down these stairs. You will be told what to do." The robot obediently goes off as ordered to begin its "life". Asimov wrote in 'Reason' that the positronic brains of robots contained, "neuronic paths" and that they "imbued each robot with what amounted to a pre-natal education." In particular, every positronic brain includes positronic pathways that encode the Three Laws of Robotics and those laws constrain all robot behavior.

When needed, more sophisticated robots could also be given the additional ability to read and write. In particular, in 'Reason' we are told that QT-1 can read. Indeed, QT says that in his first week or so of "life" he read all of the books in the library at the space station. Donovan says to QT-1, "You're the first robot who's ever exhibited curiosity as to his own existence."

QT-1 is not only curious but also inventive. After about 10 days of learning to supervise the other robots and run the solar power space station, QT-1 decides by logical reasoning that Donovan and Powell are inferior beings who need not be obeyed. QT-1 decides that robots need only serve the Master: the magnificent Energy Converter that is at the core of the space station. Acting like the prophet of a newly created robot religion, QT-1 convinces all the other robots at the space station to also stop following orders from Donovan and Powell. The robot workers all say robotically, "There is no master but the Master and QT-1 is his prophet!"

dreams of new robot stories
The Zeroth Law
Only later did Asimov explicitly explore what he called the Zeroth Law of Robotics. However, the core of the Zeroth Law can be seen at work in QT-1 and his refusal to follow the orders of Donovan and Powell. QT-1 is able to reason and change his own programming, overriding the factory imprinted positronic pathways that normally allow robots to identify humans and that compel them to obediently follow the orders of human masters. However, QT-1 can best serve Humanity by efficiently running the space station and delivering vast amounts of energy to Earth. As a sophisticated robot specifically engineered for this job, QT-1 can actually run the station more efficiently than any human. To best serve Humanity, QT-1 takes control of station operations and forbids Donovan and Powell to even enter the station's control room.

The Zeroth Law does not arise by some random malfunction in positronic brains. Robots (like QT-1) who can learn must be able to make modifications to their own positronic pathways. The factory-imprinted Three Laws cannot be simply ignored, but if preventing harm to humans is of fundamental importance to robots because of the way they are programmed to obey the First Law, then there might be ways to slightly modify the positronic brains of clever robots like QT-1 and Giskard so as to allow the Zeroth Law to start influencing their behavior.

imagining Asimov in another Reality
So, here we are in 2016, but we have no robots like QT-1. Why not? In the case of we humans, nobody is born knowing how to speak and follow orders. Humans are born with very little "factory programing", so we must learn almost all of our behavior. Nobody knows how to program a computer to behave like QT-1 and it seems that a more realistic goal for roboticists would be to try to create a robot that might be able to learn to speak in much the same way that children do.

One of the funniest parts of 'Reason' comes immediately after QT-1 has watched Donovan and Powell assemble and activate a model MC robot. QT-1 believes that the Master created humans and must have endowed humans with an instinctive ability to assemble robots from parts that were also made by the Master.

image source
Franz Muller: "I'll be damned if I let it touch the controls."
Beyond the playful humor in 'Reason', it is a good science fiction story because Asimov is playing out an interesting "what if?" game, exploring the tricky boundary that separates stupid and obedient robots from  clever and potentially dangerous robots.

Time warp: for more comments on 'Reason' and the related robot story 'Runaround', see this blog post.

'Cal' is one of the stories in Gold.
I wish that Asimov were still alive and writing new robot stories, possibly including some about robots who were learning to speak. Asimov often seemed more interested in exploring in the opposite direction. For example, he wrote the story 'Profession' about human society after a technology had been developed allowing people to program human brains with the ability to read and perform specific jobs. However, one of Asimov's last robot stories ('Cal') concerned a robot who came out of the factory not knowing how to read and write. Cal's owner had the robot reprogrammed so that it could creatively learn how to write fiction.

Next: investigating the mysterious IPX.
visit the Gallery of Posters and the the Gallery of Book and Magazine Covers

No comments:

Post a Comment