Artificial Intelligence

Perhaps the week’s biggest and scariest robot news, though—certainly for journalists—was the robot reporters story.

Thomson Financial has been using automatic computer programs to generate news stories for almost six months. The machines can spit out wire-ready copy based on financial reports a mere 0.3 seconds after receiving the data. Thomson management likes its reporter robots so much that it has decided to expand the fleet.

Flesh-and-blood journalists were quick to decry the move. “Those editors who can’t wait to install computers at the expense of journalists should beware,” warned Mark Tran in the Guardian article “Robots write the news.”

“Look at what happened in Space Odyssey, when HAL took over the spaceship. Or worse still, think of Terminator 3, when the Skynet network of computers unleashes nuclear war.”

Tran was joking. Well, half joking. But his joke was also a poignant plea. A robot may be able to turn a share report into three pithy paragraphs in less than a second, but it can’t go and watch movies about other robots and turn that into a warning for the world.

Because it can’t live, it can’t think. Or so we think. Tran’s conclusion isn’t very reassuring. “We endangered financial journalists could prolong our lives in the short term by slapping more adjectives into our copy,” he suggests, “but the writing does seem to be on the wall, as far as earnings reports go.” If all that stands between a writer’s job and redundancy is a few adjectives, well, that’s plain scary.

”Scary”—yes, nice adjective. It’s got human emotion, empathy, experience. Good, we’re still on the right side of the Turing Test the side the robots can’t get to.

Or can they? I can hear the laments already, with 20/20 hindsight. First they came for the bomb disposal crews, and we said nothing. Then they were spot-welding and spray-painting on the auto plant assembly lines, and still we said nothing. Only now that they’ve come for the journalism jobs do the journalists scream. But it’s too late.

Mistrust and paranoia have set in. How do we know Mark Tran isn’t already a robot? “Tran”—does that even sound like a human name?

It’s a losing battle. These days, it seems, there are fewer and fewer jobs a robot couldn’t do. Even automatic translation, which some said only humans could do properly (because meaning requires context and context requires lived experience) is coming on by leaps and bounds, pulling jobs out from under the feet of the lower-level human translators.

Ours are (by nature) unusually plastic brains whose biologically proper functioning has always involved the recruitment and exploitation of non-biological props and scaffolds. More so than any other creature on the planet, we humans emerge as natural-born cyborgs, factory tweaked and primed so as to be ready to grow into extended cognitive and computational architectures: ones whose systemic boundaries far exceed those of skin and skull. (p5—emphasis added)

The conjecture, then, is that one large jump or discontinuity in human cognitive evolution involves the distinctive way human brains repeatedly create and exploit various species of cognitive technology so as to expand and reshape the space of human reason. We, more than any other creature on the planet, deploy non-biological elements (instruments, media, notations) to complement (but not, typically, to replicate) our basic biological modes of processing, creating extended cognitive systems whose computational and problem-solving profiles are quire different from those of the naked brain. Human brains maintain an intricate cognitive dance with an ecologically novel, and immensely empowering, environment: the world of symbols, media, formalisms, texts, speech, instruments and culture. (p4—emphasis added)

The cognitive anthropologist Ed Hutchins^WP, in his book Cognition In The Wild^AM depicts the general role of cognitive technologies in similar terms [i.e. as thought prosthetics], suggesting that “[Such tools] permit the [users] to do the tasks that need to be done while doing the kinds of things people are good at: recognizing patterns, modeling simple dynamics of the world, and manipulating objects in the environment.” This description nicely captures what is best about good examples of cognitive technology: recent word-processing packages, web browsers, mouse and icon systems, etc. It also suggests, of course, what is wrong with many of our first attempts at creating such tools: the skills needed to use those environments (early VCR’s, word-processors, etc.) were precisely those that biological brains find hardest to support, such as the recall and execution of long, essentially arbitrary, sequences of operations. (p4—emphasis added)

These [Alzheimer] patients were a puzzle because although they still lived alone, successfully, in the city, they really should not have been able to do so. On standard psychological tests they performed rather dismally. They should have been unable to cope with the demands of daily life. What was going on?

A sequence of visits to their home environments provided the answer. These home environments, it transpired, were wonderfully calibrated to support and scaffold these biological brains. The homes were stuffed full of cognitive props, tools, and aids. Examples included message centers where they stored notes about what to do and when; photos of family and friends complete with indications of names and relationships; labels and pictures on doors; “memory books” to record new events, meetings, and plans; and “open-storage” strategies in which crucial items (pots, pans, checkbooks) are always kept in plain view, not locked away in drawers.

Before you allow this image of intensive scaffolding to simply confirm your opinion of these patients as hopelessly cognitively compromised, try to imagine a world in which normal human brains are somewhat Alzheimic. Imagine that in this world we had gradually evolved a society in which the kinds of scaffolding found in the St. Louis home environments were the norm. And then reflect that, in a certain sense, this is exactly what we have done. Our own pens, paper, notebooks, diaries, and alarm clocks complement our brute biological profiles in much the same kind of way. Yet we never say of the artist, or poet, or scientist, ”Oh, poor soul—she is not really responsible for that painting/theory/poem; for don’t you see how she had to rely on pen, paper, and sketches to offset the inadequacies of her own brain?”

“As long as you have some history and some quantifiable data from past experiences,� Mr. Snijders claims, a simple formula will soon outperform a professional’s decision-making skills.

Something researchers have known for decades: that mathematical models generally make more accurate predictions than humans do. Studies have shown that models can better predict, for example, the success or failure of a business start-up, the likelihood of recidivism and parole violation, and future performance in graduate school.

They also trump humans at making various medical diagnoses, picking the winning dogs at the racetrack and competing in online auctions. Computer-based decision-making has also grown increasingly popular in credit scoring, the insurance industry and some corners of Wall Street.

The algorithms behind so-called quant funds, he said, act with ” much greater depth of data than the human mind can. They can encapsulate experience that managers may not have.”

Other cherished decision aids, like meeting in person and poring over dossiers, are of equally dubious value when it comes to making more accurate choices, some studies have found, with face-to-face interviews actually degrading the quality of an eventual decision.

“People’s overconfidence in their ability to read someone in a half-an-hour interview is quite astounding,� said Michael A. Bishop, an associate professor of philosophy at Northern Illinois University who studies the social implications of these models.

Max H. Bazerman, a professor at Harvard Business School, wonders how many managerial decisions can actually be modeled. “The vast majority of decisions that we make in professional life don’t have this quality,� he said.

He agrees that models can make better decisions about credit card applications and college admissions, he said, “but there are many decisions that are much more unique, where that database doesn’t exist. I’m as skeptical about human intuition as these folks, but it’s not only a computer model that we replace it with. Sometimes it’s thinking more clearly.�

Many in the field of computer-assisted decision-making still refer to the debacle of Long Term Capital Management, a highflying hedge fund that counted several Nobel laureates among its founders. Its algorithms initially mastered the obscure worlds of arbitrage and derivatives with remarkable skill, until the devaluation of the Russian ruble in 1998 sent the fund into a tailspin.

Mark E. Nissen, a professor at the Naval Postgraduate School in Monterey, Calif., who has been studying computer-vs.-human procurement, sees a fundamental shift under way, with humans becoming increasingly peripheral in making routine decisions, concentrating instead on designing ever-better models.

By making smart use of computer models’ advantages, ” you’ll become like the crafty A student who doesn’t work that hard but gets mostly right answers, rather than the really hard-working student who gets lots of wrong answers and as a result gets C’s.”

A French company, Poseidon Technologies, sells underwater vision systems for swimming pools that function as lifeguard assistants, issuing alerts when people are drowning, and the system has saved lives in Europe.

This digital age belongs to the graphic interface, and it is time for us to recognize the imaginative work that went into that creation, and prepare ourselves for the imaginative breakthroughs to come. Information-space is the great symbolic accomplishment of our era. We will spend the next few decades coming to terms with it.

One basic approach to neural nets can be described as follows. Each point of a given input (for speech, each point represents two dimensions, one being frequency and the other time; for images, each point would be a pixel in a two-dimensional image) is randomly connected to the inputs of the first layer of simulated neurons. Every connection has an associated synaptic strength, which represents its importance and which is set at a random value. Each neuron adds up the signals coming into it. If the combined signal exceeds a particular threshold, the neuron fires and send a signal to its output connection; if the combined input signal does not exceed the threshold, the neuron does not fire, and its output is zero. The output of each neuron is randomly connected to the inputs of the neurons in the next layer. There are multiple layers (generally three or more), and the layers may be organized in a variety of configurations. For example, one layer may feed back to an earlier layer. At the top layer, the output of one or more neurons, also randomly selected, provides the answer.

Because computation underlies the foundations of everything we care about, from the economy to human intellect and creativity, we might well wonder: are there ultimate limits to the capacity of matter and energy to perform computation? If so, what are these limits, and how long will it take to reach them?

Our human intelligence is based on computational processes that we are learning to understand. We will ultimately multiply our intellectual powers by applying and extending the methods of human intelligence using the vastly greater capacity of nonbiological computation. So to consider the ultimate limits of computation is really to ask: what is the destiny of our civilization?