I WAS RECENTLY standing in the hallway with my two-and-a-half-year-old neighbor. He pointed to the ceiling and said, “Smoke detector.” I was amazed and had to ask myself what kind of parents did this little boy have. Did they perhaps subject him for weeks and weeks to thousands of pictures of smoke detectors, always repeating the series of images until he was finally able to identify the similarities and characteristics of smoke detectors and to correlate the object? His father is, admittedly, a fireman and so my neighbor already has a certain predisposition toward fire safety tools. But still, had this little human really been bombarded with thousands of pictures of smoke detectors, fire extinguishers, and fire axes that then enabled him to quickly identify the required implement for the next possible crisis? And then did they send him down the hall in my direction once he had finally passed the test with flying colors? No way! That’s not how it works. But the question still remains: How was my little neighbor able to identify a smoke detector in a completely new context after only seeing a smoke detector maybe two or three times in his short life?
The answer is that my neighbor did not learn about smoke detectors in the same way that a computer does, rather he understood the idea of smoke detectors. This is something which humans are very good at and which science calls “fast mapping.” If, for example, you were to give a three-year-old child never-before-seen artifacts and explain that one very special artifact is named “Koba” or comes from the land of “Koba,” the child will remember the Koba object one month later.9 After only seeing it one time! It gets even better if the child is learning to understand new actions and not only new words. Children who are only two and a half years old require only fifteen minutes of playing with an object before they can transfer its properties to other objects. For example, a child who realizes that they can balance a plastic clip named “Koba” on their arm later realizes that a similar clip, but with a slightly different shape, is also called a “Koba” and can be balanced on one’s arm.10 The whole exchange only takes a few minutes. How would two-year-olds possibly be able to learn an average of ten new words a day if they had to practice each word hundreds of times? No brain has that much time on its hands.
Of course, the brain cannot simply learn something from nothing. From what we currently know, we assume that learning by “fast mapping” allows new information to be rapidly incorporated into existing categories (presumably without even bothering the hippocampus, the memory trainer that you learned about in the previous chapter).11 But we are even able to create these categories very rapidly—whenever we give ourselves time for some mental digestion. If you present a three-year-old with three variations of a new toy (i.e., a rattle with different colors and surfaces) one right after the other and give each of these the artificial designation of “wug,” the child will not easily be able to identify a fourth rattle as a “wug.” If, however, the child is allowed half a minute of time between the presentation of each new rattle to play with the item, he or she would then grasp the concept of the wug and be able to identify a new, differently shaped and differently colored rattle as a wug. This seemingly inefficient break, this unrelated waste of time that we would love nothing more than to rationalize away in our productivity-optimized world—this is our strength—if, in fact, we hope to be able to accomplish more than a mindlessly learning computer.
We are very quick to understand categories and are able to grasp the relationship between words, objects, and actions almost immediately. You don’t believe me? Do you still think it’s only possible to effectively learn something by repetition and practice? Then allow me to give you a counterexample: How long did it take you to understand a newly coined word like “selfie”? A single experience of seeing four posing teenagers snapping a photo of themselves on a smartphone should have been enough. How quickly were you able to understand the invented word “Brexit”? You probably figured it out fairly quickly. We often understand the world at first glance, but there’s more. Once you’ve understood something, not only can you reproduce it, you can also make something new from it. If Brexit describes the exit of Great Britain from the European Union (EU), what would “Swexit,” “Spaxit,” or “Itaxit” indicate? Or from the opposite direction, what would “Bremain” or a “Breturn” mean? It’s a piece of cake for you to grasp all of the new words because you already understand the fundamental categories of thought. You are able to take these and immediately generate a new piece of knowledge, even if you’ve never heard of “Spaxit” before in your life!
So much for the topic of frequent repetition and “deep learning.” Merely memorizing a bunch of facts is no great art. Understanding them, on the other hand, is. In the future, computers might be able to “learn” about objects and pictures more quickly, but they will never be able to understand them. In order to learn, computers use very basic algorithms to analyze an enormous amount of data. Humans do the opposite. We save much less data but are therefore able to process exceedingly more. Knowing something does not mean having a lot of information. It rather means being able to grasp something with the information in hand. Deep learning is all well and good, but “deep understanding” is better. Computers do not understand what it is that they are recognizing. One interesting indication of this followed from an experiment conducted in 2015. Researchers studied artificial neural networks that had trained themselves to recognize objects (such as screwdrivers, school buses, or guitars). The network was analyzed to find out what, in fact, it had recognized. For example, what would a picture of a robin have to look like in order for the computer program to be able to respond with 100 percent certainty that it was indeed a “robin”? If anyone had expected that a perfect prototype image of a robin would pop out, a sort of “best of” from all the robin images in the world, they would have been disappointed. The resulting image was a total chaos of pixels.12 No human would be able to identify even a very rudimentary robin in such a pixelated mess. But the computer could, because it recognized the robin only as a graphic representation of pixels and did not understand that it was a living creature. If one taught a computer that Brexit refers to the exit of Great Britain from the EU, the computer would never be able to independently draw the conclusion that Swexit means the Swedes waving goodbye.
Our ability to learn extremely quickly, or we had better say, to understand things, is only possible if we do not “learn” facts and information separately, in a way that is sterile and detached, but rather by creating a category correlation that embeds things and, thereby, leads us to understand them. Computers do exactly the opposite. They are very good at saving data quickly, but they are just as dumb as they were thirty years ago. Only now, they are dumb a little faster. This is because they never take time to reflect on all of the data they have gathered. They don’t treat themselves to a break. Computers always work at full blast until they run out (or have their power switched off). But if you never take a break, you cannot ever put the information that you possess to any use, and thus you cannot acquire any knowledge. In order to generate concepts, it is essential to have a stimulus-free space (during sleep). We are able to recognize something at first glance because we don’t allow ourselves to be flooded with facts and data but, instead, make ourselves take a break. This may initially seem to be inefficient and perhaps to smack of weakness, but it is actually highly effective. In fact, this is the only way that we are able to comprehend the world, instead of merely memorizing it.
Learning power reloaded
WE SHOULD THUS not treat the brain as though it were an information machine since the most valuable learning processes of the future don’t call for us to have flawless memories (that this isn’t even possible is touched on in the next chapter), but rather for us to adjust rapidly to new situations. If we start competing with computers, trying to use learning tricks to memorize more facts, telephone numbers, and shopping lists, we are certainly going to lose. Maybe we should let algorithms take over these kinds of basic tasks for us.
Trying to develop the latest learning techniques in order to remember more information isn’t what’s important. It’s much more valuable to improve our ability to think conceptually and to understand. The brain is not a data storage device. It’s a knowledge organizer whose major talent can only be actualized once we stop treating it like
