The Holly Problem
The prospects for building an android.
Posted Feb 06, 2013
When I agreed to write a blog for Psychology Today I told the editors that I would write about Artificial as well as Human Intelligence. The Feb. 2 2013 copy of the New York Times Sunday Review has an article that turns out to be highly relevant. It asks whether we should seriously think about that old saw of Science Fiction, the rise of the androids. Could this ever happen?
It turns out that these issues have been very nicely raised in a story from a Sci-Fi TV series, Eureka( Science Fiction channel, ran 2006-2011, now in reruns). The last few episodes both explored the issues that bothered the NY Times writer and raised two very old philosophic issues, Cartesian dualism and the paradox of Theseus’ ship. I will use the Eureka story to connect these issues to the state of modern Artificial Intelligence and Neuroscience.
Here is the story line. Evil conspirators kidnapped several main characters, strapped electrodes on them, and sent the information in their brains into a computer. (Those of you with medium memories will say this is a steal from a film series, The Matrix, itself inspired by Plato’s Cave.) Eventually all the heroes except one, Holly, escaped back into their bodies. Holly’s physical body (outside the computer) had been destroyed, leaving her mind trapped in the computer. Her friends, including her distraught boyfriend, Douglas, created something called a body matrix that, from the pictures, appears to include an electronic circuit for a brain. They then transported the information about Holly from the computer into the circuit contained in the body matrix. Upon recovery Holly was somewhat amnesic. She did not recall the intensity of her association with Douglas, but expressed a willingness to practice, confident that her memory would return.
Would this ever be possible? Let us take the story apart.
There were four steps; recording the information in Holly’s brain and body that controlled her behavior, transmitting that information into the computer, transmitting it back, and reconstructing Holly in the body matrix.
The recording process would be the most difficult task, and perhaps might be impossible. Constructing a model of the human brain, including all synaptic connections, is possible. Indeed, this has been done for the mouse brain, and there are proposals to create a similar model of the human brain. But the mice involved did not survive the process. Their brains had to be excised and perfused with various chemicals. Scenes in Eureka suggest that the mind, i.e. the information processing content of the brain), could be captured by sensors external to the skull. That is where the problem comes.
We already have external sensors that can capture fairly gross electrical signals, electromagnetic disturbances, or indicators of metabolic activity within parts of the brain. This would not be enough. Reconstructing a complete model of Holly’s brain would require measures of the strength of all the synaptic connections between neurons. And there is more. Brain activity is modulated by chemical substances in the bloodstream, such as adrenalin and various neurotransmitter precursors. There are marked individual differences both in the production of the chemicals and in the sensitivity of brain structures to them. The brain-blood chemistry interchange is highly interactive, so a simple snapshot of the current chemical status of the body and structural state of the brain at one point in time would not be enough to model Holly’s mind, and particularly her emotional reactions.
I at first thought that these obstacles would be insurmountable, on scientific rather than technological grounds. I reasoned that several different arrangements of synaptic connections might produce identical electrical, electromagnetic, and metabolic responses to any given stimulus. If this were to be true, the external signals would identify a class of human brains that included Holly’s, but would not be able to identify individual brains within the class. This would be a scientific rather than a technological barrier. If the information is not in a signal, no amount of engineering is going to extract something that is not there!
Then I began to have second thoughts. It might be possible to produce sequences of stimuli, and infer the brain structure from Holly's responses to the sequence. Doing so would be a huge technical challenge, far beyond our current capabilities. Still, it might someday be possible.
The second step, transferring the necessary information into the computer, would be a piece of cake, although it might take the combined bandwidth of Comcast, Verizon, and AOL to affect the transfer.
Now to step 3. Will it ever be possible to build a computer program capable of receiving a description of a human mind, and then using it to emulate the necessary brain functions? We are surprisingly close to doing this today. Programs to emulate the activity of brain structures during problem solving have been built. (See Nature’s report on the European Union’s investment of .5 billion euros for the project.) While these programs contain several million artificial neurons, as opposed to the brain’s several billion real neurons, it is not clear that this is a limitation. The brain has to contain many parallel neural tracts because the individual neurons are not reliable computing units. Artificial neurons, being ultimately an arrangement of transistors in a machine, are extremely reliable. Therefore in principle an artificial neural system can emulate a real one, using many fewer units than the real one does.
The problems of emulating brain-chemical interactions seem to me solvable. Remember, the program does not need actual agents in the blood. All it needs are variables representing the level of agents in the bloodstream.
According to the Eureka writers, at this point we still have Holly, for, as Descartes said, “I think, therefore I am” (which was actually quoted by a character in Eureka). The computer would be doing Holly’s thinking, therefore she would be. Descartes might argue that what we have in the computer is her soul, I would prefer the term “mind.”
Transferring the information from the computer back into the chip inside the body matrix would similarly simply be a bandwidth problem. There is no scientific issue here, although there is a technological challenge.
All four steps are theoretically possible. Are they currently possible? No way. I won’t live to see the day, my children probably won’t, but I am not so sure of my grandchildren. At this point we run into the philosophical issues.
Would the thing produced at Step 4 still be Holly? Descartes did not consider this, but his predecessors, the Greek philosophers, had, about 2000 years earlier. According to legend, the Athenians had mounted the ship of Theseus, the legendary founder of the city, as a civic monument. Over the years planks rotted and were replaced. Eventually all the original planks were gone. Was it still Theseus’ ship? (In a modern version of this example, suppose that a museum owns Abraham Lincoln’s axe, with its head and handle replaced.)
Aristotle concluded that the monument was still Theseus’ ship, because its form was the same. Aristotle’s predecessor, Heraclitus, reached a different conclusion, arguing that you cannot step into the same river twice. According to Heraclitus, if you duplicate the same pattern in two different physical creations, you have two different things.
The Eureka writers agreed with Aristotle. So do I, with one caveat. We are primates, and, with the possible exception of the orangutan, primates are social creatures. The (new?) (resurrected?) Holly has to fit in. That means that her new body must be subject to all the annoyances of mankind; from the sniffles to arthritis. She must age along with Douglas, and eventually must die, finally and irrevocably. Then, like the rest of us, she will live on in our memories while her soul will go we know not where.