Gary F. Marcus Quotes

To mention a colorful example, the nineteenth-century German scientist Karl Vogt once wrote that “thoughts stand in the same relation to the brain as gall does to the liver or urine to the kidneys.” When he expressed this idea in public, a philosopher interjected that the longer one listens to Professor Vogt, the more one tends to believe him. Clearly, more sophisticated ideas and models are in demand.

The CD-ROM's worth of information in the genome really wouldn't be enough to paint a bitmapped picture of an embryo, but it is enough to describe a process for building one. An artist who only wants to paint a picture that looks like a kind of tree has much less to remember than an artist who wants to paint a particular Ponderosa Pine from memory; in a similar way, if some alien's genome had to encode every cell in a body, it would need much more information (many more nucleotides) than our genomes do, because ours specify a general way to build a creature rather than an exact picture of every detail of the finished product. Our genomes are lossy because they specify methods rather than pictures, but it is precisely that lossiness that allows them to so efficiently supervise the construction of complex biological structure.

What propels an embryo from one stage to the next-and makes one species different from another-is not a blueprint but rather an enormous autonomous library of the instructions contained within its genome. Each gene does double duty, specifying both a recipe for a protein and a set of regulatory conditions for when and where it should be built. Taken together suites of these IF-THEN genes give cells the power to act as parts of complicated improvisational orchestras. Like real musicians, what they play depends on both their own artistic impulses and what the other members of the orchestra are playing. As we will see in the next chapter, every bit of this process-from the Cellular Big 4 to the combination of regulatory cues-holds as much for development of the brain as it does for the body.

"In universities and pharmaceutical labs around the world, computer scientists and computational biologists are designing algorithms to sift through billions of gene sequences, looking for links between certain genetic markers and diseases. The goal is to help us sidestep the diseases we're most likely to contract and to provide each one of us with a cabinet of personalized medicines. Each one should include just the right dosage and the ideal mix of molecules for our bodies. Between these two branches of research, genetic and behavioral, we're being parsed, inside and out. Even the language of the two fields is similar. In a nod to geneticists, Dishman and his team are working to catalog what they call our "behavioral markers." The math is also about the same. Whether they're scrutinizing our strands of DNA or our nightly trips to the bathroom, statisticians are searching for norms, correlations, and anomalies. Dishman prefers his behavioral approach, in part because the market's less crowded. "There are a zillion people looking at biology," he says, "and too few looking at behavior." His gadgets also have an edge because they can provide basic alerts from day one. The technology indicating whether a person gets out of bed, for example, isn't much more complicated than the sensor that automatically opens a supermarket door. But that nugget of information is valuable. Once we start installing these sensors, and the electronics companies get their foot in the door, the experts can start refining the analysis from simple alerts to sophisticated predictions-perhaps preparing us for the onset of Parkinson's disease or Alzheimer's."

Honey bees, too, use a highly specialized learning mechanism to help them figure out where they are going: the difference is that their system works based on the trajectory of a single star, our very own sun. Once again, part of the system is prewired, but part of it requires learning. The prewired bit is a mathematical function that relates the sun's position on the horizon to to a bee's orientation-but some of the values of the equation must be set, which is where learning comes in. What the bee learns is a highly specific bit of information about the sun's trajectory at the bee's particular latitude at a particular time of year. A five o'clock winter sun in Boston means something very different from a five o'clock summer sun in California, and a highly focused learning mechanism allows honeybees to take advantage of that information. We know that bees don't simply memorize a correspondence between particular places on the horizon and particular headings, because bees that have been raised in conditions in which they are exposed only to morning light can accurately use the sun as a guide during evening light.

In essence, the bee's azimuth system acts like a sundial run in reverse, and like a sundial, it has to be calibrated. A sundial, which must be oriented with respect to a known compass direction, calculates the time of day based on where the sun is; the navigational centers in the bee's brain calculate where the sun should be based on the time of day. As a consequence, the one thing that bees can't cope with is the discalibration that results from jet lag. In a famous 1960s experiment, Max Renner packed up a hive of bees in Long Island, New York, flew them to Davis, California, and tested their ability to navigate with the sun as a landmark. The jetlagged bees consistently misoriented themselves by 45 degrees, precisely as though they believed it was three hours later. The complex circuitry that allows the bee to use the sun as a guide is built in, but it is not that genes trump the environment (or the other way around). Instead, genes enable creatures to make sensible use of their particular environment. Learning is not the antithesis of innateness but one of its most important products.

Our sense of a composition largely inheres in how we feel about the individual parts; narrative arcs are almost always essential in drama but (unless there are lyrics involved) often less essential in music. All of this is, I suspect, again symptomatic of human memory limitations. We live, to a remarkable degree, in the present; what happened thirty seconds ago is already rapidly fading from our memory (or at least rapidly becomes harder for us to retrieve).

Memory, perception, coordination, strength: guitar is just plain hard.

But the real story is how narrow Duplex was. For all the fantastic resources of Google (and its parent company, Alphabet), the system that they created was so narrow it could handle just three things: restaurant reservations, hair salon appointments, and the opening hours of a few selected businesses. By the time the demo was publicly released, on Android phones, even the hair salon appointments and the opening hour queries were gone. Some of the world’s best minds in AI, using some of the biggest clusters of computers in the world, had produced a special-purpose gadget for making nothing but restaurant reservations. It doesn’t get narrower than that.

More broadly, formal logic of the sort we have been talking about does only one thing well: it allows us to take knowledge of which we are certain and apply rules that are always valid to deduce new knowledge of which we are also certain.

Because genes work in combination, the incremental effect of adding a new gene to a genome may be not linear, but exponential.

Language builds on our cognitive capacities to reason about the goals and intentions of other people, on our desire to imitate, our desire to communicate, and our twin capacities for using convention to name things and sequence to indicate differences between differing possibilities.

Biology doesn't know in advance what the end product will be; there's no Stuffit Compressor to convert a human being into a genome. But the genome itself is very much akin to a compression scheme, a terrifically efficient description of how to build something of great complexity-perhaps more efficient than anything yet developed in the labs of computer scientists (never mind the complexities of the brain, there are trillions of cells in the rest of the body, and they are all supervised by the same 30,000-gene genome). And although there is no counterpart in nature to a program that compresses a picture into a compact description, there is a natural counterpart to the program that decompresses the compressed encoding, and that's the cell. Genome in, organism out. Through the logic of gene expression, cells are self-regulating factories that translate genomes into biological structure.

"Sometimes guitar riffs get repeated over and over ("vamping," in the lingo of musicians), but generally there is a soloist proving variation that runs above that background, lest the song sound monotonous. Philip Glass's minimalist compositions (such as the soundtrack to 'Koyaanisqatsi') deviate from much of the classical music that preceded them, with much less obvious movement than, say, the Romantic-era compositions that his work seems to rebel against, yet his works, too, consist not only of extensive repetition but also of constant (though subtle) variation. Virtually every song you've ever heard consists of exactly that: themes that recur over and over, overlaid with variations."

One key lesson learned from mapping the genome is that access to a rough initial map proved crucial to developing more detailed maps of small individual human differences.