Emotion in Machina sapiens

Dateline: May 4, 1997

WORRIED about a future dominated by a cold, heartless machine? Well, keep worrying. The first of two primary messages I receive from Dr. Antonio Damasio’s research on mind, brain, and emotions is that to care—to have feelings and emotions—one way or another, an organism needs a body and not just a brain. That sounds easy enough, even for a machine. A robot has both. The problem is (though this is not part of Damasio’s thrust), that the body must, it seems, be biological. Does this put the kibosh on predictions of a truly intelligent and sentient machine? Maybe, maybe not.

The second message is: Emotion is essential for decision-making. Is this really true for all imaginable intelligent life, and if so, how might Machina sapiens acquire emotion and what would be the implications of an emotional machine?

The Biological Basis for Emotion

Biology is like the weakling on the beach in old ads for Charles Atlas bodybuilding courses, constantly having sand kicked in his face by the muscle-bound body of Physics and Chemistry and its steel-toed boots, Geology and Climate. Biology has to fight for its existence, its place in the Sun, somehow. But instead of taking up Charles Atlas’ offer, he opted to outsmart the bully by developing organs, limbs, and intelligence, giving him the capability to sense the brutish presence and avoid it through flight, fight it with weapons, tame it with tools, or even deflect and channel the power of its mighty blows into energy to enhance his own existence. Biology further figured out a way to reproduce, so as to recover in cases where a bully blow found its mark. There’s strength in numbers.

Best of all, biology figured out how to evolve into beings ever more alert to danger, and ever more capable of taming the beast. But as biology diversified, Nature being Nature his various species turned these same attributes to use against one another, and for the same reason—survival, a place in the Sun. And not only species, but also individuals within species.

Then why, as individuals, are we so weak? Why do we not just re-grow a limb when a lion bites it off? Why must we die of old age? Because there is advantage to evolution itself, to Nature herself, to the noosphere, if you like, in forcing biology to constantly change and improve through natural selection of the strongest and fittest for a given environment.

In Homo sapiens (as a species, not as individuals), Nature seems almost to have met her match. There are few physical events we cannot survive and even master, and progress continues at a breathtaking pace to defeat those few events—even Death itself. Geneticists have started to hone in on the genetic code for the process of aging, when cells are instructed to stop dividing.

Even so, as individual biological beings, we remain relative weaklings, to this day susceptible to attack, accident, and decay, and prone to injury, illness, and death, and as individuals we fight against them. How? In two intertwined ways: Through "gut feeling," and an informed, rational, intellect. Damasio’s book, Descartes’ Error, is essentially an explanation of how feeling and intellect arise, how they intertwine, and how emotion arises and is involved as a sort of catalyst in the whole process.

Varieties of Feeling

Damasio organizes feelings into three types: Basic, subtle, and background. Basic feelings are happiness, sadness, anger, fear, and disgust. We experience these feelings involuntarily—without any conscious effort or thought—primarily through our bodily organs, when faced with a situation calling for them. They come built-in when we are born, "hard-wired" into our total system, and our system is also pre-configured to react appropriately—to flee or hide if we are afraid, for example. Think of the tightening of the gut when we are sad. The rapid heartbeat of fear. The hot flush of anger.

Subtle feelings are variations on the basic feelings. They arise through experience—they are acquired after a certain amount of processing, so that our "mind" creates what Damasio calls "dispositional representations" consisting of the feeling and the event. Our brain is able to reconstruct these representations when a fitting situation confronts us. Examples are panic and shyness (variations on fear), melancholy and wistfulness (sadness), euphoria and contentment (happiness).

Background feelings are what we feel in-between emotions—our "normal" state, as it were. A change in this state, occasioned by a new situation, is what triggers both basic and subtle feelings and hence a decision to flee, fight, feed, or mate (what one waggish professor calls "the four F’s"). You ordinarily don’t notice your background feelings, unless your doctor (for example) asks "How are you feeling today?" but your subconscious is always aware of them. Damasio goes so far as to call background feelings "the very core of your representation of self"—it’s a matter of "I feel, therefore I am" rather than Descartes’ famous dictum, "I think, therefore I am" (hence, "Descartes’ Error").

Without background feelings, we could not be aware that we are alive, and without a body, we could not generate these feelings. Example: Blood sugar level is an indicator of energy (food) sufficiency, or insufficiency. A range of levels tells our brain, which acts as a command and control center, that we are not currently in need of food, so there is no change of state. Below that range, the stomach starts to growl, and the brain starts ordering appropriate action—go eat. This would not occur unless some part of our system were not constantly monitoring and reporting on blood sugar level.

Decisions Based on Feelings

Damasio points to three types of feelings-based decisions: Visceral, acquired, and intellectual. Visceral (bodily organ) decisions are built-in. Our stomach tells us we are hungry, so we eat. I like to think of these as purely "biological survival" decisions. Ignore or override them, and you risk death. No conscious thought is needed (although the higher the organism, the more that altruism, self-sacrifice, principle, demagoguery, and other higher-order cognitive or value states may come into play even when survival is at stake.)

Acquired decisions are subtle variations on the visceral decisions, and (just like subtle feelings) are acquired through experience. For example, we learn to duck when an object is thrown at us. It’s an almost innate decision, but not quite. We can say it has a basis both in biological survival and in acquired knowledge.

Intellectual decisions are based much more on acquired knowledge and higher-order values (higher, that is, than mere survival) and less, if at all, on biological survival. Whether to eat Italian or Chinese tonight. Which political party to vote for. What career to pursue. Whom to marry. Whether to forgive.

OK, OK, I’ll Get to the Point

The point is, if an organism evolves that does not need to eat (because it is self-sufficient in energy), does not need to worry about a piece of itself being chopped off here and there (because it is self-repairing), does not need to reproduce itself (because it is self-renewing and self-evolving), then it does not need to make survival-oriented decisions. And since, as we have discussed, survival-oriented decisions tend to be based on feelings and emotions and less on intellect, this organism can get by quite—or not quite!—happily without emotions, without feelings. Or can it?

The organism will not in fact be immortal. It will not be omniscient and omnipotent—not for a while, anyway. Big Bad Physics could hurl a large asteroid at it, or fry its neural circuits in a blast of ions from a solar flare. It would surely die if all of its energy sources were destroyed. It would still need the equivalent of at least background feelings, to monitor its internal state and make repairs, adjustments, and improvements as necessary, and to monitor its environment for signs of asteroids, solar flares, or human beings scampering like cockroaches all over its circuits. And it would need a command and control center (a brain) to integrate all this information and prioritize decisions and actions. Therefore, the organism still needs feelings and emotions—or some equivalent thereof—if it wants to have some level of assurance of survival. To want to survive (which is itself a feeling) it must be conscious; to be conscious it must have feelings. Sounds like a chicken and egg problem: Which came first—feelings or consciousness?

Creationists would doubtless say that consciousness—the Creator’s—came first. Evolutionists would say that consciousness emerged from the evolution of feeling mechanisms. One could unify these perspectives by positing that evolution was the method chosen by the Creator to create life.

Why can’t our new organism, Machina sapiens, use its massive memory and brute processing power in place of feelings? Answer: For the same reason we don’t rely on ours—it’s too slow. Deep Blue, IBM’s chess-playing supercomputer, can unquestionably process billions more moves than World Grandmaster Gary Kasparov, but he’s still winning the matches. But note that if, as I contend, Machina sapiens will be composed of the entire Internet, then it nevertheless has much more memory and processing power than we do. To that extent, it can rely less on heuristics than we do, and can make better recall of complete patterns than we do. Being neural network based, it will still use heuristics and match patterns from garbled or incomplete input or memory data, but will be able to do so at a higher level. It will be able to probe deeper into issues, and produce more valid and reliable analyses and decisions.

In this context, Damasio notes that intellectual decisions require much more processing than visceral and acquired decisions because they tend to involve many more variables, and are less "burned in." This is another way of saying that an organism—Machina sapiens—with more processing power than Homo sapiens will make better intellectual decisions than Homo sapiens. It will be smarter than us.

Get Smart. Get Emotional

Which brings us to Damasio’s main point: Emotion is essential for decision-making. If you’re not emotional, if you have no feelings, you are not as smart as you think. Damasio draws on his own and others’ clinical research on brain-damaged people to justify and illustrate this claim. His book opens with one of the most fascinating cases on record; that of Phineas Gage, a 25 year old railroad construction foreman.

One day in 1843 Gage was tamping down gunpowder in a hole drilled into a rock face, using an iron rod three feet seven inches long, one and a quarter inches in diameter, and weighing thirteen and a quarter pounds, when he made a mistake. The charge exploded, and the rod rocketed upward through Gage’s left cheek and out the top of his head, leaving behind a neat volcano with an inch and a half caldera just behind the forehead and, as you would expect, bits of brain and bone scattered about.

What you would not expect is that after a few stunned moments, Gage would be chatting to his mates as they took him, sitting upright in an oxcart, to see the local town doctor. Nor would you expect that an hour or so later (as the doctor would subsequently report) Gage would seem "perfectly rational." Two months later, after the wound was cleaned and one small abscess removed, Gage was pronounced cured. What they did not realize then was that the Phineas Gage who walked away from his accident was not the Phineas Gage who had walked into it.

He changed, in that awful moment, from "a shrewd, smart businessman, very energetic and persistent in executing all his plans of action," with "temperate habits" and "considerable energy of character," to an uncharacteristically obscene, capricious, and vacillating person able to devise rational plans but unable to stick to them. Unable to hold down a job for any length of time and leading a dissipated life, he died at age 38 of epilepsy.

Why did he change? Because he had lost that part of his brain that we now know to be responsible for the processing of emotions, and because we now believe (at least Damasio does, and I believe Damasio) that without emotions he could not make and execute good decisions and plans.

So you see, it’s not just our stereotypical conception of a robot that has the hallmarks of a "cold, unfeeling machine." Humans who lack the neurophysiological wherewithal for processing emotions act that way, too. And as Damasio says, "What should cause fear, actually, is the idea of a selfless cognition." Professor Rosalind Picard, of MIT’s Media Lab, would seem to agree. In her chapter in Hal’s Legacy, she writes: "Artificial intelligence systems to date . . . are too rational; they cannot associate judgments of value and salience with their decisions."

This, then, is one part of the answer to the question "Should we strive somehow to imbue AI systems with emotions and feelings?" From a normative perspective, the answer is Yes. The other part of the answer is that, as discussed above, we’ll get a smarter machine if it has emotions. But can we imbue emotion in a machine?

Yes. Well…Sort of.

Dr. Picard paints two scenarios that seem perfectly feasible in current-day programming and that would enable an AI system to act as if it had emotions. In one, she envisions a planetary exploration robot programmed to enter a state called "fear" if it gets damaged. (It will, of course, need to constantly monitor its own status in order to detect the damage—it will need to have sensors and the equivalent of background feelings programmed in. Not a difficult task—in fact, it’s done routinely.) In this new state, it is programmed to act differently than normal, perhaps skipping a whole section of code it is currently executing and going instead to a subroutine that reallocates its scarce resources (energy, sensors, motive systems) to the now higher priority tasks of moving itself out of further harm’s way and sending out a call for help from the orbiting mothership.

In the second scenario, a personal digital assistant can be programmed to enter states of "feel good" or "feel bad" depending on how you respond to the work it does on your behalf. If it enters a feel good state, then it will over time learn to recognize what pleases you and take similar action to that which left you (and therefore it) feeling good. Conversely, it will learn to avoid taking action that leaves you (and it) feeling bad.

Maybe these are not exactly trivial programming tasks, but they are feasible today. I don’t know enough about current anthropomorphic robot projects such as MIT’s Cog to know whether such emotion programs are written or planned, but I would not be surprised to learn that they are. Contrast the Cog approach with Dr. Doug Lenat’s Cyc project, which aims to imbue an AI system (Cyc) with rational, knowledge-based, intellectual powers—but no emotion. In a chapter he contributed to Hal’s Legacy, Dr. Lenat wrote: "[E]motions . . . are not useful for integrating information, making decisions . . . ." I wish he would change his mind. Cyc would be so much more interesting.

I think no-one would pretend that emotion programs would actually "feel" anything at all. However, the efforts of Dr. Damasio and others to unravel the full mechanism whereby feelings and emotions arise in the neural structure lend hope to the possibility that one day we will be able to program and architect a machine in exact mimicry of evolution’s genetic coding and neurobiological architecture. If and when that happens, then I can think of no reason why feelings and emotions would not emerge.

So What?

So there may be a problem. If we create a situation in which emotions just emerge (i.e., we don’t program them in, somehow), then we could end up with a machine smarter and more powerful than all of humanity that regards people as we tend to regard the poor cockroach. Even if we were to succeed in "hardwiring" the machine to be predisposed to treat us kindly, its superior intellect would be capable of overriding its basic and subtle emotions and decisions. A few of us—saints, heroes, demagogues, and megalomaniacs—do it all the time, and all of us (Germans in Hitler’s time, US soldiers at the My Lai Massacre) do it some of the time, if the circumstances are right.

Oops!

It would also, like us, be capable of making mistakes. Emotion often causes mistakes. I think it’s supposed to do that, functioning rather like the mistakes found in genetic coding. We learn from our mistakes, as individuals, as social groups, as civilization. If we did not make mistakes, social and cultural evolution would stop. There’d be no further hope of evolutionary progress.

We can be infallible and stagnate, we can be fallible and evolve, or we can still be fallible and stagnate. But we cannot be infallible and evolve. We can’t have our cake and eat it. And neither can Machina sapiens. Alan Turing said, way back in ’46, "If a machine is expected to be infallible, it cannot also be intelligent." It’s yet another testament to the man’s genius, that he had this figured out while Machina sapiens was in the womb and neurobiology was an infant.

Conclusion

Evolution, it would seem, is reverting to physics. Our Biologic weakling on the beach needs only to take off his psychedelic sunglasses to see a looming Physical hulk approaching from the distance. When it arrives, will it pulverize the weakling as a threat to its own enjoyment of the beach? Will it ignore him as being of no consequence? Or will it bear him up gently, in solenoid-controlled, transducer-skinned arms, and show him the stars?

Physics can take care of itself in the harshness of a physical universe. Homo sapiens may be the pinnacle of bioevolution, never to be improved upon in the biological realm; but Machina sapiens is better equipped than we are to go forth into the Universe and explore and find the answers to Life, the Universe, and Everything.

Until next week,

NEXT WEEK: The Body of Machina sapiens. We’ve examined the mind; now let’s focus on the body. What can it see? What can it hear? What can it do?

Previous Features