by Jeff Stibel
III
We can’t create intelligence until we know the answer to a fundamental question: What characterizes human thought? “A period of mulling,” said the late University of Chicago professor Howard Margolis, “followed by periods of recapitulation, in which we describe to ourselves what seems to have gone on during the mulling.” In other words, the human mind thinks in a series of loop-de-loops. The brain is wired to be parallel, which allows our thoughts to be recursive.
As a result, the human brain is a lousy computing machine. To behave like the human brain, a computer would have to do this: Start searching for an item with fierce concentration, then back off a little, then jump back in, then find itself staring blankly out the window, then off to a warm reverie—shafts of sunlight bouncing off the green grass or something like that—then suddenly, bang, back to reality with an abrupt epiphany: “Got to put Puppy Chow on the grocery list.” That’s how humans think. Even logical thinking, the kind you might expect from a rocket scientist or a McKinsey strategist, is more like a playful dolphin doing loop-de-loops and acrobatic tumbles than a shark torpedoing toward prey. We can’t help it. That’s the way the brain works.
Pulitzer Prize–winning brain scientist Douglas Hofstadter describes the process in his book I Am a Strange Loop. He argues that consciousness is an endless loop, where the brain takes the information it’s fed from the environment and other brains and constantly edits it in an elusive and self-repeating manner. In other words, thoughts travel from brain to mouth to ear to brain, round and round, until internal consciousness arises.
Not coincidentally, this is also how we learn. “We human beings have used our plasticity not only to learn, but to learn how to learn better,” Dennett says. We repeat and repeat and repeat something until it becomes better and better and better. This is the primary thesis behind Malcolm Gladwell’s bestseller Outliers, in which he argues that the greatest achievements of humankind have come not from genius but from reiterative practice.
For an artificial brain to be similar to a human brain, it must have the same loopy and iterative processes. Humanlike thinking will not come from more powerful computers or from building on the strengths of artificial intelligence but from a network approach that mimics the weaknesses of human thinking. In other words, we need to create a machine that stops calculating from time to time to gaze out the window. We don’t need supercomputers to do that.
If we can create a machine that guesses, fumbles, rounds off, and is not very good with numbers, we will be closer to replicating the human mind. We will also need the machine to be recursive: it should continuously edit itself, make little changes, test out potential answers against problems, and discard losing ideas. Most importantly, we need a machine that learns through repetition, and that would prefer to be half-right than completely right. In other words, we need a prediction machine.
IV
The brain is a slow, inefficient machine. Transmissions to the cerebral cortex range from one to thirty meters per second along the axons and about one-third of a meter per second along dendrites. In comparison, light travels at about 300 million meters per second. So the brain is quite sluggish compared to the transmission speed of a computer or a fiber optic network. Moreover, it takes a neuron two thousandths of a second to snap on and off in your head. A computer does the same thing a million times faster. Neurons fire on average at about one hundred times per second (technically they can get up to several hundred pulses per second, but then they retreat in exhaustion). This pales in comparison to the transistor in your smartphone.
Given the brain’s slow processing speed, humans must try to keep predicting what will happen. For its part, the brain helps us to guess better by rewarding our correct guesses. The rewards are little spurts of dopamine (the same substance that is overproduced when an addict uses a drug like heroin) that are distributed throughout the brain’s 500,000 dopamine neurons. This reward process coaxes out smart thinking.
Suppose you have a piece of chocolate in front of you. Your brain predicts that it is going to taste good, based on past experience, and your hand reaches for it and drops it into your mouth. Over time, the brain makes other associations and connections so that, eventually, even the word Godiva yields a similar biological reaction.
The results are similar for emotionally laden concepts such as freedom of speech, higher taxes, motherhood, or apple pie. For many of these, the brain’s reaction is preconfigured based on past experience: the pattern is set beforehand because the brain has already placed a value on the concept. Therefore, the brain doesn’t have to do any fresh thinking. It has predicted its reaction.
Humans use anticipation every minute of every day, for both big and small things. When you spin a cup of coffee around so that the handle has moved from the left to the right side, your mind doesn’t go about reexamining the entire cup. It doesn’t need to start from scratch. It knows from previous experience that the mug of coffee is the same. It just fills in the change in the handle. Similarly, when we step out the front door in the morning, our brains know what to expect. Memory serves us well; the patterns are there. If you step out your front door in the morning and find a dead body on your sidewalk, you will certainly note that. But you don’t examine the oak tree in the front yard as though it had never existed before.
What’s really interesting about this is that Plato, in his theory of forms, stated that a perfect tree, a perfect flower, a perfect model of everything exists in the ether of the heavens. For centuries, philosophers have explored the meaning of this. But now, brain science offers some new enlightenment: our brains hold perfect representations of things—memory patterns—that can be efficiently called upon. To that prototypical image, the brain merely makes a quick comparison, noting whatever is new. These memory patterns, of course, enable predictions, preemptive expectations of what we will see. They are, in the words of Steven Pinker, “the internal simulation of possible behaviors and their anticipated consequences.”
Making predictions is not just the work of the brain regions that enable reasoning and decision making. Many predictions are driven by the amygdala, an almond-shaped cluster of interconnected structures perched above the brainstem near the bottom of the limbic ring. The amygdala is the seat not of reason but of passion and emotion. It turns out that emotion plays an important role in predictions. In fact, in times of emergency, the amygdala often springs into action before the rationally thinking neocortex has time to make a decision.
Imagine seeing a rattlesnake in your path: the visual signal instantly goes from the retina to the visual cortex, which gets the first pass at the coiled object. Then it passes the information on to the neocortex for analysis and to the hippocampus for storage in memory. All of that is straightforward. But recently, researchers have noted that part of that signal goes in a different direction. Here we have a perfect example of parallel processing in action. The information also goes directly from the visual cortex to the highly emotional amygdala. Since the amygdala can house memories, those memories can make us respond even without “knowing” the reason. We don’t wait around to finish analyzing the slithering object on the ground; we jump.
“While the hippocampus remembers the dry facts, the amygdala retains the emotional flavor that goes with those facts—that means that the brain has two memory systems, one for ordinary facts and one for emotionally charged ones,” says psychologist and author of the widely acclaimed book Emotional Intelligence, Daniel Goleman. “Just as there is a steady murmur of background thoughts in the mind, there is a constant emotional hum.” What’s particularly interesting is that, unlike other parts of our brains, the amygdala is fully formed at birth. It’s so important to our survival that it was given developmental priority.
Our brains are far more distributed than we once thought. We have a brain that sees patterns rather than individual pixels of information; a brain that is anticipating with pre-stor
ed knowledge; a brain that has intuition and insight. And the really good news is that, with time, it only gets better. Sure, the neurons in our brains die, but our wisdom blossoms.
V
The brain can arrive at answers through intuition that no computer, regardless how powerful, could ever conceive of. The internet, in contrast to computers, is already scratching the surface of the brain’s insight through intuition. Several companies are building software online that leverages the brain to create what we think of as human consciousness. Brain scientist Doug Lenat has been working on one such intelligent system, called CYC, for the past 20 years. Ask Lenat when he thinks CYC will be conscious, and his response is bold: “I think it’s conscious now.”
For the internet to be a true brain, it needs to combine calculation, communication, prediction capabilities, and loopiness. When these functions can perform in parallel in the haphazard way of the brain, humanlike intelligence will spread out across the internet. As Google’s chairman Erik Schmidt highlighted many years ago, “when the network becomes as fast as the processor, the computer hollows out and spreads across the network.”
Some systems on the internet will almost certainly reach the level of consciousness that we reserve only for the smartest of animals, including humans. But despite the optimism of Doug Lenat, we are not there yet. We are close, however. We have perfected calculation and memory to a level greater than human capacity. The internet’s ability to communicate is also advanced; in many respects, it is approaching 80 percent of human capacity. The prediction capabilities of the internet have grown stronger over time, although they are still only about 30 percent of what a human mind can do; good but not yet good enough. As scientists and businesses continue to work on these elements, we move closer to the gray area of intelligence, and eventually the internet will have enough gray matter itself to emerge with wisdom and consciousness.
By far the weakest component is the concept of loopiness—the ability for us to combine disparate pieces of information into a coherent pattern. But even there, great strides have been made. The journal Science reported in 2012 on a new brain model called Spaun. This neural network can mimic behavior, recognize syntax, and determine patterns by replicating the loopiness of the human brain. Spaun achieves this using only 2.5 million synthetic neurons. While this pales in comparison to the human brain, Spaun can predict the answers to IQ questions such as filling in the pattern of “2, 4, 8, 16, 32, __.” Spaun has a 94 percent accuracy rating for image recognition, as compared to a control group of humans who have roughly a 98 percent accuracy rating. It is not 100 percent accurate, and that is the point—Spaun is replicating an imperfect mind.
Spaun’s methods are loopy: the scientists did not use standard computer software to teach it how to calculate; instead, they created spiking neurons that mimic the behavior of humans. As the lead researcher, Dr. Chris Eliasmith, remarked, “This model is trying to address that issue of cognitive flexibility. How do we switch between tasks, how do we use the same components in our head to do all those different tasks?” He went on to add that Spaun’s mistakes, not its accomplishments, are what is most surprising. The process of finding a solution is quite haphazard for Spaun—it uses endless loops and makes the same errors that the human brain does. It takes in all the available information, discards what it deems irrelevant, meanders a bit, and comes up with an answer that is occasionally foolish, but often correct.
The New York Times reported that, “from the billions of documents that form the World Wide Web and the links that weave them together, computer scientists and a growing collection of start-up companies are finding new ways to mine human intelligence.” We are living in an epic time, in which machines are growing more intelligent every day. We are approaching the time in which all the information on the internet will be crumpled like our paper model of the brain, where patterns will be established, and where multiple drafts will live and die. The result will be the emergence of intelligence.
Acknowledgments
Breakpoint grew out of an article I wrote for the Harvard Business Review, and much of the thesis has been influenced by the articles I have written since, as well as the feedback I have received from my editors, peer reviewers, and commentators at Harvard. The book could not have been completed without the tremendous help of Lindsey Long. When she started this process, she was my executive assistant, but she has found her calling working on the book, which will inevitably lead to more writing-intensive roles. Lindsey’s tireless work improved the book in more ways than I can describe, and she participated in virtually every aspect, including challenging me on many of the underlying theories. I also owe thanks to my editor, Laurie Harting, whose efforts improved the book with each of the many drafts she reviewed. Early versions of the book were also reviewed by a number of individuals who provided constructive feedback: William Allen, Paul Allopenna, Lydia Blalock, Erik Calonius, Peter Delgrosso, Dan Dennett, Reid Fahs, Finn Faldi, John Griffen, Judy Hackett, David Hughes, Alex Kazerani, Caitlin Mason, Sara Mathew, Lars Pettersson, Cheryl Stibel, and John Suh. The graphs and illustrations for the book were produced by the tremendously gifted Mike Samuelsen. Additional graphics, the Breakpoint website, and related marketing materials for the book were produced by a number of incredibly talented individuals: Don Berkman, Bernice Brennan, Chad Buechler, Tatiana Camacho-Daniel, Melissa Halim, Jordana Haspel, Brittany Johnson, Samantha Lim, Dustin Luther, Catherine Mangan, Brandon Mills, Mike Samuelsen, Lauren Simpson, Alexander Staikos, and Sam Yacov. This team was led by Judy Hackett and Liz Gengl, both of whom I owe much gratitude. Particular thanks as well to my agent, Jim Levine, who has now shepherded me through two successful books. My first book was edited and published by the fantastic team at Harvard Business Press, and I owe a special thank you to them for allowing me to reuse and adapt some material that appeared previously in Wired for Thought.
In many respects, I tried to take readers of Breakpoint to the intersection of the brain, biology, and technology. As a result, I relied on a number of experts in those fields to help guide my writing. They include Paul Allopenna, Jim Anderson, Dan Ariely, Dan Dennett, Andrew Duchon, Carl Dunham, Deborah Gordon, David Landan, George Miller, Steve Reiss, and John Santini.
I owe a debt of gratitude to the entire team at the Dun & Bradstreet Credibility Corporation but in particular to the senior team—Bill Borzage, Sue Collyns, Pete Delgrosso, Hari Ganapathy, Judy Hackett, Wisdom Lu, Chris Nowlin, Sam Paisley, Aaron Stibel, and Gonzalo Troncoso—who put up with me while I was working on the book during many nights and weekends. I also want to thank Great Hill Partners in general and my board in specific (Michael Kumin and Chris Gaffney, as well as Peter Garran and Chris Skarinka). They encouraged me to complete Breakpoint in part because they knew it would improve our business: the partners at Great Hill are the rarest type of investors who understand the benefits of stepping back to gain perspective.
Nothing that I do would be possible without the support of my family. Thank you to my mom, dad, Aaron, Travis, Kristen, and Rick for always being there. I owe my wife and kids a particular thanks. Cheryl read numerous drafts of the book, and my 5- and 7-year-old children, Dennett and Lincoln, sat through two hour-long presentations on the subject and managed to provide a surprisingly critical review. My family contributed significantly to the book, but more importantly, they contribute significantly to my life.
Appendix
To learn more about Breakpoint:
Visit the website at http://www.BreakpointBook.com.
Become a fan at our Facebook page http://www.facebook.com/BreakpointBook.
Follow our Twitter feed @Breakpoint for breaking news.
Use the hashtag #Breakpoint to send a tweet to the Breakpoint community.
To learn more about Jeff Stibel, please visit www.Stibel.com or follow him on Twitter @Stibel. Learn more about his 2009 book, Wired for Thought (Harvard Business Press
), at www.WiredForThought.com.
Expand your network by following these great thinkers and organizations on Twitter:
Dan Ariely (@DanAriely)
Richard Dawkins (@RichardDawkins)
Dan Dennett (@DanielDennett)
Daniel Goleman (@DanielGolemanEI)
Steven Johnson (@StevenBJohnson)
Kevin Kelly (@Kevin2Kelly)
Ray Kurzweil (@KurzweilAINews)
Marissa Meyer (@MarissaMeyer)
Steven Pinker (@SAPinker)
Matt Ridley (@MattWRidley)
The Defense Advanced Research Projects Agency (@Darpa)
Discovery News (@Discovery_News)
Harvard Business Review (@HarvardBiz)
National Science Foundation (@NSF)
Nature (@NatureMagazine)
Popular Science (@PopSci)
ScienceDaily (@ScienceDaily)
SciAm Mind (@SciAmMind)
Scientific American (@SciAm)
TechCrunch (@TechCrunch)
Wired (@Wired)
Wired Science (@WiredScience)
Others not on Twitter and websites of interest:
Association for the Advancement of Artificial Intelligence
http://www.aaai.org
BrainGate
http://www.BrainGate.com
Malcolm Gladwell
http://www.gladwell.com
Deborah Gordon’s Lab
http://www.stanford.edu/~dmgordon/