I’ve been enjoying the debate about when our robot masters will take over the world for quite some time. And despite the fact that really smart people like Ray Kurzweil are convinced the singularity will take place in our lifetimes, I have to disagree. Vehemently. For the simple reason that we still don’t know what intelligence actually is, neither its baseline nor its limits, regardless of whether we’re talking about humans, animals, or machines.
There’s an equally simple reason why we don’t know what human intelligence is – we’ve been looking for it in all the wrong places for centuries, using the wrong tools, the wrong measures, and the wrong assumptions. And this false quest continues, as far as I can tell, in the current frenzy about AI. We’re still acting pretty stupid when it comes to understanding the concept of intelligence. Consistently stupid, if that’s any comfort at all.
One of the best ways to understand what little we know about human intelligence is to look at how humans understand animal intelligence. The quick answer to whether we grok animal intelligence is found in the title of a very engaging book by an eminent primatologist and cognitive scientist, Frans de Waal: Are We Smart Enough to Know How Smart Animals Are?
In posing this as a question, you can tell that de Waal plans to answer it in the negative. In the process of doing so de Waal exposes a huge gap in how humans – in the form of cognitive scientists, behavioral scientists, and the like – define human intelligence. As you may imagine, many of the idiotic studies of animal intelligence cited by de Waal started with embarrassingly false assumptions about what makes humans intelligent, and then went on to “prove” that animals, based on this false metric, were infinitely inferior to humans.
A simple example was the oft-repeated claim that animals are incapable of facial recognition for the simple reason that even higher-order primates can’t tell one human face from another. Which is of course, the wrong question to ask: why would any animal know how to recognize human faces? The reverse is certainly true. Humans suck at recognizing non-human faces: Unless you’re an experienced animal behaviorist, you’re also going to have trouble telling different members of a non-human species from one another (other than domestic animals, which live and work with us.)
But if you ask – experimentally – an ape, a monkey, a crow, even a wasp, to identify the members of its pack or herd or nest or “murder of crows,” these and other highly social animals are not just extremely adept at it, they are able to use facial recognition exactly as we do, as a means to manage an individual’s position in the social hierarchy (who is she relative to who am I and what should I do or not do about that in light of the current situation) and otherwise interact with the group.
How about that “unique” exemplar of human intelligence, tool use? In the late 1950’s it was assumed that only humans made tools, now we know that corvids and many other species, are not just tool makers, they’re meta-tool makers and users. In other words, they can make a tool that manipulates another tool in order to accomplish a particular task, usually involving acquiring food. It turns out crows can do this better than monkeys, which are capable of tool use but are not so good at understanding that meta-tool use involves a sequence of actions that start with finding the meta-tool and then using it to manipulate the original tool in order to get the object. Corvids, by this important measure of intelligence, are smarter than monkeys.
And both species, by the meta-tool use standard, are much smarter than human children, which can start using tools at 12-24 months, but need several more years to master meta-tool use.
I can go on and on. Peter Wohlleben writes about intelligent trees and forests in his book, The Hidden Life of Trees. It turns that trees of different species cooperate to share resources – water, nutrients, and even access to sunlight – amongst one another in order to maintain the collective health of the forest. This notion of altruism – yet another form of intelligence that was once the exclusive domain of humans – is mediated by a third party, a ubiquitous fungus called Rhizopogon . The filaments produced by this truffle-like fungus permeate the root structures of the forest trees, and become the conduits of the nutrients and water that are shared by the trees. (There’s an article from Scientific American that explains this phenomenon here.)
Wohlleben – and the author of the article above – describe what is clearly altruistic, collaborative behavior on the part of different species of trees in order to reach common goals, like closing a hole in the forest canopy that can dry out the forest floor and lead to the invasion of parasites and plants that could endanger the existing trees. Their altruism can be measured by the differential flow of nutrients along the filament paths laid down by the fungus, a flow that scientists have been able to quantify and correlate with activity that is clearly in the category “all for one and one for all.”
Altruism is a clear sign of intelligence – it takes varying degrees of planning, foresight, self-awareness, and inference to do it right – and either the forest is able to mechanistically respond to threats or it does so through some form of intelligence. Either way, it’s another form of intelligence once deemed to be the sole purview of humans that, in this case, exists in the interplay between a fungus and a group of trees.
Even more interesting is the visualization of this network: lay this forest network (on the right) side by side with a visualization of a human social network, such as the one I selected that shows the interlocking board memberships of major US corporations (on the left), and the similarities are obvious.
(source: Corporate network: http://plutocratsandplutocracy.blogspot.com/2016/05/the-power-elite.html. Wood wide web network: https://blogs.scientificamerican.com/artful-amoeba/dying-trees-can-send-food-to-neighbors-of-different-species/)
Assuming there is intelligence driving our network of the corporate world, can we surmise that there is a form of intelligence at work in the forest? And if we do, where does the intelligence come from in the forest network, which is referred to, tongue in cheek, by researchers as the wood wide web? Is it in the trees? The fungus? Both? Neither?
Many scientists, philosophers, and ethologists are tempted to assume that this wood wide web must be mediated by a mechanism of chemical imbalances, osmosis, and other non-intelligent forces. This mechanistic view of the animal world, promulgated by Descartes in the 17th century, certainly makes it easy to ignore or dismiss Wohlleben’s premise. But, like the case of animal face recognition and meta-tool use, Occam’s razor isn’t the best tool to use when looking at the intersection of behavior and intelligence: the simplest solution turns out to be too simplistic.
Considering how literally every passing year brings us more and greater revelations about how non-human intelligence is encroaching on our sacred perception of an anthropocentric world of human masters and animal automatons, we should try to apply a little skepticism about what the advent of so-called artificial intelligence is really about. Call it automation, call it the new process efficiency, anthropomorphize it with names like Einstein, Leonardo, Claire, or Screaming Jay Hawkins, but don’t call it intelligence. Not until we understand what human intelligence really is.
This is why I take great comfort in the response by the most intelligent Grady Booch when I asked him about the singularity on stage last year. “We will never see it coming,” Booch opined, “…because we ourselves will have co-evolved.” Whatever our notion of intelligence is today, it will be different, and continue to be different, as we evolve. I kind of like that idea, it gives me hope for the human race.
So when I see headlines like this one, Computers are Getting Better at Reading than Humans, I wince. The article imputes intelligence where there isn’t, and impugns human intelligence in the process. One example of this machine that is supposed to be smarter than humans is that it successfully scanned an entry in Wikipedia on Dr. Who and then correctly identified a discrete piece of information from the article, in one example the name of Dr. Who’s ship. In no way does this justify the breathlessly hyperbolic tone of the headline.
The reason such headlines exist is precisely because we hardly know what human intelligence really is, and so we mistake gimmicky behavior on the part of a machine as a sign of superior intelligence. Having a machine best a human at chess or the game of Go sounds impressive, and it is on a certain level. But, as de Waal reports, a highly trained chimp was able to beat a human memory champion at a memorization“ game” with relative ease. Should we assume that the chimp is more intelligent than the human? It depends on the task at hand, doesn’t it?
And is winning at Go more impressive than the dog that, running at full tilt across broken ground, calculates the precise speed and trajectory of a ball and leaps at the precise moment to catch it in its mouth, and then nails a perfect four point landing? It takes some seriously furry calculus to pull that trick off.
If you’re not impressed, watch the scene in Hidden Figures when the “human computer”, Katherine Goble (a black woman played by actress Taraji Henson, and please remember what much of society thought then about the intelligence of non-white races,) goes to the blackboard and computes the exact landing point of the Mercury capsule in front of an astonished room full of space program officials. The scene, though apocryphal, signifies the moment when human intelligence evolved to where it could model something that every spit-covered tennis ball-loving dog does as a matter of course. Ms. Goble’s on-the-fly calculation is truly a tour-de-force of human intelligence, and obviously, Fido can’t calculate on a chalkboard where and when and how high she needs to jump to snag the ball. But, then again, she doesn’t need to, she just does it “naturally”.
So what is intelligence? Beats me. But what I do know is that it’s easy to think we know something when we don’t. Danny Kahneman, a cognitive psychologist, was awarded a Nobel prize in Economics for proving that we humans are hardly as intelligent as we think we are, and, looking around me, I’m definitely with Kahneman on that one.
What I do know is that we really need to dial back on what we think so-called AI can do. I like the conceptualization that my friend Trevor Miles, the resident deep thinker at supply chain software vendor Kinaxis, uses for AI. The goal of AI, Trevor likes to say, is to “take the robot out of the human,” as opposed to replacing humans with robots. Works for me. So let’s cut out the nonsense about superior intelligence, that’s really not what’s at play in the world of AI. Nor is it necessary to even think in those terms. If all we do is remove the robot from the human and reduce the drudgery and risk of error that comes with trying to handle too much information to quickly, it will be a huge win.
I’ll finish my rant about what is intelligence with a personal note. My youngest sister, Susannah, is developmentally disabled, what we used to call in pre-PC times “retarded.” It was assumed 50-plus years ago when she was born that people like her, just like people on the autism spectrum, the visually impaired, Helen Keller, people of color, and anyone else who is seen as the “other”, are inherently stupid and sub-human. We now know better. Much better. Susannah is able to lead a relatively rich life, certainly not the institutionalized one that we were told was our only recourse when she was born, and in the process has grown to demonstrate traits like sympathy, a sense of humor and a sense of wonder, and other forms of intelligence that people like her weren’t supposed to have.
All this is to say that we when we get all a-flutter about the march of AI towards the inevitability of a singularity that will leave us as a bunch of neutered idiots – to join, no doubt, those benighted fools, idiots, cretins and retards of a previous era – facing a purposeless future, or we’re tempted to bow and scrape at the prospect of a machine that “reads better than a human,” we need to stop and ask ourselves the following question:
Are we smart enough to know what artificial intelligence is?