machine learning

by

Turing the world

AI will bring us many new understandings. And confusions, as Joseph Weizenbaum warned.

As we think our electronic world is becoming more human, often it’s becoming less.

In the past few years, Siri and Alexa and their kin have shot past the Turing test, proposed by British mathematician Alan Turing in 1950. We can’t always tell if there’s a human or machine on the other side of conversation. The raw power of the underlying artificial intelligence keeps accelerating, especially for the type of AI known as deep learning, built on connections between layers of neural networks. Deep learning systems already can beat humans at making predictions from, say, medical images. And they can make findings that humans wouldn’t attempt—for instance, tapping ECG data to predict patient sex and age.

Although some of these models try valiantly to explain their decisions, more often than not it’s a mistake to think we understand what’s going on under the covers. “A full explanation might require looking at thousands or tens of thousands of variables and complex probabilistic relationships that connects things where we don’t see any connections,” says David Weinberger, author of Everyday Chaos. “You have to look at all of that, and in many instances we just can’t.”

It’s also a mistake to believe deep learning and other AI technologies actually understand our world. They don’t see a kitten or a tumor or your favorite Calvin and Hobbes collection. All they see are patterns of swirls in their oceans of data.

When chatting with Siri and Alexa and our other semi-loyal cloud servants, though, we tend to anthropomorphize these beasts. Seeing the world as human-like has been a common human trait for longer than we can track. We imagined supernatural beings based on the worst human patriarchs; now we teach our children that dolphins are happy to be enslaved so that they can entertain us. Back in the 1980s as he introduced a crude personal robot, Nolan Bushnell remarked that the robot’s bugs were what gave it personality. We’re still there, looking for personality as we try to tease Siri.

So it’s good to think carefully about the right roles for the strange computing power lurking so many places. In medicine, better ways to figure the around-the-clock insulin dosing for people with type 1 diabetes would be great. Ditto a tool to predict if someone in the ICU will go into cardiac arrest shortly. But forget any chatbot “therapist” that claims to understands us.

Back in 1966 Joseph Weizenbaum wrote the first chatbot, Eliza, with one variant called Doctor modeled on simple psychotherapy. Weizenbaum was horrified when his secretary didn’t want him to see her conversation with the Doctor and then when other computer scientists suggested building clinical versions.

“What I had not realized is that extremely short exposures to a relatively simple computer program could induce powerful delusional thinking in quite normal people,” he wrote in Computer Power and Human Reason, published in 1976. “Computers and men are not species of the same genus… However much intelligence computers may attain, now or in the future, theirs must always be an intelligence alien to genuine human problems and concerns.”

Joseph Weizenbaum 77

by

Working out

Okay, what really happens down the road to all our jobs?welcomeWe know that automation replaces many human jobs and generates many others, and that artificial intelligence will accelerate this creative destruction. Historically, the default view among business and technology leaders, supported mostly by hand-waving, is that this unstoppable march will bring a wealth of new jobs, if only the masses somehow can receive proper technological education.

It’s hard to assess the recent historical record on job loss versus gain, although today’s New York Times offers an interesting take. And while we can easily spot job losses, new jobs created by machines, “almost by definition, are harder to imagine,” as MIT economist Erik Brynjolfsson pointed out in a session at the American Association for the Advancement of Science (AAAS) annual meeting in Boston on Saturday.

But in the past couple of years the public discussion has grown more worried, with one dark perspective on implications well described in a poorly titled essay by Rutgers historian James Livingston.

At the AAAS session, Harvard computer scientist David Parkes presented some relevant thoughts from the 100 Year Study on Artificial Intelligence project. Here are a few quotes from the study’s report on AI and real life in 2030, published last September:

  • “AI will gradually invade almost all employment sectors, requiring a shift away from human labor that computers are able to take over.”
  • “To date, digital technologies have been affecting workers more in the skilled middle, such as travel agents, rather than the very lowest-skilled or highest skilled work. On the other hand, the spectrum of tasks that digital systems can do is evolving as AI systems improve, which is likely to gradually increase the scope of what is considered routine. AI is also creeping into high end of the spectrum, including professional services not historically performed by machines.”
  • “A spectrum of effects will emerge, ranging from small amounts of replacement or augmentation to complete replacement. For example, although most of a lawyer’s job is not yet automated, AI applied to legal information extraction and topic modeling has automated parts of first-year lawyers’ jobs. In the not too distant future, a diverse array of job-holders, from radiologists to truck drivers to gardeners, may be affected.”
  • “As labor becomes a less important factor in production as compared to owning intellectual capital, a majority of citizens may find the value of their labor insufficient to pay for a socially acceptable standard of living. These changes will require a political, rather than a purely economic, response concerning what kind of social safety nets should be in place to protect people from large, structural shifts in the economy. Absent mitigating policies, the beneficiaries of these shifts may be a small group at the upper stratum of the society.”
  • “Longer term, the current social safety net may need to evolve into better social services for everyone, such as healthcare and education, or a guaranteed basic income. Indeed, countries such as Switzerland and Finland have actively considered such measures. AI may be thought of as a radically different mechanism of wealth creation in which everyone should be entitled to a portion of the world’s AI-produced treasure.”

At another packed AAAS session, Alta Charo, professor of law and bioethics at the University of Wisconsin at Madison, gave a masterful quick summary of the history and findings of the report on human genome editing from the National Academy of Science. Released last week, this report’s recommendations drew plenty of public attention—far more than last fall’s AI in 2030 report, although AI will have much greater impact in the next decade or two or three.

by

The write stuffing

Backlit_keyboard

When I graduated from high school, all I really knew professionally was that I wanted to write on many topics. Last weekend when people at my high school reunion asked politely what I wrote about, I did find myself saying, many topics—in fact, way more now than when I worked as a staff journalist. Okay, I’m not covering the full human condition. Much of the universe is unexplored. But so far this year I’ve done stories about medical hackathons and crowdsourced scientific challenges, global data security and global financial crises (still separate topics so far!), drug development crises, the future of suburbia, steam power, gene therapy,  agricultural particulates, the challenges of small data in healthcare, chemical sensing on a chip, employee cross-training,  urban carbon dioxide release, jet engines, zebrafish brains, surgery by telemedicine and robotics manufacturing, among others.