Reporter: Jieshu Wang | Editor: Jiaxin Su , Rita Chen. Provided by Synced
- To read Part I of this interview, please click this link :Interview with John Markoff : A prophet in the age of AI (1)
- To read Chinese edition, please click this link:专访「硅谷老炮」John Markoff：人与机器谁主未来？
As the public started to realize that AI’s super intellect might surpass human beings’ intelligence, a question like “Will AI’s super intelligence take the place of human being’s survival skills?” emerges to the surface. Markoff had the same question before. He shared his experience in an electric shaver factory in the Netherlands. 128 machine arms were doing the same work men and women used to do with a more yoga-kind of agility. The operations that the camera guided them were way better than the most skillful worker. Machine arms swung through two assembly lines, drawing three eclipse curves, and then plugged the product parts into the tiny holes that human eyes could not see. These machine arms ran crazily fast, so they would be put into the glass cabinets in order to prevent their managers from getting hurt. Machine arms kept working for three time a day, 365 days a year without rest. These machine arms were seen as the ideal assembly “workers,” and they did not need payment as sophisticated workers did. It seemed as if AI was taking over.
Nonetheless, after a long time of reporting and investigating, Markoff realized that there were jobs that could not be replaced by automation. In 2010 and 2011, Markoff wrote about e-discovery, a software that had stronger ability in reading legal documents than humans. Many people’s first reactions included, “this software will affect the legal labor market.” But, it did not. The actual job of a lawyer could be identified to 12 things, including counseling, going to trial, and convincing juries, document reading among others. From this perspective, automation machines only automate one tiny part of the task, not the entire task. This was why Markoff also states that even though there were machines that could write stories and paint, skilled jobs like journalists and lawyers would not be replaced by AI technologies in short term. In this way, even though Artificial intelligence seemed to be taking over many aspects of the work, not all jobs will be replaced by machines.
AI vs. IA
In addition to the conflict between machine automation and human’s skills, Markoff discussed the relationship between AI and Intelligence Augmentation(IA). AI researchers focused on developing technologies that mimic human’s capabilities, while IA researchers paid more attention to Human Computer Interaction, utilizing computing to extend human’s intelligence ability. Popular robots can be counted as a subset of AI, but most of the technologies the public use were all from IA research. Even though AI and IA seemed incompatible, Markoff mentioned that there was not a clear cutting line between these two fields due to the deeper functionality intersections between them. For example, self-driving car is a good IA model, using AI to enhance human’s capabilities. However, Markoff added, “if you can basically make completely self-driving cars, you need fewer cars, you need less parking space, you can redesign cities. At the same time, you lose millions of millions of jobs.” To him, this is one of the interesting dichotomies that rise when the technologies of AI overlap with those of IA. “It’s a puzzle basically,” Markoff added.
Reporting about AI
At the end of the interview, Markoff expressed his opinions about reporting about AI. He thought that even though there were considerable amount of news organizations covering technologies, the quality of their reporting was uneven — “A lot of them focus on sort of business and consumer stuff and it’s not focusing on technical detail.” Markoff emphases that people and our society need journalists to observe and explain these things. Regardless of reporting AI or IA, reporting itself is to provide readers a new perspective to view our technologies and prompt readers to think about the impacts our technologies might bring. The best thing for journalists could do is to “give us a real view of exactly the technology is.”
It is Synced’s pleasure to have John Markoff, a journalist at the New York Times and winner of the 2013 Pulitzer Prize for Explanatory Reporting, share his professional experience about Artificial Intelligence(AI).
In your book, you mentioned two distinct technical communities: artificial intelligence and intelligence augmentation. We are familiar with AI, but not so familiar with the concept of IA, because when we talk about IA, many people may think of some Frankenstein style or cyborg, something genetically modified or being planted with electrodes in the brains. So, could you please describe what AI and IA are exactly?What is the relationship between them? What do you expect to see about them in the future?
Markoff: I guess, in terms of communities, it has to do what they focus on as researchers and designers. So the AI community has for a long time basically focused on building technologies that mimic human capabilities, everything from the physical robots, I consider robotics to be a subset of Artificial Intelligence and that was originally true, to all the intellectual things like cognition and reasoning, all the things humans do. IA focuses on using computing to extend humans’ intellectual powers. It started with Engelbart’s group, and Engelbart as a research scientist, basically decided to devote his life to building systems that would allow small group of human intellectual workers to collaborate effectively and basically “bootstrap” human knowledge. That was his passion. Like Internet. If you go back, so the internet is a set of protocol. Basically it’s a set of documents, and they are called RFCs. If you go and read RFC Number 1, the reason for the ARPAnet which was the of the internet was to use Doug Engelbart’s technology remotely. It was called oN-Line System, NLS. So in a sense, that was the first killer app, the very first killer app.
I mean, that was the internet was for initially was to build a set of tools that would allow human knowledge we could use to work more collaboratively, which was a really nice idea. I’ve watched these tools of all over a long time, so his ideas were first borrowed by the researchers at Xerox Palo Alto Research center. You know, Alan Kay worked there to develop personal computer. And then the ideas were coalesced to run the computer called Alto and the software run on Alto. And then both Microsoft and Apple borrowed these ideas and commercialize them. And that was how the computing world was reshaped.
So you think IA actually shaped most part of the technology we use today.
Markoff: Examples of sort of IA approach, taking technology to extend the human mind. But you know, they are not like black and white ideas. First of all, there are these two communities that generally have philosophical orientation. The AI guys have their view and HCI (Human Computer Interaction) community has been a community that really wants to use computing tools to facilitate humans to work with computers. So I’m not the only one taking this position. I mean, there are some other very interesting computer scientists who have written about this dichotomy. But it’s a paradox, because if you augment the power of a human you also need fewer humans because humans are more powerful. So that was the puzzle I try to work on and try to understand in the book. And I have models basically. I think Siri is a great IA model. It is using AI technologies to extend the human power to basically build a partner for human. And I completely understand that it’s not a black and white issue. The self-driving car issue is a really fascinating one because if you can basically make completely self-driving cars, you need fewer cars, you need less parking space, you can redesign cities. At the same time, you lose millions of millions of jobs. So there’s this really interesting dichotomy and attention basically on the ideas. It’s a puzzle basically; I don’t think there…. I think it’s a good way of viewing the problem and thinking about the impact of technology. That’s sort of what I’m trying to frame.
Do you think AI and IA will have more intersection in the future ?
Markoff: I hope that the designers of the systems were actually thinking about these questions. I was looking for a way to synthesize IA and AI and bring them back together. So yes, in a sense I hope that’s true but it has to be done by the individual actions of designers. I know some of them very well. The two designers of Siri really are model from that kind of thinking. I mean there are deeply schooled AI researchers who want to use the techniques to do Engelbart style things, extend human power. That really excites me. So the potential is there.
In “Machines of Loving Grace” you acknowledged the likelihood of job losses due to prevalence of automation technology, but you did not further explore the subject. Other books such as Jerry Kaplan’s “Humans Need not Apply” argues that robotics will increase inequality and therefore have profound impact on society. You don’t seem to share that thinking and you seem to think that concerns of inequality can be addressed by social measures such as new tax codes. Could you elaborate on your position on this and give us the reasons for your thinking?
Markoff: Jerry is a good friend of mine. We have this debate all the time. We have very different perspectives. Both of us have shifted our views. Books come out and time goes by. I finished the book at the end of 2013, so it’s like almost two year. And the same is true for him. And now we think we both come to feel that the threat of automation is catastrophic. It’s not gonna destroy all jobs anytime soon, but there is still an active debate. I’ve been reading a lot about economists’ writings. There’s a growing interests in the field of economic inequality in America, becoming a very hot topic. Jerry makes this argument about technology and inequality and I think that the issue is controversial. The debate is fascinating to read, but I do not think it’s clear. I think it’s possible he is right but I’m still reading on it. The best labor economist I read on this is David Autor. He has written some really thoughtful pieces about the limits of AI and in terms of automating jobs. He was the one who popularized the notion of “hollowing out” the American economy. You know there’s a great concern about the loss of middle class in America. Some thought that maybe in part technology is responsible for that. For example, there were many well-paid but repetitive white collar jobs and a lot of them have gone away in the last half decade to a decade. The people who did things in corporations that were repetitive have largely been replaced by software that just plugs things together. That’s automation but I wouldn’t call it AI. You can use based software and effectively automating interchange converse and things like that. And it takes away jobs. It’s computer automation but it’s not AI automation. I don’t think it’s super clear. I mean, Jerry is very adamant he is right. I just haven’t seen the evidence. So we are still discussing that. There’s a big debate among the economists about what’s driven this growing inequality in American society. It might be changing tax policy and it can be other things expect science and technology.
What kind of job will not be replaced by robots and AI?
Markoff: Part of the answer of the question over what time period. It depends. There is a very interesting report just done some MIT labor economists (Eric Brynjolfsson and Andrew McAfee) looking at the jobs of the lawyer. Actually they started off by citing my reporting in 2011 and 2010 when I reported about the use of e-discovery software which is document reading software. You have software that can do it demonstrably better job in reading documents than humans. A lot of people looked at that and said: “Oh, it’s gonna affect the growth of the legal market.” They looked at the actual job of a lawyer and it turns out that reading documents is only one part of probably I think they identified 12 things that lawyers do including counseling, going to trial, convincing juries. The only thing that the AI technologies have impact on is reading documents. So that’s called task automation as opposed to job automation. We are seeing in American economy, that in the white-collar world, lots of examples of task automation for skilled jobs, doctors, lawyers, and people like that, programmers, journalists. For example, so much was made in neuroscience, that the fact that machines can write stories. But when will machines be able do interviews. Maybe someday, but not anytime soon. So once again, it’s automating a part of the task, which is the writing task, not the complete task. So I was wondering one of the things that got me change my perspective on what fraction of the American work force, which is more spread out at risk been automated.
There’s an interesting recent McKinsey report. They have a very different approach. The debate is shifting in America now. I can see there’s a growing consensus that the economy is not gonna be decimated by automation any time soon. The American economy recovered and there are more people working in America now than ever in history. There is something called labor participation rate, that is the percentage of the possible workers that actually working. In America, the rate is actually very low. It’s not about technology; mostly it’s about the aging American work force. As my generation begins to retire, they leave the work force and that lower the participation rate. So it’s a complicated question.