This might remind us of our traditional confidence in the universality of laws governing the physical universe and its many subdomains, and remind us as well of the difficulty of applying general laws to a contingent, interoperating, generative universe of spinning dust, some of which happens to have motives and loved ones. We may be learning that the particulars that generalities whisk off their shoulders like dandruff turn out to count for everything.
The most important recent movement in philosophical ethics accords with this as well. Virtue ethics notes the problems with deontological and utilitarian approaches and instead asks what Aristotle took to be the fundamental question of ethics: What does it mean to lead a good life? The answer does not lay out the principles to be followed or the calculus to be computed. Modern virtue ethics instead says that the good life is one in which we flourish. Flourishing by its nature is open ended. How you flourish depends on the particularities of who you are and which virtues—“excellences” in the ancient Greek sense—you cultivate. Flourishing is not an end state but a response to the unpredictable opportunities and obstacles that happen to face us. The fact that modern virtue ethics was initiated by a woman—Elizabeth Anscombe in a famous 1958 paper28—and that much of the most important work on it (especially on the ethics of care) has been done by women is not an accident.29
These movements away from principles and from a cold calculus of pleasure and pain brings morality more in line with the transformations we have seen under way in realm after realm: a turn from reducing complex phenomena to instances of general rules and laws, and toward acknowledging the particulars that make each case unique.
Machine learning systems are profoundly nonmoral. They are just machines, not Just machines. But the need to operationalize our morality for them is leading us ever further down the path away from the principle-based morality that governs the Land of Ought. This can lead, perhaps simultaneously, to two contradictory outcomes. If we outsource morality to AI unchecked, the vulnerable can be tyrannized by faceless statistical engines that literally do not hear their voices. We could also lazily cede control to AI in cases where fairness and flourishing would be better served by insisting that the decisions be left up to us. At the same time, our machines’ ability to process individual cases according to models that account for more detail and particularity than the human brain can follow may shift our own model, encouraging us to attend more closely to the particular and personal details that make moral situations as unique and real as each inhabitant of the Land of Is.
Meaning
[T]he network failed to completely distill the essence of a dumbbell.
That’s the conclusion reached by the Google computer scientists who fed images of dumbbells into a deep-learning system and then asked it to draw an image of what it thinks a dumbbell is.30 As we noted in chapter 2, the system succeeded in putting together images showing dumbbells at different orientations, but many of the images had a weight lifter’s detached arm eerily gripping the dumbbell. The results were greeted as an amusing failure.
But was it a failure? That depends on what “meaning” means.…
* * *
We have long tried to understand meanings with something akin to the sine qua non approach to explanation. Aristotle helped set us down this road by telling us that what a thing is—its essence—is the category it’s in plus what distinguishes it from other items in that category. For example, we human beings are in the category of animals, but we’re distinguished from other animals by our ability to reason. We are the rational animals.
For a couple of millennia, we found comfort in this idea of meaning: not only was there an order, but the principle of order was simple and consistent. For example, in the eighteenth century, Carl Linnaeus placed each animal, vegetable, or mineral in a category and placed each category in a hierarchy that looked like the Org Chart of Everything. Scientific genus-species names still reflect Linnaeus’s Aristotelian-style classification.31
But in the late nineteenth century, a different idea began to emerge. The Swiss linguist Ferdinand de Saussure proposed that every word exists in a web of words that are related to it but that are different from it. The meaning of sneaker is its similarity to, and difference from, shoe, boot, high heel, and so forth. Each of those words is at the center of its own web of similarities and differences.
The notion of meaning as a messy relational context—a web or network—has become quite pervasive, in part because it’s become possible to put this context to work. For example, Facebook’s social graph and Google’s Knowledge Graph connect atoms of information without regard for what single categories they should be filed under. A graph can connect information about, say, the Apple Watch to snippets about other smartwatches, digital watches, analog watches, descriptions of how digital watches are manufactured, histories of timepieces, philosophies of the clockwork universe, the physics of time, maps of the sources of the raw materials required, the sources that use forced labor, literary references to digital watches, photos of people wearing watches, the science of digital displays, the way high school boys used to try to get their calculators’ LCD displays to display 80085 because it looks like it spells “BOOBS” … anything related in any way. Each of those nodes is itself connected to many more pieces, like words in Saussure’s webs of meaning, like hyperlinked pages on the web, and like data in the models machine learning systems build for themselves.
Our old technology was not nearly as generous with meaning. The beatniks of the 1950s were on to something when they insisted, “I am not a number,” in between bongo solos. The first generations of computers indeed reduced things—employees, inventory, processes—to the handful of fields that the technology could manage. Now we worry not that computers have reduced us to what fits onto punch cards but that they know far too much about us and how we’re connected. We at times, understandably, yearn for the good old reductive days.
Being overly inclusive in the data we collect and connect raises obvious issues about the loss of privacy, but we are at the same time gaining galaxies of meaning. Aristotle and Linnaeus tried to describe what a thing essentially is by referencing exactly two relationships: how it is like the other things in its category and how it is distinguished from them. At its heart, this approach assumes that each thing is essentially distinguishable from all the rest of creation that is not that thing. Our new view expresses meaning in the overwhelming and unsystematic connections of things to everything else in every way imaginable, including some that only our machine learning systems see. In a connected world, the boundaries between things are drawn not by those things’ essential essence but by our intentions.
So did Google’s AI fail at the task of identifying dumbbells? Yes, if we take things to be what they are only when they’re apart from everything else. But if you were an alien, which photo would give you a better idea of what a dumbbell is, a dumbbell in isolation or Google AI’s image? Is a dumbbell a dumbbell apart from its complex web of relations to human bodies, exercise equipment, health, mortality, and vanity?
There are, of course, times when we want the pared-down meaning—for example, when you’re trying to check out a dumbbell’s handgrip in a product catalog. But you are checking out handgrips presumably because you already know that a dumbbell is a weight intended to be grasped and lifted in order to become strong, to become attractive, or to finally win the approval of your mother, the competitive weight lifter. The pared-down meaning only makes sense within the thing’s place in the messy, generative set of implicit and explicit connections of everything to everything else. Precision comes at the cost of meaning. Messiness is the root of all.
In this way, the internet’s collaborative, cacophonous chaos of links and machine learning’s model of models unrestrained by complexity are far more representative of what things are than Aristotle’s or Linnaeus’s attempt to clarify meaning with the edge of a scalpel.
The Future
If the globe that ornaments your desk h
as ridges where there are mountains, that globe is bumpier than the Earth it represents. In fact, if your globe is the size of a billiard ball, to be accurate it should be smoother than a billiard ball.32
Our calculators assume we need only so many digits of magnitude or precision, and make us trade off between the two.
We’ve been able to program traditional computers only because we’ve been willing to specify a relative handful of stepwise rules, load in readouts from the dials we’ve planted across the planet, and handcraft the exceptions we can anticipate.
From these sorts of peeks through the slats at our overwhelming world, we have confronted the future by ascertaining its possibilities and relentlessly reducing them as best we could.
Now we have new tools. They sometimes come to conclusions that surpass our ability to comprehend them. They express their truths in probabilities and percentages; certainty has come to flag that an error is about to be committed. They create a place of connection and creativity that thrives on particularity. They open a world in which every mote depends on every other in ways that explanations insult.
These new tools are far from infallible. In fact, unchecked they can visit unfairness with especial ferocity on the most vulnerable. But we built these tools because, overall and most of the time, they work. They have shown us that we no longer have to reduce the future to survive it. We thrive in our new future by making more of it.
This future isn’t going to settle down, resolve itself, or yield to simple rules and expectations. Feeling overwhelmed, confused, surprised, and uncertain is our new baseline stance toward the world because that expresses the human truth about the world.
We are at the beginning of a new paradox: We can control more of our future than ever, but our means of doing so reveals the world as further beyond our understanding than we’ve let ourselves believe.
We have a category for this sort of paradox: the awe that first roused humans to look up and to begin to grow into what we are.
Awe abides. It can be a gracious awe that gives thanks for a gift we did nothing to deserve. It can be awe at the preposterous improbability that billions of years would lead to this exact us standing at this precise here. It can be awe at the privilege of understanding so little, or so much, in the face of all there is to know. But awe always opens outward, letting the unthought ground our ideas and the winds wash through our words. One way or another, awe opens the more of the world.
Now, at last, our tools are complicit in our awe.
NOTES
Introduction
1. Riccardo Miotto, Li Li, Brian A. Kidd, and Joel T. Dudley, “Deep Patient: An Unsupervised Representation to Predict the Future of Patients from the Electronic Health Records,” Scientific Reports 6 (2016): article 26094, https://perma.cc/R2GY-YBQQ.
2. There are many ways of computing this, but one calculation says that there are 10800 possible moves, which works out to 10720 for every atom in the universe. “Number of Possible Go Games,” Sensei’s Library, last modified June 25, 2018, https://perma.cc/2JPY-KMVF. Some estimates put the number of chess moves at 10120. The number of atoms in the universe is generally estimated at around 1080. To get a sense of how vast these numbers are, keep in mind that 1081 is ten times larger than 1080.
3. Cade Metz, “The Sadness and Beauty of Watching Google’s AI Play Go,” Wired, Mar. 11, 2016, https://perma.cc/UPD4-KVUR.
4. Big Grammar has declared that Internet is no longer to be capitalized. That is, I believe, a mistake. Likewise for the Net and for the Web. But I have lost this battle with my culture. So, in this book the Internet and the Web—capitalized here for the last time in this text—will be treated as if they were just pieces of technology and not unique, lived-in domains.
5. Dan Siroker, “How Obama Raised $60 Million by Running a Simple Experiment,” Optimizely Blog, Nov. 29, 2010, https://perma.cc/TW5M-PHJ5. See also Richard E. Nisbett, “What Your Team Can Learn from Team Obama about A/B Testing,” Fortune, Aug. 18, 2015, http://perma.cc/922Z-5PMA.
6. Brian Christian, “The A/B Test: Inside the Technology That’s Changing the Rules of Business,” Wired, Apr. 25, 2012, http://perma.cc/H35M-ENAA.
7. For example, “Baltimore after Freddie Gray: The ‘Mind-Set Has Changed’ ” increased readership by 1,677 over “Soul-Searching in Baltimore, a Year after Freddie Gray’s Death.” Mark Bulik, “Which Headlines Attract More Readers,” Times Insider, June 13, 2016, https://www.nytimes.com/2016/06/13/insider/which-headlines-attract-most-readers.html.
8. Katja Kevic et al., “Characterizing Experimentation in Continuous Deployment: A Case Study on Bing,” ICSE-SEIP ’17: Proceedings of the 39th International Conference on Software Engineering: Software Engineering in Practice Track (2017): 123–132, https://doi.org/10.1109/ICSE-SEIP.2017.19.
9. Sean Hollister, “Here’s Why Samsung Note 7 Phones Are Catching Fire,” CNET, Oct. 10, 2016, https://perma.cc/HKM2-VQBB; Kate Samuelson, “A Brief History of Samsung’s Troubled Galaxy Note 7 Smartphone,” Time, Oct. 11, 2016, https://perma.cc/NQ7F-9ZCT.
10. See Elizabeth Landay, “From a Tree, a ‘Miracle’ Called Aspirin,” CNN, Dec. 22, 2010, http://perma.cc/HTT2-FR5C; and J. M. S. Pearce, “The Controversial Story of Aspirin,” World Neurology, Dec. 2, 2014, http://perma.cc/2TAJ-RC2T.
11. Josefina Casas, “5 Tricks for Writing Great Headlines on Twitter and Facebook as Awesome and Clickable as Buzzfeed’s,” Postcron, accessed Nov. 2, 2018, https://perma.cc/JE59-CLT8.
12. “Every Drop Adds Up,” ALS Association site, https://perma.cc/V8T7-XSAN.
13. Braden R. Allenby and Daniel Sarewitz, The Techno-Human Condition (Cambridge, MA: MIT Press, 2010).
14. Edward Lorenz, “Predictability: Does the Flap of a Butterfly’s Wing in Brazil Set Off a Tornado in Texas?,” address at the American Association for the Advancement of Science, Dec. 29, 1972, https://perma.cc/L5J3-BSF7. Also see Christian Oestreicher, “A History of Chaos Theory,” Dialogues in Clinical Neuroscience 9, no. 3 (Sep. 2007): 279–289, https://perma.cc/6U5L-QKXH. Here are two excellent explanations and explorations: James Gleick, Chaos: Making a New Science (New York: Penguin, 1987); and Steven Johnson, Emergence (New York: Simon and Schuster, 2001).
15. See Jurassic Park. No, really, you should see it. We watch it every year at Thanksgiving.
16. Rachel Carson, Silent Spring (New York: Houghton Mifflin, 1962). On the term ecosystem, see A. J. Willis, “Forum,” Functional Ecology 11 (1997): 268–271, 268.
17. Roger Abrantes, “How Wolves Change Rivers,” Ethology Institute Cambridge, Jan. 13, 2017, https://perma.cc/3364-BUSZ.
18. Nassim Nicholas Taleb, The Black Swan: The Impact of the Highly Improbable (New York: Random House, 2007).
19. Daniel Pink, Free Agent Nation (New York: Warner Business Books, 2001).
20. Obviously, what I say in this book does not necessarily represent the opinions or ideas of any of those groups.
Chapter One
1. Kasha Patel, “Since Katrina: NASA Advances Storm Models, Science,” NASA, Aug. 21, 2015, http://perma.cc/RFN4-94NZ. See also Kelsey Campbell-Dollaghan, “Here’s How Much Better NASA’s Weather Models Have Gotten since Katrina,” Gizmodo, Aug. 24, 2015, https://perma.cc/A7QU-RWTK.
2. Patel, “Since Katrina.”
3. Progress is being made in earthquake prediction. For example, a team led by researchers at Los Alamos National Laboratory has successfully used machine learning to analyze acoustic signals to predict when an earthquake will occur … in the laboratory. Bertrand Rouet-Leduc et al., “Machine Learning Predicts Laboratory Earthquakes,” Geophysical Research Letters 44, no. 18, Sept. 28, 2017, 9276–9282, https://perma.cc/566D-JAB4. There’s also been progress in using deep learning to predict when an earthquake’s aftershocks will occur. See James Vincent, “Google and Harvard Team Up to Use Deep Learning to Predict Earthquake Aftershocks,” The Verge, Aug. 20, 2018, https://perma.cc/RJX5-DUCX.
4. Statisticians distinguish between predictions and forecasts, using tomorrow’s weat
her and long-term climate change as their standard example, but for our purposes here we don’t need to. See Nate Silver, The Signal and the Noise (New York: Penguin Books, 2012), for an excellent discussion.
5. G. J. Whitrow, Time in History (New York: Barnes and Noble, 1988), 25.
6. In an article in the Atlantic, Eric Weiner says of Athens in particular, “[I]n their efforts to nourish their minds, the Athenians built the world’s first global city. Master shipbuilders and sailors, they journeyed to Egypt, Mesopotamia, and beyond, bringing back the alphabet from the Phoenicians, medicine and sculpture from the Egyptians, mathematics from the Babylonians, literature from the Sumerians.” Eric Weiner, “What Made Ancient Athens a City of Genius?,” Atlantic, Feb. 10, 2016, https://perma.cc/QE9X-TSZV.
7. Actually, maybe not so literally, since the ancient Greeks didn’t have a word for blue and there is debate about their perception of color. See Ananda Triulzi, “Ancient Greek Color Vision,” Serendip Studio, Nov. 27, 2006, https://perma.cc/XDU7-LDFJ. Also, RadioLab has an excellent podcast episode about the perception of the color of the sky: “Why Isn’t the Sky Blue?,” RadioLab, May 20, 2012, podcast, 22:23, https://perma.cc/239Y-L2C6.
8. According to Lisa Raphals, the gods couldn’t reverse the Fates’ decrees, but they could at times postpone them. See her “Fate, Fortune, Chance, and Luck in Chinese and Greek: A Comparative Semantic History,” Philosophy East and West 53, no. 4 (Oct. 2003): 537–574, https://perma.cc/6WGB-7H5T. On the broader question of the forces affecting ancient Greek life, see Martha C. Nussbaum, The Fragility of Goodness: Luck and Ethics in Greek Tragedy and Philosophy, 2nd ed. (Cambridge: Cambridge University Press, 2001).
Everyday Chaos Page 21