LEARNING CODE
I appreciate “Making AI More Human,” Alison Gopnik's article about the two ways that artificial intelligence is being configured to approach learning. In top-down methods, such as Bayesian models, abstract concepts are used to create a hypothesis and predict which patterns of data should be seen if it is true. Meanwhile in bottom-up methods, such as “deep learning,” abstract concepts are derived by looking for patterns in concrete data.
Whereas the top-down approach makes lots of sense because that's how we learn most things in school, the bottom-up one remains quite mysterious, as mysterious as how a child learns his or her mother tongue. So a machine is shown thousands of pictures of, say, the letter A, and through brute repetition it starts to know that when the pixels are arranged just so, that's an A. But what is going on inside the machine? Is it postulating a series of guesses and thus creating its own rules? And if that is the case, then what logical tools does it have available to create and test the guesses?
On supporting science journalism
If you're enjoying this article, consider supporting our award-winning journalism by subscribing. By purchasing a subscription you are helping to ensure the future of impactful stories about the discoveries and ideas shaping our world today.
JAMES LOEWEN Oakland, Calif.
Assumptions about fabricating an artificial brain ignore what little we know about how the human brain thinks, feels and acts, and we cannot view it as a slower version of a computer. The brain is a soft, squishy organ, which has evolved over millions of years, and its synaptic connections are not electronic but electrochemical (hence our sluggish ability to solve equations, compared with a computer's speed).
Scientists working on AI are aware of this disparity and are trying to build ever more humanlike central nervous systems, such as computers that “learn” through trial and error. But these machines can only duplicate certain brainlike functions.
I wonder why so much attention (and consequent funding) is spent trying to mimic the human brain instead of, for example, researching practical medical advances. We make hundreds of thousands of brains every day; they are called babies.
BARRY MALETZKY via e-mail
BUBBLE TROUBLE
In “The Quantum Multiverse,” Yasunori Nomura discusses the classic idea of a multiverse in which cosmic inflation led to an infinite number of “bubble universes” and an alternative theory in which such universes do not coexist in real space but rather are potential outcomes of observations, or “probability space.”
Nomura notes that we might be able to observe “a remnant from a ‘collision’ of bubble universes in the sky.” Is it therefore implicit that our “bubble” could collide at any time with another one? And if so, would our bubble (and our existence) simply “burst” without any advance warning?
E. DENNIS KELL Mays Landing, N.J.
Nomura states that superdistant galaxies are moving away from Earth faster than the speed of light and therefore cannot be observed, a limit called the cosmological horizon. Yet Albert Einstein put a speed limit on everything in the universe: the speed of light. Thus, the theory of relativity would be violated if anything receded from Earth faster than that speed.
BRUCE BARNBAUM Granite Falls, Wash.
NOMURA REPLIES: Regarding Kell's question: Because of the eternally inflating nature of the space in which our bubble resides, the probability of our universe colliding with other universes is almost certain. It is very unlikely, however, that our bubble would “burst”—the effect would be diluted by the many things that have occurred within our universe. In fact, the dilution is expected to be so strong that the possibility of finding even faint evidence of a bubble collision is (unfortunately) low.
In response to Barnbaum: There is no contradiction here. If we define the velocity as the change of the physical distance divided by time, then distant objects do recede from us faster than the speed of light, but this is only because space is expanding. The objects are not actually propagating faster than light.
COSMOS CONSENSUS?
Nomura describes the concept of a multiverse as arising from the theory of inflation. But in the February issue [“Pop Goes the Universe”], Anna Ijjas, Paul J. Steinhardt (one of the originators of the theory) and Abraham Loeb described themselves as now questioning the inflation idea.
I realize that there are many opinions and competing theories at the fringe of our knowledge about the cosmos and our existence, but you could at least reference this conflict and give readers some context for this latest venture into the unknown. If inflation is passé among those at the forefront of the quest for understanding the nature of our ultimate environment, then does it not follow that theories built on that idea are equally suspect?
J. A. SCLATER Aldergrove, British Columbia
THE EDITORS REPLY: Ijjas, Steinhardt and Loeb did raise objections to the theory of inflation, but as the authors pointed out in their February article, their view is a minority opinion. In our July issue, we printed a letter responding to that article that was co-signed by 33 scientists who support inflation, including Nomura. Although the ultimate verdict on inflation is still out, we believe both articles represent important and legitimate scientific viewpoints, so we invited the authors to present their ideas to readers directly.
ALTERNATIVE ARCHAEOLOGY
In “Romance of the Vanished Past” [Skeptic], Michael Shermer argues against my book Magicians of the Gods, which describes the possibility of a forgotten episode of civilization in prehistory. Shermer's article is a shallow and tendentious treatment of a complex subject that does not take proper account of rebuttals to critical attacks on the Younger Dryas impact hypothesis, in which a comet strike more than 12,000 years ago caused the megafaunal extinction in North America, and misrepresents the state of the argument around my theory that this event wiped out an advanced human society as well.
Please inform your readers that in May, Shermer held a live online debate with me on the Joe Rogan Experience, Episode 961, that covered these subjects in much greater depth and afforded me the possibility of presenting a proper rebuttal. It is available at http://bit.ly/2rr6ivF
GRAHAM HANCOCK via e-mail
SHERMER REPLIES: If I had to distill the hours of my debate with Hancock into one central point, it would be the importance of philosopher Karl Popper's idea of falsifiability in science: that scientific theories make predictions that observations can prove to be incorrect. What would it take to falsify Hancock's theory?
Further, outsider scientists in general, and alternative archaeologists such as Hancock in particular, can and do make important contributions but only if their paradigm-challenging ideas not only explain why an accepted theory is wrong but also why the evidence better fits their theory. In my opinion, Hancock's idea is based entirely on negative evidence—what he thinks is wrong with the accepted archaeological timeline—and he offers no positive evidence of this purported lost civilization: no metal, no writing, no tools and not even pottery.
CLARIFICATION
“Lost at Sea,” by Danielle L. Dixson, refers to GABAA as a neurotransmitter. Technically, it is a neurotransmitter receptor.