用户名:  密码:   
网站首页即时通讯活动公告最新消息科技前沿学人动向两岸三地人在海外历届活动关于我们联系我们申请加入
栏目导航 — 美国华裔教授专家网科技动向学人动向
关键字  范围   
 
Rodney Brooks:A.I. Will Not Conquer Humanity
Rodney Brooks:A.I. Will Not Conquer Humanity
1/8/2019 10:40:20 AM | 浏览:71 | 评论:0

Rodney Brooks:A.I. Will Not Conquer Humanity

If you imagine the technologies that will define daily life in 50 years, it’s tempting to think of Arthur C. Clarke’s dictum that advanced technology is indistinguishable from magic. You might envision the world of 2069 bursting with things we would consider fantastical today.

The problem with that is, technologies don’t just magically appear. They come from the clever refinement and recombination of previously existing technologies. Even when powerful innovations do arise, it can take decades for them to replace old stuff that works well enough.

To wrap my mind around where we’re actually heading, I called Rodney Brooks. Brooks, 64, is both a technological optimist and a realist. He’s the former head of the Computer Science and Artificial Intelligence lab at MIT and a co-founder of two robotics companies — iRobot, maker of the Roomba floor cleaners, and Rethink Robotics, which until recently made robots that could work closely with people. He also has written extensively about why A.I. is overhyped, and how people misunderstand the uneven pace of technology.

This conversation has been edited for clarity.

Medium:Where would you begin to think about what daily life will be like in 50 years?

Rodney Brooks: Self-driving cars are going to be a big thing 50 years from now. From the late 19th century to the mid-20th century, cities got transformed all over the world by automobiles. We are going to have another transformation of our cities, I think, from autonomous driving and personal transportation, although it may clog things up so much that we’ll get much better public transportation. I don’t know yet.

The city is going to be rather different, and there’s going to be a lot more cities. We’re going to have twice as many people in cities, across the world, as we do now. And then:the impact of climate change, which is undeniable. I was in a hotel just south of D.C., in August, where there were trees, and little village walks, and it was all under a roof, because the climate is so shitty there most of the year. They built a little town that’s part of a hotel. We may see more of that.

What else?

In North America, in Europe, Japan, and China, we are going to be a much older population. In 50 years, Africa’s going to be half the world’s population and is going to be very, very young. So maybe all the innovation will be coming from Africa 50 years from now.

Someone who fell asleep 50 years ago and woke up today probably would be blown away by social changes, medical improvements, and our communications devices. But I don’t think the world would look fundamentally different to that person. If we were to fall asleep now and wake up in 2069, would we feel more lost?

Some things will still be around that look pretty much like they are today. Fifty years ago, we had airplanes with wings, and jet engines hanging off of them. And that’s what our airplanes still look like. They’re much better engines, they’re much more fuel-efficient, they’re bigger planes in general. Way more people fly on them, so there’s a lot more of them. But that hasn’t fundamentally changed in any surprising way.

Now, you point to the iPhone, etc. In the last 50 years, we’ve seen a unique technological change, and that is the sustainment of Moore’s Law. It’s a unique event in history that computation got so much better, on a regular basis, for 50 years. Most people, if you’re thinking about the future, apply Moore’s Law to every technology, which is totally wrong. Moore’s Law was unique in why it could happen, and the whole misunderstanding of that has distorted our view of how things will change in the future.

How was Moore’s Law deceptive?

Moore’s Law was based on the fact that you could change the physical structure [of computer circuits] without changing information content. You could halve the physical structure. Is there a pile of sand on my desk, or not? You take away half the grains, there’s still a pile of sand on my desk. You take away half of the remaining, there’s still a pile of sand on my desk. And we were able to do that, over the 50 years, 25 times or so. Until we got down to a single grain, which is where we are now. And we can’t halve the pile of sand anymore.

“The exponential economy is oversold. There is no process in the universe, ever, that’s been exponential for a sustained period of time.”

We haven’t seen the end of the effect of Moore’s Law, as we figure out how to apply these vast amounts of computation to more and more problems. That’s where it will still have an impact. But there hasn’t been a sustained halving or doubling in any other technology.

So, the idea that we are entering a period of exponential progress — if that were true, there’d be an unbelievable amount of technological change in the world every year.

Yeah, I think the exponential economy is oversold. There is no process in the universe, ever, that’s ever been exponential for a sustained period of time, because you run out of stuff locally to do with it, or to use.

Nonetheless, could it be true that technological change will accelerate now, because there’s still so much room for applications of information technology to diffuse throughout society?

We’re in an exploitation phase. In Massachusetts the year before last, we got rid of toll booths. We didn’t get rid of toll takers by having robots take tolls from people. We used a whole array of digitalized supply chains. You no longer need to have your credit card physically present for transactions. Everyone has access to the web, so you can go and register your transponder to your license plate. And then there is a little bit of A.I. in reading the characters of a license plate, if there’s no transponder, and that relies on the databases of the department of motor vehicles being available to connect the license plate to the address to send the bill, if someone’s not registered. So all these different pieces came together, and they’re all a result of information technology. We’re going to see a lot of exploitation, where new pieces get put together by different players, and it will change the way services work.

But we haven’t had a big breakthrough in artificial intelligence for nine years now. People have said to me, “But of course we’re going to a breakthrough every year.” No, what we’re seeing is exploitation of machine learning at the moment — at a grand scale. I don’t think we can automatically assume we’re going to get breakthroughs on any schedule at all. And that’s another reason I say Moore’s Law was unique. It had scheduled breakthroughs. We haven’t seen technologies with a schedule at any other time.

Is fiction also tripping people up when it comes to understanding the path A.I. is on?

Well, I repeatedly hear people saying, “A.I. is going to take over the world and dominate us,” and they refer to science fiction stories. We don’t read ghost stories and say, “The ghosts are coming! They’re going to take over!” We don’t see “first contact” stories(and say), “The aliens are coming! They’re going to take over!” Because people fundamentally know that they shouldn’t believe in ghosts, or aliens showing up any day. But they get so confused about A.I. that they take science fiction as fact. It’s a little hard for me to understand. But there are many things in the world of humans that are hard for me to understand.

So, is it fair to say A.I. doesn’t even really exist at the moment?

Oh, not in the way that people think of it. Right now, A.I. is a classification system with deep learning; there is no intent, there is no understanding of the world. Defense Advanced Research Projects Agency(DARPA)just announced a $2 billion program whose aim, over many years, is to try to give A.I. the common sense understanding of the world that an 18-month-old has. The idea that intelligent, thinking, conniving beings exist in A.I. is total, total, total science fiction. We don’t have anything with the intent of a slug at this point.

Might that still be true in 2069?

Absolutely, it could be.

“The idea that intelligent, thinking, conniving beings exist in A.I. is total, total, total science fiction. We don’t have anything with the intent of a slug at this point.”

Do you think making computers truly intelligent will require them to have a physical presence in daily life; for example, robots moving among us could accumulate more data about the world than what you could simulate in a server or in a data center.

That’s certainly what I have been arguing since the 1980s. It’s what Alan Turing suggested, in 1948, in a paper of his that didn’t see the light of day till 1970. I think, ultimately, getting a robot a simple grounding in reality will require a physical presence. But I may be wrong.

Could it possibly require some blending of biological material and electronics, computers with living cells, or something like that?

It may. My next blog post is going to talk about that, actually. My dining room table is covered in books and old papers as I’m trying to figure this one out.

Back to self-driving cars, and whether they’ll be everywhere in 2069. So far, the rollout has gone slower than some companies have promised.

Well, we’ve already passed some of the deadlines when they said we’re going to have autonomous cars on the road. And if you look carefully over the last six months, what you see the big players saying is, “Our first rollout is going to be in a geographically isolated or bound area, under good weather conditions.” And some of them are saying all the human-driven vehicles in that space will be shuttles driven by employees. So, it’s not having to interact with the general public driving. And many of the rollouts are in places where the vehicle can afford to stop dead in its tracks at any point and not cause an accident. You can’t do that on the freeway — you cannot stop dead.

So, on campus-like environments, they changed it from a dynamic problem to a static problem, at least initially. And I think that’s smart. I’m not saying that we won’t get there in the future, but it’s going to take a lot longer, and a lot more experience. It may require transformation of our cities, and our roads, to get it everywhere.

How likely is that? If the existing infrastructure works well enough, it doesn’t always get replaced when a better technology is available.

No, it doesn’t. We’ve had self-driving trains for 40 or 50 years and we haven’t put many on. We’ve got 15 self-driving train systems in the U.S., mostly in airports.

One of the companies you co-founded, Rethink Robotics, shut down this fall because demand was too soft for its collaborative robots that could work alongside people in factories and packaging facilities. How does the fate of Rethink shape your thinking on the uneven pace of technological development? This particular company didn’t work out, but the broader idea behind it feels like a winner over the long term.

The way I’m starting to think about it now is in terms of the Segway.

The Segway was a radical idea:an electrical personal transportation device. But it wasn’t what took off. Now, we’re starting to see personal transportation devices take off — these electric scooters that are everywhere in some cities. Segway got some things right, but they didn’t get other things right. For instance, under the Segway model, each individual owned the device, and had a pretty high capital cost, $5,000 to get one. These scooters are much lower capital, much lower mass, and you don’t own one, you just rent it by the minute, or 10 minutes or whatever. So the idea of a personal electric transportation device was pioneered by Segway, but didn’t have everything together to make it widely adoptable.

Will the homes of 2069 have robots doing chores?

I think we will have more robots cleaning, and more than just the floors. Toilet bowls will be cleaned by robotic devices, but they might be permanently attached to each toilet, rather than wandering around. I’m pretty sure our rooftop solar panels and our windows will be cleaned by robots. And our indoor personal “farms,” growing fresh food on site, will be tended by robotic devices.

Whether people think of the devices that do it as robots is another matter. Remember in the 1970s when computers were in magnetic tape cabinets? If we said that we’d have computers in our kitchens in 20 or 30 years, people, thinking about the magnetic tape cabinets, would have said “no way.” But, of course, we do now have lots of computers in our kitchens — I have them in my oven, my microwave, my dishwasher, my fridge, my Alexa, and quite possibly some in the lighting system that I haven’t identified — but they don’t look anything like those 1970s mainframes. Likewise for the robotic devices populating our houses in 50 years. They won’t look like what we imagine robots look like today.

 

Rodney Brooks:自动驾驶可能还得等50年

作者:Brian Bergstein 
在谈到未来50年AI如何发展时,“现代机器人之父”Rodney Brooks称摩尔定律并未终结,但未来摩尔定律不会适用于每种技术,这些技术不太可能继续呈现指数级增长。同时,自动驾驶最终可能会实现,但要经过比50年更长的时间,积累更多的经验。

科幻电影中的技术都会实现吗?在21世纪末期,我们能看到无人驾驶汽车在高速公路上奔跑吗?

针对未来50年内的人工智能发展,Medium专栏作者采访了顶级机器人专家Rodney Brooks。

现年64岁的Rodney Brooks是美国著名机器人制造专家,斯坦福大学博士,麻省理工学院计算机科学和人工智能实验室(CSAIL)的前负责人,同时还是美国人工智能研究协会(AAAI)的创始会员。20 世纪 90 年代设计了第一个火星机器人,他同时也是著名机器人公司 iRobot 和 Rethinks Robotics的创办者,也被业界称为“现代机器人之父”。

摩尔定律并未终结,但未来技术不太可能呈现指数级增长

如果想象一下50年后那些定义日常生活的技术,就很容易想到Arthur C. Clarke的格言——“先进的技术与魔法无法区分”。到2069年,世界会充斥着我们现在看起来觉得不可思议的事物。

关键在于,技术不会是神奇地突然出现。它们通常是现有技术的巧妙改进和重新组合。即使确实出现了新的强大创新,它们也需要几十年的时间才能取代现有的“旧”事物。

问:如果让你想象一下50年后的日常,你最先想到什么?

答:自动驾驶汽车将在50年后成为一件大事。从19世纪末到20世纪中期,世界各地的城市由于汽车的出现发生了变化。我认为,我们将从自动驾驶和个人交通中对我们所生活的城市进行另一次改造,尽管它可能还会有交通堵塞问题,所以我们还需为了更好的公共交通而努力。


如今的城市将会变得相当不同,并且会有更多的城市出现。正如我们现在,未来全世界的城市将拥有两倍的人口。其次是气候变化的影响,这不可否认。今年八月我住在华盛顿特区以南的一家酒店,那里有树木,还有小村庄可以散步,但你可能很难想象,这些都是在屋顶之下,因为一年中的大部分时间气候都很糟糕。他们建了一个小镇作为酒店的一部分。这种奇怪现象以后还会层出不穷。

在北美,欧洲,日本和中国,我们将面临一个老龄化更加严重的社会。在未来的50年内,非洲将占据世界人口的一半,而且都非常非常年轻。因此,也许在50年后,所有的创新都将在非洲上诞生。

问:如果一个人在50年前睡着了,他今天醒来后,可能会被社会变革、医疗改善和我们的通讯设备震撼住。但我不认为现在的世界看起来与50年前并无联系和共同点。如果我们现在睡着并在2069年醒来,那是的我们会感到更失落吗?

答:(到2069年)有些事情仍然会像今天一样。五十年前,我们就有了喷气式飞机,而如今我们的飞机看起来仍然如此,只不过它们有着更好的发动机,它们更省油、体型更大,而它们也成为了更多人选择的出行方式。但和50年前,这并没有什么根本性的改变。

但不是什么事物都是如此,比如iPhone等。在过去的50年里,我们看到了独特的技术变革,这就是摩尔定律的持续性。


Rodney Brooks:A.I. Will Not Conquer Humanity

摩尔定律

这也是人类历史上一个独特的事件:计算得到了极大发展。果你正在考虑未来时,将摩尔定律应用于每种技术,这是完全错误的。对于如何解释某些科技为何发展如此迅速的原因,摩尔定律是非常独特的,但如果将其应用到整个科技界、整个社会的话,会扭曲我们对未来事物将如何变化的看法。


问:摩尔定律是如何“迷惑”大众的?


答:摩尔定律基于这样一个事实,摩尔定律是指IC上可容纳的晶体管数目,约每隔18个月便会增加一倍,性能也将提升一倍。如果这个趋势继续的话,计算能力相对于时间周期将呈指数式的上升。


我们没有看到摩尔定律效应的终结,因为我们可以弄清楚如何将这些大量的计算应用于越来越多的现实问题,这仍会产生影响。但是,任何其他技术都不可能持续减半或翻倍。


因此,我们正在进入一个指数级的增长时期——如果这是真的,那么每年世界上都会出现令人难以置信的技术变革。


我认为指数型经济是被夸大了,宇宙中没有任何过程能在一段持续的时间内一直呈现指数级增长,因为你可能已经将资源用光了。


问:尽管如此,技术变革现在还在加速,是因为信息技术的应用仍然有很大的空间传播到整个社会吗?


答:我们正处于开发阶段。前年在美国马萨诸塞州,我们摆脱了收费站。我们并非是通过机器人收费来摆脱这项工作,而是使用了一整套数字化供应链。你不再需要实际操作信用卡进行交易。每个人都可以访问网络,因此您可以将转发器注册到你的车牌。如果没有转发器,那么在扫描车牌时会应用上AI,并且依赖于机动车部门的数据库可以将车牌连接到地址来发送账单。所有这些都是信息技术的成果,我们也会看到更多的开发。


但是,九年来我们在人工智能方面没有取得重大突破。


人们对我说,“当然,我们每年都会有一个突破。”不,我们所看到的是目前对机器学习的大规模开发。我认为我们不能自动假设我们将在任何领域上都取得突破。


AI不会统治世界,甚至对这个世界还没有完善的认知


问:当了解到AI的发展路径时,小说是否也会让人们沮丧?


答:好吧,我反复听到有人说,“AI将接管世界,支配我们”,他们指的是科幻故事。但我们不会在读鬼故事时说:“幽灵来了!他们将接管世界!”,我们也不会看着《异形》说:”外星人来了!他们将占领世界!“因为人们从根本上就知道他们不应该相信鬼魂或者外星人的出现。但他们对AI非常困惑,因为AI是现实存在的,所以人们容易把科幻当作事实。这对我来说有点难以理解,但人类世界中有很多东西让我难以理解。


问:那么,说人工智能目前还不存在是否恰当?


答:哦,不是人们想得那样。目前,AI并不拥有什么自己的意识以及对世界完善的理解。美国国防高级研究计划局(DARPA)不久前宣布了一项耗资20亿美元的计划,该计划多年来一直致力于让AI对世界有一个常识性的认知。有着像人类那样的智慧、独立思考能力甚至生命的AI完完全全则是科幻小说里杜撰的。在这一点上,不用怀疑。


问:2069年仍旧如此?


答:当然。


问:你认为让计算机真正智能化会要求它们在日常生活中有实体存在吗? 


答:这是自上世纪80年代以来我一直在讨论的。这也是阿兰·图灵在1948年的一篇论文就提到的。我认为,让机器人最终在现实中成为一个常见的存在是需要有实体存在的。当然我的观点不见得对。


问:是否可能需要将生物材料和电子设备结合在一起?例如,带有活细胞的计算机?


答:可能需要。实际上,我的下一篇博文将讨论这个问题。


实现无人驾驶仍有很长的路要走


问:回到自动驾驶汽车,以及它们是否会在2069年无处不在。到目前为止,自动驾驶汽车的推出速度比一些公司承诺的都要慢。


答:的确已经过了他们承诺的截止日期了。如果你仔细观察过去六个月,那么大家的统一口径都是:“在天气条件好的情况下,我们第一次推出的自动驾驶汽车测试是在地理位置偏远的地区。”


根据这些公司的说辞,这些汽车安排了员工来当驾驶测试员。因此,它不需要与一般公众驾驶互动、磨合。许多部署都可以实现让自动驾驶车辆在允许测试的领域内随时停下来,并且不会导致事故发生。这当然不可能在高速公路上这样做,那样的话,谁都无法阻止死亡。


因此,在这样的环境下,这些公司将动态问题变为静态问题,至少在最初阶段是如此,我觉得这么做很聪明。

Rodney Brooks:A.I. Will Not Conquer Humanity


 

谷歌母公司Alphabet旗下自动驾驶公司waymo


我并不是说我们将来不会实现完全的无人驾驶,但它会花费比想象中更长的时间,而且需要积累更多的经验。这可能还需要改造我们的城市和道路,以便实现自动驾驶汽车在哪里都可以上路。


问:(实现处处都有全自动驾驶)有多大可能?如果现有的基础设施运行良好,当有更好的技术可用时,旧事物并不总是被取代。


答:可能性不是很大。像自驾列车已经出现很多年了,而且我们还没有在此投入很多。在美国,我们有15个自驾列车系统,而它们主要集中在机场。


问:在2069年,机器人是否为我们做家务?


答:我想将有更多的机器人为我们清洁,而不仅仅只是打扫地板,它们也会为我们刷碗、打扫卫生间,它们可能会永久地固定在厨房和卫生间里,而不是像扫地机器人那样四处移动。我很确定未来我们的屋顶太阳能电池板和窗户都会被机器人清理干净。而我们还可以拥有室内个人“农场”,在房间内种植新鲜食物,由机器人照料。


人们是否认为这些家用设备可以被视作一种机器人的设备则是另一回事。上世纪70年代,计算机都庞大无比,如果对那时的人说20年或30年后我们的厨房里都有电脑,那么估计那时候的人也会表示:“不可能”。但是,现实是,如今我们的厨房可以放很多台电脑。同样,50年后,我们屋内的机器人设备也是如此——它们看起来不像我们今天想象中机器人的样子


参考链接:

https://medium.com/s/2069/a-top-roboticist-says-a-i-will-not-conquer-humanity-133f2611d035

相关栏目:『学人动向
日本摄影师RK眼里的中国 2019-01-22 [103]
陈黎芳:华为对美国技术只是从十万八千里追至两万五千里! 2019-01-22 [105]
何帆:未来30年大转型, 敢问路在何方?如何应对挑战? 2019-01-20 [100]
赵天宇 :缺钱还烧钱, 生物药企如何活下去? 2019-01-20 [43]
李碧武:“互联网+教育”究竟是什么? 2019-01-19 [51]
中国文人百年学术抄袭史(1891-2018) 2019-01-16 [270]
耶鲁大学教授威德雷谢维奇《反思精英教育》:赢在起跑线,输在终点 2019-01-16 [163]
张秋红:40%的中国城市人口流失 2019-01-15 [152]
沈晓明:海南的三大痛点 2019-01-15 [210]
刘亚东:我的国, 有那么厉害吗? 2019-01-15 [170]
相关栏目更多文章
最新图文:
周舟:苏黎世理工、洛桑理工和剑桥联合研发出可在体内游走的柔性微型机器人 :台灣政客競當網紅 走火入魔 :陈黎芳:华为对美国技术只是从十万八千里追至两万五千里! :NYO青少年交响乐团2019欧洲巡演报名正式启动 (3/11截止) :中北大学信息商务学院面向海内外引进各类优秀人才 :CES《 2019年技术趋势 》:用核心数据预测全球电子消费产品技术趋势 宋楷文:美前駐華大使鮑卡斯: “美國優先”是錯誤的 :何帆:未来30年大转型, 敢问路在何方?如何应对挑战?
更多最新图文
更多《即时通讯》>>
 
打印本文章
 
您的名字:
电子邮件:
留言内容:
注意: 留言内容不要超过4000字,否则会被截断。
未 审 核:  是
  
关于我们联系我们申请加入后台管理设为主页加入收藏
美国华裔教授专家网版权所有,谢绝拷贝。如欲选登或发表,请与美国华裔教授专家网联系。
Copyright © 2019 ScholarsUpdate.com. All Rights Reserved.