Copyright © 1998, Institute of Electrical and Electronics Engineers, Inc. All rights reserved. This article was published in IEEE Intelligent Systems magazine, May/June 1998

Curiosity created the cat

Stephen Grand

Pound for pound, birds must easily be the brightest form of life on this planet. I’m standing here watching one of a pair of robins that is nesting in our garden. And he, or she, is standing watching me. Since we discovered the nest full of chicks and befriended their overworked parents, we’ve been throwing out the odd crumb for them to help with feeding the brood. Yesterday’s crumbs have long since vanished, and now Robin wants more. He’s standing three feet away from me, alternately pecking furiously at the obviously crumb-free concrete and looking up at me with one quizzical eye. "More," he’s saying; "give me more." It’s completely obvious to me that he’s not pecking compulsively; this is not the automatic, involuntary action you normally see in birds.

"Bird-brained" is no insult

At my most mechanistic, I’d have to say that he associates my presence with food, and that the pecking is like the salivation of Pavlov’s dogs. Nevertheless, even the most ardent behaviorist would acknowledge that this is a complex and sophisticated piece of processing, and anyway, I don’t believe it’s that simple for an instant. For a start, Robin knows what kind of a thing I am. He’s not looking at my knees, but straight into my eyes. He knows I’m another creature, and his behavior is taking that into account. We’re also not having this conversation where we throw the crumbs; he’s pecking away right in front of my feet, where he knows I can see him. Finally, he’s basing this association on a very small number of significantly varying trials and yet managing to extract the key features. Perhaps later I’ll do some experiments to see if I can fool him, but nothing will diminish my awe.

Whatever is going on in Robin’s head, no matter how trivial we try to make it sound, it’s all happening inside half a cubic inch of brain. That kind of processing power just totally staggers me. Despite the obsession anthropologists have with cranial capacity, size isn’t everything when it comes to brainpower. Just behind Robin is a pond full of large goldfish, whose brains are barely any smaller than his. Goldfish are, however, a great deal more stupid than birds - I despair of ever teaching mine to talk. Furthermore, I’m sure I know plenty of humans who seem to barely justify the fact that their brains are several hundred times the size of Robin’s!

Curiouser and curiouser

Human brainpower comes in many forms. Some people, of course, are both dull and stupid. Others are highly intelligent but not very bright - their minds are competent but not lively. On the whole, though, intelligence is proportionate with brightness, and bright, lively minds are by definition curious. Among higher animal species, curiosity and intelligence are quite often closely linked. Cats and birds are prime examples of curious and highly intelligent life forms. Goldfish are not curious - they might well come to the surface to find out if a disturbance means they’re about to be fed, but this is simple conditioning.

Real curiosity is all about persistently meddling with the world. In the absence of stimulation, uncurious creatures just sit there, like couch potatoes, until something happens. Curious animals, faced with a static environment, will go and perturb it, even at great risk to their safety. My avian friend Robin is taking a huge risk in trying to tempt me, a large predatory creature, to part with some food. Cats frequently get themselves stuck up trees or jammed in trash cans, or peppered with hedgehog or porcupine quills, all in the name of curiosity.

So curiosity and intelligence often go hand in hand. Is there a causal relationship here? Any coincidence between two occurrences always has at least four possible explanations:

• A and B are unrelated, despite the coincidence;
• A causes B;
• B causes A; or
• An unknown C causes both.

Which of these cases applies in what I perceive to be a relationship between intelligence and curiosity? Assuming they are linked, does intelligence cause curiosity, does curiosity cause intelligence, or does some unknown factor happen to create both? I believe that, to a degree, the answer is "yes - all of the above." If some factor C provides a trigger that encourages either A or B, then A will promote B will promote A, onwards and upwards in a spiral of positive feedback.

Curiosity has the effect of generating novel experiences, which lead to learning and thus improve mental competence. High mental competence in turn fuels curiosity, because it provides the wherewithal to manipulate the world and discover new things. Dumb creatures have only limited means of exploring their world; intelligent creatures such as cats can do much more. They can figure out how to open doors, remove lids, poke at porcupines, and thus open up many more fruitful opportunities to be nosy. Highly intelligent creatures such as humans can invent the microscope, take atoms to bits, and visit other planets, giving them huge (and productive) scope for their curiosity.

Curiosity increases risk to an organism, yet under some circumstances it can be a powerful survival trait. It really comes into its own in the case of scavenging omnivores, who are living life on the edge, and repeatedly have to discover novel sources of food or obtain food by devious means. Curiosity and ingenuity could well have been vital requirements for early hominids, for example, stressed by a changing climate and trapped by their geography. Those who looked, found. Those who figured out how to look better, found more. The need to find novel sources of food or other resources might have been the trigger that favored curiosity, which in turn led to a feedback loop of ever-increasing brainpower. Curiosity might occasionally kill the cat, but curiosity also created the cat, and maybe created us too.

The benefits of boredom

But this is not an anthropology magazine, so why am I mentioning this? Well, I wonder whether true intelligence will ever emerge from artificial systems that show no curiosity. It is now well recognized that intelligence cannot exist in a vacuum - that systems need to be grounded in some kind of rich, noisy environment. Intelligence cannot manifest itself in the absence of environmental complexity on which to act. Moreover, I seriously doubt if systems embedded in sparse, simple, or abstract environments can ever even become intelligent. This environmental demand is, more than anything, why birds are smarter than fish. Autonomous agent, animat, and robotics researchers are becoming increasingly cognizant of the need for a rich playground for their creations, even the need for it to be populated by other, socially interacting agents. But perhaps even this is not enough. Even a complex, nonlinear environment can go horribly quiet after a while, and agents can turn into little couch potatoes, sitting idle and waiting to be spoon-fed. We’ve seen what a couple of generations of TV can do for us humans, feeding us knowledge but draining away our self-motivation. I’m frequently dismayed by the lack of curiosity I see in people, and sometimes fear that the upward cycle of curiosity feeding intellect feeding curiosity has been broken in many cases, to be replaced by a downward spiral into comfortable oblivion. So what hope do synthetic systems have, if they aren’t given the gift of curiosity?

Part of the trick to making machines curious is simply to make them self-motivated, and the key to that is to get them bored. The little artificial life forms I created for a computer game (see "Creatures: an exercise in creation," IEEE Expert, Vol. 12, No. 4, July/Aug. 1997, pp. 19?24) had a wide range of drives and needs, linked to the reinforcement scheme that fed back learning into their neural networks. One of those drives was boredom; anything that increased boredom caused punishment, while things that decreased the drive caused reward. This motivated the creatures to learn to do things that increased their level of stimulation. Even then I had to ensure that they were kick-started into action in the first place, by genetically programming in some "instincts" - force-learned reactions such as "if you are bored, don’t stand still." I’ve no doubt that cats, robins, and humans are born with similar circuitry wired into their brains.

Thanks to the boredom drive and a few basic instincts, my little creatures show many of the symptoms of curiosity. Because they are more acutely aware of objects they have never seen before, they do tend to satisfy their drive to avoid boredom by experimenting with new things. What I didn’t manage to instill was the ability to seek out new things that aren’t already visible. I did add another instinct to encourage bored creatures to travel, but I don’t think they’d be capable of, say, breaking something open to see what’s inside, like a cat or a curious child would. Nevertheless, satisfying their curiosity leads them to experience new things and learn new facts about their world, making them more competent. At the same time, the rewards that come from such "intellectual" inquiry teach them that being curious is a good thing. Consequently, there are the glimmerings of a feedback loop, in which boredom triggers a bootstrapping of both curiosity and intelligence.

To compute, perchance to dream

How you would translate these ideas into a nonembodied (for example, desktop computer-based) intelligent system, I really don’t know. A desktop computer is perhaps the ultimate couch potato - unless you do something to it, it just sits there. Perhaps the design of even the most abstract and specialized intelligent system should include some attempt to overcome this and to enable software to be self-motivated and curious. Desktop computers lack motor systems, so they cannot direct their senses or manipulate their surroundings to show curiosity toward their external environment. Perhaps they should at least try to show an interest in their internal environment? Perhaps instead of just crunching and regurgitating the data they’ve been spoon-fed, they should retain it, play with it, and test out hypotheses on it. Perhaps their showing curiosity about their "memory environment" is equivalent to dreaming.

Either way, a machine that demonstrates nothing but passivity hardly looks very intelligent, and perhaps cannot actually be intelligent, regardless of what its designers might claim or hope for it. Proactive behavior seems like an essential component of an intelligent system. In the form of curiosity, it is highly characteristic of the more intelligent natural systems, including us. Why else are any of us involved in trying to create artificial intelligence, anyway, if it’s not to satisfy our curiosity?

My curious little robin has gone, now, perhaps scared off by an equally nosy cat. No, wait, I see why he’s vanished - a dark shadow flitting overhead resolves itself to reveal a large Heron, who’s obviously curious to discover what’s in my fishpond. Not such a sublime kind of curiosity as a robin who is testing to see if I’ll give him breadcrumbs, nor as profound as an astronomer who wants to discover what goes on in the farthest reaches of space, but it’s a kind of curiosity nonetheless, and seems like the mark of a truly intelligent creature. Hey! Wait a minute, though! What are you doing? Come back here with that goldfish!

Copyright © 2004 Cyberlife Research Ltd.
Last modified: 06/04/04