科学巨匠：杰弗里·辛顿

BBC之声，音乐、广播、播客。

BBC Sounds, music, radio, podcasts.

大家好。欢迎收听本周的《五维生命科学》。我是《裸体科学家》节目的克里斯·史密斯。接下来，领导旅行者探测器任务的爱德华·斯通去世，享年88岁，这些探测器至今仍在运行已半个世纪，我们将回顾他的遗产。此外，科学家们发现了他们认为的首例患有唐氏综合症的尼安德特人，显然得到了族群的照料。

Hello. Welcome to this week's five Life Science. I'm Chris Smith from the Naked Scientists. And coming up, Edward Stone, the man who led the Voyager probe missions, which are still working half a century on, has died at the age of 88, and we reflect on his legacy. Also, scientists discover what they think is the first Neanderthal with Down syndrome, clearly cared for by his community.

如果你要去参加大型音乐节，你的急救包里应该准备些什么？还有...

And what should your first aid kit have in it if you're heading to a major music festival? Plus.

我们必须学会不要像对待传统计算机程序那样对待聊天机器人——过去你可以依赖程序，但现在不行了。

We have to learn not to treat the chatbots like you would have treated an old fashioned computer program where you could rely on it. You can't.

《科学时代》节目回归，本期嘉宾是人工智能先驱杰弗里·辛顿。

Times of Science is back with the artificial intelligence pioneer, Geoffrey Hinton.

这里是五台直播的《裸体科学家》节目。

The naked scientists on five live.

本周首先，人们向爱德华·斯通致敬，这位主导了非凡的旅行者探测器任务的人去世了，享年88岁。出生于爱荷华州诺克斯维尔的斯通曾就读于加州理工学院，后加入NASA喷气推进实验室。正是在那里，他成为定义其职业生涯的旅行者太空任务的项目科学家。

First this week tributes have been paid to Edward Stone, the man who oversaw the remarkable Voyager missions. He's died and he was 88. Stone, who was originally from Knoxville and Iowa attended the California Institute of Technology before joining the NASA Jet Propulsion Laboratory. It was there that he became project scientist for the Voyager space missions that were going to define his career.

四、三、二、一。点火成功，升空开始。泰坦半人马座火箭搭载着两艘旅行者号航天器中的第一艘升空，将人类的感知范围扩展至太阳系前所未有的远方。传回的报告显示，那对大型固体火箭发动机运行完美，每台产生120万磅推力。

Four, three, two, one. We have ignition and we have liftoff. We have liftoff of the Titan Centaur carrying the first of two Voyager spacecraft to extend man's senses farther into the solar system than ever before. Reports coming back indicate those twin large solid motors are functioning perfectly producing 1,200,000 pounds of thrust each.

1977年8月20日，旅行者2号发射升空。为了更深入了解爱德华·斯通及其卓越成就，我拜访了皇家天文学家马丁·里斯，他不仅认识斯通，还对宇宙有着更全面的认知。

The launch of Voyager two on the 08/20/1977. So to find out more about Edward Stone and his remarkable achievements, I went to meet someone who knew him and knows a lot more about the cosmos in general, our Astronomer Royal, Martin Rees.

是的。我很荣幸认识他，因为学术研究我常去加州理工学院，而他是那里举足轻重的人物。当然，我特别清楚他在旅行者项目中的伟大成就，但他的领导价值更为重要——他真诚迷人，是位出色的团队领袖。更了不起的是，他执掌这些旅行者项目长达五十年左右。

Yes. I'm privileged to know him because I went to Caltech for academic reasons quite often, and he was an important presence around there. And, of course, I was aware of his great achievements with the Voyager project in particular, but the value of his leadership was very important because he was very genuine and engaging, a very good team leader. And, of course, it's remarkable that he was in charge of these Voyager projects for fifty years more or less.

七十年代可谓是NASA的黄金十年，对吧？但为什么他们选择在1977年实施旅行者计划呢？

It was NASA's kinda golden decade, wasn't it, in the seventies? But why did they do Voyager when they did in 1977?

正如埃德等人所认识到的，当时存在进行'盛大巡游'的构想。外行星的排列位置使得航天器可以借助它们的引力弹弓效应，更快抵达太阳系边缘。这种每170年左右才出现一次的星体排列是绝佳机会。如果当时不尝试向外行星发射探测器就太遗憾了，而埃德正是实现这一构想的关键人物之一。

Well, as Ed in particular realized and others did, there was at that time an interest in a possible grand tour. This is the idea that the alignment of the outer planets was such that by bouncing off them, by going close to them, it might be possible to give an extra push to a spacecraft headed to the edge of the solar system and get to the edge rather quicker. And so this alignment which happened every one hundred and seventy years or thereabouts was a good opportunity. And it'll be sad if there weren't efforts made to send some probes to the outer planets at that time, and Ed was one of the people who actually implemented this.

不过为什么要发射两艘旅行者探测器呢？

Why two Voyager probes though?

它们遵循略有不同的轨道，其中一艘后来超过了另一艘。我记得一艘重点探测土星和天王星，另一艘则计划飞得更远。

They went on slightly different orbits, and one of them overtook the other. And I think one was concentrating more on Saturn and Uranus, and one hoped to get further.

他们教了我们什么？

What did they teach us?

当然，我们获得了一些外行星的照片，而且确实还有其他探测器后来提供了外行星和冥王星的图像。但旅行者号真正特别之处在于，它经过近五十年仍在运行。它早已越过冥王星，仍在传回数据。对于这样一个遥远的探测器而言，最关键的是它已脱离太阳系的影响范围。我们通常将太阳系视为行星、小行星带、柯伊伯带等构成的系统。

Well, of course, we got some pictures of those outer planets, and of course, there have been some other probes that have given us later pictures of outer planets and of Pluto. But of course, what's been really special about Voyager is that it is still going after nearly fifty years. It's way beyond Pluto, and it's sending back data. And the particular thing, which is important for a very remote probe, is that it gets out of the influence of solar system. The solar system, obviously, we regard as the planets and the asteroids and the Kuiper Belt and all that.

但存在一个太阳风推挤星际介质的区域，只有到达这个区域的边缘才算真正进入星际空间。人们普遍认为需要飞到比冥王星更远的位置，才能观测到太阳风逐渐消失、真正接触原始星际物质的过渡带。显然旅行者号穿越了这个边界，当时探测到了密度、温度和磁场的变化。这确实是标志性成就——人类制造的物体首次突破了太阳系的影响范围。

But there is a region where the solar wind pushes out the interstellar medium, and so we're not really in interstellar space until we get to the edge of of this region. And it was generally thought that you had to get further away than Pluto before you would encounter the transition between seeing the wind of the sun still going out and genuine pristine into stellar matter. And it's clear that the voyagers did get across this barrier and found changes in the density, the temperature, and the magnetic field at that time. So that was the distinctive achievement to actually be the first human human based objects that got outside the influence of the solar system.

今早我查看了NASA官网，可以追踪探测器的飞行轨迹。数据显示它们距离地球约240亿公里，大约是地球到冥王星最近距离的四倍。这真是...相当遥远。

I had a look at NASA's website this morning because you can chart the progress of the voyages, and it says they're 24,000,000,000 kilometers away. So that's roughly four times the Earth to Pluto distance at closest. It's it's a it's a long way.

没错。信号传到那里需要超过一天时间。最近几个月新闻里报道的惊人事件是：旅行者号的信号突然中断了，出现了某些故障。

No. That's right. So it takes more as a day to get a signal to them. And what's been remarkable, and that's been in the news in the last few months, is that the signal from the Voyager went dead. Something went wrong.

但令人惊叹的是，他们似乎成功恢复了通信。想想这其中的难度：从地球的射电望远镜向如此遥远的目标发送信号，而那个仅靠几瓦电力运行的设备竟能从宇宙深处回应。这实在不可思议。我不禁想到：用1970年代的技术尚能取得如此成就，若能用现代手机级别的材料发射探测器舰队，我们的能力将会有多大提升——更精密、更紧凑。

But amazingly, they seem to have managed to revive it. And if you just think of what's involved, sending a signal to something that far away from a radio transmitter, some dish here, and getting back a signal from that enormous distance from something where the battery gives just a few watts of power. It is amazing, really. And of course, the other inference I draw is if we think of the worthwildness of what can be done even with the technology of the nineteen seventies, think how much better we could do if we were able to launch a whole flotilla of probes with the kind of material that we have in our mobile phones. Much more sophisticated, much more compact.

单是这项工程就令我肃然起敬。四十六年后——准确说是四十六年，在辐射、温度等最严酷的极端环境下，这些设备大体仍在运作。他们不得不关闭部分功能。旅行者1号甚至还有台磁带录音机，用磁记录数据后以2千比特的速率回传。这太惊人了，它至今仍在工作。

I'm in awe just of the engineering. Fifty years later, well, forty six years to be precise, in the worst imaginable environment really, in terms of radiation, temperature, and so on, this stuff is still working mostly. They've had to shut down some of the There's even a tape recorder on Voyager one, magnetic tape that records things and then sends it back at the princely rate of two kilobits, I think. I think and and it get sends it back, but that's amazing. It still works.

是的。限制在于重量上限，这意味着由于重量限制，你无法让这些仪器达到理想的坚固程度。而巧合的是，当新型SpaceX火箭投入使用后——它能将100吨载荷送入太空——这个问题将大大缓解。目前，我们正期待能开展更宏大的项目，发射更重的设备。但这曾是个非常微小的物体，必须设计简单，使用的还是五十年前就过时的技术。

Yes. The constraint is the weight limit, which means you can't be as robust as you'd like to be with these instruments because of the weight limit. And incidentally, that's going to be a problem that's gonna be much eased when we have the new SpaceX rocket which can launch up to a 100 tons into space. So at the moment, we're looking forward to much bigger projects where we can launch heavier things. But this was a very small object, which had to be simple and use a technology that's fifty years out of date.

埃德·斯通让我惊叹的是，当许多科学家早已挂起试管架退休时，他仍在坚持工作。88岁高龄（或大约这个年纪）时，他仍在领导一项太空任务。这相当了不起。确实如此。

What I found amazing about Ed Stone was that he was still going at a time when many scientists have hung their test tube rack up. He was still going, still leading a mission at the age of 88 Yes. Or thereabouts. That's pretty good. Yes.

确实如此。当然，很少有人能像他这样幸运——年轻时开创的事业到耄耋之年仍在继续，仍硕果累累，仍被视为充满价值。他是个了不起的人，始终保持良好健康状况，是位卓越的领导者。整个职业生涯中，他担任过帕萨迪纳大型实验室主任，参与过十余个不同项目。

Well, that's right. And, of course, very few of us have the privilege of having work we started when we were young, still going and still being fruitful and still being deemed interesting when we're as old as he was. But he was a great guy. He continued in fairly good health, and he was a very good leader. And over his career, he was the director of the big lab at at at Pasadena, and he was involved in more than a dozen different projects over the his time.

因此他对太空探索事业做出了巨大贡献。

And so he made a great contribution to a space exploration in general.

多么伟大的遗产啊。而旅行者号仍在以近4万英里的时速飞向星际空间，并将持续飞行数百万年。这是皇家天文学家马丁·里斯对爱德华·斯通生平的追忆——这位领导旅行者号项目的科学家已于88岁逝世。新研究发现，一块尼安德特儿童头骨化石碎片呈现的结构变化，与唐氏综合征特征高度吻合。据信这名儿童存活超过六岁，这表明旧石器时代社群存在集体照护行为。

What a great legacy to leave. And, of course, the voyages continue on into outer space traveling at nearly 40,000 miles an hour and will do so for millions of years to come. That was the astronomer Royal Martin Rees on the life of Edward Stone, who led the Voyager project and has died at the age of 88. A new study has found that fossilized fragments from a Neanderthal child's skull bear the structural changes that are compellingly consistent with that individual having Down's syndrome. It's believed the child lived beyond the age of six, which suggests communal caregiving within the Palaeothic community.

此前在解剖学意义上的现代人类祖先中发现过类似案例，但在尼安德特人中绝对是首次。剑桥大学考古学家、尼安德特人专家艾玛·波默罗伊为我们解读了这项研究。我们沿剑河散步，共同探讨了这个发现。

This has been seen before in anatomically modern human ancestors, but definitely not in Neanderthals. Emma Pomeroy, who's an archaeologist and a Neanderthal specialist at the University of Cambridge, has been taking a look at the study for us. We went for a walk together along the River Cam to reflect on it.

这篇论文研究的是一块在西班牙科瓦内格拉遗址发现的头骨碎片。最初的问题之一是：这块碎片属于哪个物种？它混杂在多种骨骼中，研究者不确定是尼安德特人还是现代人类。此外，骨骼本身存在某些变异——这是靠近耳朵的碎片，包含半规管等与听觉平衡相关的耳道结构。

The paper is about a fragment from the skull that they found at a site called Cova Negra in Spain. And one of the questions to start out with was what species did this skull fragment come from? It was in a mixed collection of bones, they weren't sure whether it might be Neanderthal or modern human. And then also, there were some changes in the bone itself. So it's a fragment from near the ear, and it contains a sort of semicircular canals, the various bits of the ear canals involved in hearing and balance.

他们还发现这些骨骼与典型解剖结构存在许多差异。通过大量测量和主要对耳道的评估，他们确认这确实是一个尼安德特人个体的骨骼碎片，并发现其生长方式存在广泛异常，这些异常很可能与唐氏综合症相关。

And they also found that there were some, a lot of differences there from the typical anatomy. So what they did, they were able to establish that it was a fragment indeed from a Neanderthal individual. And by taking lots of measurements and assessing the ear canals essentially, they were able to show that there was a wide range of anomalies in the way it had grown and linked those to most likely Down syndrome.

那他们现阶段还没提取到DNA吗？

They haven't got DNA at this stage then?

是的，还没有。我认为通过古DNA诊断唐氏综合症可能相当复杂，因为这涉及多条染色体或染色体片段的异常，而高度碎片化的DNA会使这项工作变得棘手。

No, haven't. And I think a diagnosis such as Down syndrome might be actually quite complicated from the ancient DNA because it's to do with multiple numbers of chromosomes or or extra bits of chromosomes. And that's kind of tricky to to do with very fragmented DNA.

那为什么说这是突破性/里程碑式的发现呢？

So why is this a breakthrough slash a landmark discovery?

这非常有趣，因为据我所知，在考古学和古人类学记录中，我们尚未发现任何确诊的唐氏综合症案例。虽然我们知道其他类人猿可能患有此症（比如黑猩猩就有病例），但这可能是我们首次在人类祖先中发现。更值得注意的是，研究者推测这个个体死亡时仍是儿童（可能超过六岁但年纪尚小），这具有多重意义——因为骨骼的这些变化意味着这个孩子可能患有严重症状，包括平衡障碍、眩晕等问题。

Well, it's very interesting, because I don't think we've actually got any identified cases of Down syndrome, in the archaeological the paleoanthropological record. We know that other great apes can have Down syndrome. We've got examples of chimpanzees, but this is, perhaps the first example that we have from human ancestors. It's also really interesting because they've suggested that it's from an individual who was still a child, perhaps older than six year old, but still pretty young. And that has various implications, because the nature of these changes in the bone meant that the child probably would have had quite severe symptoms in terms of real trouble with their balance, problems with vertigo, things like that.

实际上，这个孩子很可能难以完成日常活动。这意味着他需要母亲的大量照料，但作者们认为，所需的照顾可能超出了母亲单独能提供的范围。因此可以推断，当时必定有更广泛群体的照顾或对母亲的协助，才能使这个孩子在缺乏现代医疗支持的情况下存活到相对较大的年龄（以唐氏患者的标准而言）。

So really, they wouldn't have been able to do normal everyday activities very well. And the implication of that is that they would have needed substantial care from the mother, but probably, the authors argue, that the care they would have needed would have gone beyond what the mother alone could have offered. So in that sense there must have been care from a wider group or assistance for the mother enabling this child to survive actually to a fairly good age for someone with this syndrome at a time when there wasn't really any treatment and medical support.

这让你感到惊讶吗？根据我们对这些尼安德特人群体结构、社会结构的了解，以及与他们在解剖学上重叠的现代智人的研究，我们已经从中获得很多认知。他们彼此照应的行为会让你觉得意外吗？

Does that surprise you? Based on what we know about the likely community structure, the social structure of these individuals, and also anatomically modern humans who overlapped with them, we've learned quite a lot from them, Does that surprise you that they should be looking out for one another?

就我个人而言，没有。但这个问题在人类进化及我们何时开始互相关怀方面确实存在很大争议。纵观现代社会，包括现存的狩猎采集社会，我们确实看到大量关怀行为，不仅针对幼童——显然儿童需要照顾，他们很长时间内无法自立——还包括病弱或年长的个体。我们知道尼安德特人等人类祖先也生活在社会群体中。有证据表明他们会相互协作，比如在狩猎活动中。

Me personally, no. But this has been a really controversial question in terms of human evolution and when we started to care for one another. So if we look across modern human societies, including living hunter gatherer societies, we do see a great deal of care, not only for the young, which obviously children need care, they can't fend for themselves for a long time in our species, but also for individuals who are unwell or elderly. We know that other human ancestors like Neanderthals were living in social groups. We've got evidence that they were collaborating with each other, so for hunting, for example.

我们确实知道尼安德特人像现代人类一样需要很长时间成长。因此他们在童年和婴儿期需要持续数年的精心照料。根据对年长尼安德特人及其他更早期人类祖先的研究，我们也发现了需要长期照护的个体证据，包括严重受伤、重大健康问题、感染、骨折，甚至可能是截肢和瘫痪的情况。鉴于这些发现，我认为这个结论或许并不完全出人意料。

And we do know that Neanderthals took a long time to grow up just like modern humans do. So they would have needed, as children and as infants, quite substantial care for a number of years. Based on sort of elderly Neanderthals as well, and in fact other earlier human ancestors, we do find evidence of individuals who also must have required substantial care. So with major injuries, major health problems, infections, bone fractures, even perhaps amputations and paralysis. So given we have those findings, I think perhaps it's probably not entirely surprising.

作者们将这个具体发现置于这样的背景中：试图理解这种社会关怀为何及何时演化出现。是否与预期互惠有关？即如果我现在受伤你帮助我，未来你受伤时我也可以帮助你，从而实现长期互利。还是更多与对群体成员真正同理心的演化相关？

The context in which the authors are putting this particular finding is in trying to understand why this kind of social care and when this social care might have evolved. Is it to do with expected reciprocity? So the idea that, you know, if I get an injury now and you help me out, in the future, you might get an injury and I can help you out. And so it benefits us both long term. Or is it more to do with sort of evolution of true compassion for fellow members of our group?

他们提出，由于这个特定孩子永远无法真正回报，或许这能帮助我们理解这种关怀行为在人类进化过程中是如何形成的。

And they're arguing because this particular child would never really have been able to reciprocate, that perhaps helps us to understand how this caring behaviour might have come about in our evolution.

引人入胜的报道。以上是艾玛·波默罗伊的见解，她为我们评论的这项研究刚发表在《科学》期刊上。这里是克里斯·史密斯主持的《五维生命科学》。接下来，人工智能教父杰弗里·辛顿将作为本周科学巨匠登场，他将带我了解AI的实际运作原理。

Fascinating story. Emma Pomeroy there, and the study she was commenting on for us has just come out in the journal Science. This is five Life Science with me, Chris Smith. On the way, the godfather of AI, Geoffrey Hinton, is our titan of science this week. He takes me through how AI actually works.

但首先关注正在萨默塞特郡皮尔顿附近举行的全球著名音乐节——格拉斯顿伯里。我们的同事詹姆斯·蒂特科正在现场，因此我们认为让他咨询全科医生兼作家亚当·斯塔滕应该携带哪些物品前往沃西农场会是个好主意。

But first, Glastonbury, which is one of the world's most famous music festivals, is underway near Pilton in Somerset. Our colleague, James Titko, is currently there, so we thought it would be a good idea for him to ask GP and author Adam Staten what he should be taking with him to Worthy Farm.

确实。我想我们国家通常不太适应炎热晴朗的天气，人们往往准备不足。脱水是个大问题，特别是当你整天在阳光下活动时。应对方法其实很简单，就是确保持续补充足够水分。

Yeah. I suppose we don't often cope with hot weather and sunny weather much in this country, so people are often not very well prepared for it. Dehydration is a big factor, especially when you're active all day and out in the sunshine and out in the heat. Know, You I suppose the way to deal with it is pretty straightforward. It's just to make sure you're keeping a good intake of fluid up.

通常我们会建议在这样的天气里喝大约三升水。但如果你整天都在户外且活动量特别大，喝四到五升可能更合适。另外需要特别注意的，当然是这类音乐节上免不了的酒精摄入——显然酒精会加剧脱水。所以最好交替饮用酒精饮料和非酒精饮料，保持体内水分充足。

We'd normally say drink about three liters of fluid on a day like that. But if you're gonna be outside all day and gonna be particularly active, even four or five liters is probably a good idea. The other thing to bear in mind, obviously, that goes in hand with these sort of music festivals is alcohol intake, and, obviously, the alcohol can add to dehydration. So it's a good idea to try and intersperse any alcoholic drinks with nonalcoholic drinks to keep your fluid levels up.

我会尽力。不过不能保证周末期间我不会喝上几杯啤酒。所以确实需要增加水分摄入，这是肯定的。另外，我的皮肤相当苍白。

I will do my best. I can't promise that I won't be enjoying a few beers, though, over the course of the weekend. So be good to up my water intake. No doubt. Another thing, I'm have quite pasty white skin.

我想我们大多数人都有过晒伤的经历。程度从轻微到严重不等。只要注意防护并管好自己的行为，其实很容易避免。简单的措施包括戴帽子、使用高倍数防晒霜，以及尽量待在阴凉处——特别是在一天中最热的时段，大约上午11点到下午3点，那时阳光最强烈。

I think most of us have probably experienced sunburn at some time or another. Mean, I obviously, range from being fairly mild to pretty severe. It's fairly easily avoided if you pay attention and think about what you're doing. Simple things to do is wear things like a hat, obviously, use a high factor sun cream, and just try and stay in the shade, particularly in those hottest parts of the day from about eleven in the morning till three in the afternoon when the sun's really out high and strong.

对我这样的人来说，晴朗天气的三重威胁中最后一项是花粉指数。

The final prong of this triple threat of sunny weather for someone like me is pollen count.

任何患有花粉症的人，最好确保每天服用抗组胺药，理想情况下选择每日一次的无嗜睡型。提前每天服用也能在一定程度上预防昆虫叮咬可能引发的严重过敏反应。如果不幸被叮咬，它能防止症状过度恶化。当然你可能还需要

Anybody that suffered with hay fever, it's worth just making sure you're taking antihistamine every day, ideally a sort of once daily non drowsy one. Just take that each day in anticipation. It would also help defend you a little bit against the nasty reactions that you might get from an insect bite. If you're unlucky enough to get bitten, it will stop those flaring up so badly. And obviously, you might want

考虑使用鼻喷剂和眼药水，如果这些部位的症状特别严重的话。长时间在户外活动，我大部分时间都会站着，估计还会跳很多舞。过去我遇到过脚部疼痛的问题，特别是连续几天累积下来。最好的办法

to think about things like nasal sprays and eye drops if those are particularly troublesome symptoms for yourself. Long days out in the sun. I'll be on my feet for the majority of it, dancing a lot too, hopefully. A problem that I've experienced in the past is my feet getting really sore, especially when you're compounding this over a few days. Best thing

是穿舒适合脚的鞋子，避免摩擦和异常压力点。如果不幸起了水泡，药房能买到很好的水泡贴。晚上休息时，不妨脱掉鞋袜让双脚透透气，避免整晚处于潮湿闷热的状态。

is to wear some sort of comfortable, well fitted footwear that's not gonna rub and not gonna put unusual pressure points on your feet. If you are unlucky enough to get things like blisters, then there are some good blister pastas you can get over the counter from most pharmacies. Probably in the evening, if you sat down somewhere for a while, it might be worth taking shoes and socks off and listening your feet get some air so they're not sort of sweaty and damp throughout the evening.

是啊。我是说，我想不出有什么比严重的脚气更能毁掉我周末的了。另一个必然情况是震耳的音乐和大型音响系统，在某些时刻，你可能会发现自己离大喇叭很近。持续的噪音会损害听力。在这样的情况下，

Yeah. I mean, I can't think of a much easier way to ruin my weekend than a severe bout of athlete's foot. Another inevitability will be loud music, big sound systems, potentially, at points, you might find yourself quite close to big speakers. Prolonged loud noises can damage your hearing. In the context of a of

在音乐节期间，你愿意对听力造成的伤害最好是短期的。你知道，可能会耳鸣几天，听力稍微有些模糊，但这些应该会恢复。当然，如果这种情况经常发生或持续时间很长，那可能会导致长期的听力问题。如果你经常参加音乐节和演唱会之类的活动，那么某种形式的耳朵保护可能是个不错的主意。

a festival, the damage you like to do with your hearing is good to be sort of short term, really. You know, you might be left with some ringing ears and slightly muffled hearing for a few days, but that that should recover. Obviously, if it's something that happens really regularly or it's happening on a prolonged basis, then then that can lead to long term hearing problems. If you're someone that's gonna go to a lot of music festivals and a lot of gigs and things, then some kind of ear defense is is probably not a bad idea.

最后我想听听你的看法，关于音乐节的厕所，它们可是臭名昭著。有些人可能会决定，你知道吗？我就是不想经历那种体验。太折磨人了。我知道有些药物可以让你在音乐节期间忍住，等到回家再解决。

The last thing I wanted to get your opinion on, going to festivals, the toilets are infamous. And some people, they may decide, you know what? I just don't wanna go through that experience. It's too harrowing. I know there are drug treatments out there to kind of bang yourself up for the duration of a festival to wait till you get home to do your business.

在这种情况下使用这些药物合适吗？是的，你说得对。有些品牌比如易蒙停，通用名是洛哌丁胺，是一种止泻药，它会让你放慢速度，倾向于

Are those advisable in that context? Yeah. You're right. There are brands like Imodium, which is generically called loperamide, an anti diarrhea medicine, and it'll slow you down and tend

导致便秘。我是说，这其实是个人的选择，但它的缺点是如果你服用了它，我们所有的腹泻都会停止，可能会让你相当便秘，这样回家后可能会带来另一个问题，疼痛和排便困难。所以詹姆斯，我得把这个决定权交给你，稍微权衡一下。如果你觉得值得的话，我想看看你的选择。但这确实是一个选项。

to constipate you. I mean, it's a bit of a personal choice, really, but the downside to it is if you take it, we have all diarrhea, it can make you quite constipated, which then might give you the other problem when you get home with pain and and difficulty getting going. So I would have to leave that in your hands, James, to shine a bit. That's what I'm looking if you feel that's worth it. But it it, suppose, is an option.

说到这里，亚当，非常感谢你所有的建议。我现在要带着这些知识去药店了，我想在这种情况下我可能会把易蒙停留在货架上。非常感谢你和我交谈。

On that note, Adam, thank you so much for all your advice. I'm gonna go to the pharmacy now armed with the knowledge, and I think I'll probably be leaving Imodium on the shelf in that instance. Thank you so much for speaking with me.

詹姆斯·蒂科与全科医生兼作家亚当·斯塔滕的对话。剑桥大学的科学家发现，一种名为希思巨蜥的大型蜥蜴可能每年为澳大利亚的绵羊养殖户节省大量资金，因为它们能控制绿头苍蝇的数量。这些巨蜥作为自然的清洁工，能清除遍布蛆虫的动物尸体，但在试图恢复自然环境的野化项目中，它们却被严重忽视了。为了了解更多，我去见了剑桥大学动物学系的汤姆·杰米森。

James Tickow in conversation with GP and author Adam Staten. Scientists at the University of Cambridge have found that giant lizards called Heath goannas could be saving Australia sheep farmers a lot of money every year by keeping down blowfly numbers. The goannas act as a natural cleanup crew by clearing maggot ridden animal carcasses from the landscape, but they've been sorely overlooked by rewilding projects that are trying to restore the natural environment. To find out more, I went to meet Tom Jamieson from the University of Cambridge's Department of Zoology.

‘回报项目’是一种景观恢复形式，我们旨在重新引入当地已灭绝的物种，以恢复它们为生态系统提供的功能和服务。我真正感兴趣的是研究爬行动物在这类项目中可能扮演的角色，因为它们长期被严重忽视。所以我们确实想研究爬行动物的作用，当它

So Rewarding Projects is this form of landscape restoration, where we're looking to reintroduce locally extinct species in order to restore the functions and services they provide to ecosystems. And what I'd be really interested in studying is the role that reptiles might play in these sort of projects as they've been massively overlooked. So we've, yeah, wanted to study the role of reptiles when it

涉及到野化项目时。具体是哪些爬行动物？根据你提到的南澳大利亚地理环境，我能猜到你要说什么，但具体是哪些种类的爬行动物呢？

comes to rewilding projects. Which reptiles? I can imagine, given the geography you've referenced, South Australia, I can imagine what you're gonna say, but what sort of reptiles specifically?

我一直在研究巨蜥，作为这个项目的案例研究对象。这类巨蜥或澳洲巨蜥，包括科莫多龙及其近亲，在澳大利亚具有极高的多样性。全球80种巨蜥中，澳大利亚就有约30种。因此我们重点研究如何将它们纳入野化项目并进行管理。

So I've been working on looking at monitor lizards, a bit of a case study to this project. So these are monitor lizards or goannas. This is the group that includes the Komodo dragon and its relatives, and they have a really high diversity in Australia. Of the 80 species world round worldwide, there's around 30 or so in Australia. And so we've really focused in to see how they could be managed and included in rewilding projects.

有些体型确实巨大。你提到科莫多龙。我在澳大利亚见过的一些巨蜥体型非常庞大。

Some of them are really big. You say Komodo dragons. Some of the ones I've I've come across in Australia are massive.

完全正确。比如栖息在中部地区的澳洲巨蜥，它们非常美丽，是仅次于科莫多龙的世界第二大蜥蜴，体长能超过两米。

Absolutely. Yes. So we've got species like the perentie that live right in the center, and these are, you know, beautiful. They're the second largest lizard in the world after Komodo dragon. They can get over two meters long.

我还在研究科莫多龙的另一种近亲——希斯巨蜥。这是非常可爱的物种，体色从亮黑色到橄榄棕不等，全身布满美丽的黄色斑点与条纹，生活在南海岸地区，体长约1.5米。

And I've been looking another relative of the Komodo dragon, the Heath glanor. So these are a lovely lovely, very pretty species. They sort of range from a glossy black to an olive brown covered in these these beautiful yellow spots and stripes, and they live down on the South Coast. They grow to about 1.5 meters long.

你们具体是怎么开展研究的？

How have you gone about it?

我们参与了Marnabungurra项目。这是一个位于澳大利亚南海岸中心的约克半岛上的生态恢复项目，旨在修复17万公顷的土地，面积大约相当于伦敦的大小，确实非常广阔。

We worked down with the Marnabungurra project. So this is a rewilding project down on the South Coast Of Australia, right right in the center of the South Coast Of Australia on a place called the York Peninsula. This project aims to restore a 170,000 hectares of landscape. So this is roughly the the footprint of the size of London. So it's a really, really massive area.

这片区域大部分本土哺乳动物已经消失，约90%的原生哺乳动物因入侵物种狐狸和猫的捕食而灭绝。整个项目的核心是通过重新引入这些本土物种来修复生态系统。虽然项目重点在哺乳动物，但我们特别关注了爬行动物的作用——尤其是那些1.5米长的石龙子，它们如何通过生态服务功能促进景观恢复。

And from this area, most of the native mammals have been lost. Most a lot of the native species have been lost, so 90% of the native mammal fauna is gone from this huge region, mainly as a result of predation by invasive foxes and cats. And so the wider project what they seek to do is to be reintroducing these native species in order to try and restore that landscape. So the project was very focused on mammals and what we wanted to focus on was what the role of reptiles might be in this project, particularly these heath gwanas I've mentioned, 1.5 meter long lizards, how they might be able to contribute to restoring this landscape, to these ecosystem services.

你是指随着生态恢复，它们的数量变化会引起景观的连锁反应吗？还是说它们本就存在，在生态恢复过程中发挥作用？或者是两者兼有？

Do you mean as in in terms of as you restore the landscape, their numbers change and the landscape responds to their numbers changing? So it's like a domino effect. Is that what you mean by what they do or is it that they're already there and as the landscape rewilds they make a contribution to that process or both?

两者都有。我们的研究重点是：通过重新引入个体或采取保护措施来增加其数量后，这些种群增长会因其在生态系统中的行为产生怎样的连锁效应。具体而言，我们想研究种群增长如何通过食腐行为影响更广泛的生态系统。

It's a little bit of both. The idea is what we want to study is if we're looking to boost their numbers, so by either reintroducing individuals to help grow the population or by doing specific conservation measures to help those populations grow, the knock on effects that they will then have as their populations grow by the different things they do in the landscape. And so in this case we wanted to look at specifically how as those populations grow through conservation measures how they're contributing to scavenging and what benefits that might have for the wider ecosystem.

它们具体有什么贡献？

And what do they contribute?

我们发现它们是极其高效的食腐动物，能大量清理动物尸体，从而减少尸体滋生的有害生物（如传播疾病的丽蝇）。这些石龙子实际上提供了至关重要的生态服务。

So what we found is that they are really, really effective scavengers. They remove large amounts of carcasses from the landscape and that reduces a lot of the problematic things that might breed in these carcasses. So things like blowflies, animals that cause disease, we find the presence of these Heath goannas removes this from the landscape. So they're providing a really, really important service.

说到苍蝇——就像此刻罗切斯特这里困扰我们的这些——既然它们能控制苍蝇数量，而澳大利亚另一大产业是牛羊养殖，这是否也会对这些产业产生连带效益呢？

And presumably, I mean, enormous number of flies around us here probably preying on us here in Rochester, but presumably, if you remove those flies, because one of the other big Australian industries is gonna be things like sheep and cattle, does that have a knock on benefit for those industries?

确实如此。我们之所以关注石龙子对丽蝇数量的影响，是因为在我们研究的这片广阔区域内有大量绵羊农场，而丽蝇正是这些农场面临的主要问题。它们会引发一种名为蝇蛆病的可怕疾病——丽蝇将卵产在绵羊的羊毛上，孵化后的蛆虫会钻入羊的皮肉，活生生地啃噬它们。这种疾病还会造成巨大的经济损失。

Absolutely, yes. So I mean, one of the reasons we were interested in looking at the effects that these heath goannas might have on blowfly populations is because in this wider landscape we're working in, there's a lot of sheep farms and a big problem for these sheep farms is these blowflies. They cause a disease called flystrike. A really horrible disease, the blowflies will lay their eggs on the wool of sheep, those eggs hatch into maggots and they burrow into the sheep's flesh and start eat them alive. It's a very costly disease as well.

澳大利亚绵羊养殖业每年因此损失2.8亿澳元。因此，任何能减少丽蝇数量、降低该病发生率的本土物种都可能产生积极影响，甚至带来显著的经济效益。我们的研究发现，这些石龙子确实能减少丽蝇数量，从而可能降低疾病发生率，对农业产生重大积极影响。

This costs the Australian sheep farming industry $280,000,000 a year, and so any role of, a native species that might be to reduce these numbers of blowflies, reduce the incidences of this disease could have really positive effects. Yeah. Quite big financial effects. And so so that's what we find. We find these go on as they reduce the numbers of these blowflies, and that potentially reduces incidences of this disease that might have some big positive effects for agricultural industry too.

所以请善待你们的蜥蜴。以上是汤姆·詹米森的研究，该研究刚发表在《生态与进化》期刊上。《科学时代》本周回归，将带来人工智能先驱杰弗里·辛顿的专题报道。

So look after your lizards. That was Tom Jamieson, and his study has just been published in the journal Ecology and Evolution. Times of Science is back this week and returns with the artificial intelligence pioneer, Geoffrey Hinton.

这里是《裸体科学家》节目，正在五台直播。

The Naked Scientists on five Live.

杰弗里·辛顿1947年6月12日出生于温布尔登。他是数学家兼教育家玛丽·艾佛勒斯·布尔与逻辑学家乔治·布尔的玄外孙。其显赫家族成员还包括外科医生兼作家詹姆斯·辛顿，以及数学家查尔斯·霍华德·辛顿。杰弗里曾就读于布里斯托尔的克利夫顿学院，后在剑桥大学攻读哲学与自然科学，1970年获得实验心理学学士学位。

Geoffrey Hinton was born on the 12/06/1947 in Wimbledon. He's the great great grandson of the mathematician and educator Mary Everest Ball and her husband, the logician, George Ball. Other notable family members include the surgeon and author James Hinton and the mathematician Charles Howard Hinton. Geoffrey attended Clifton College in Bristol before embarking on an array of undergraduate studies at Cambridge that took him philosophy and natural sciences. He graduated with a degree in experimental psychology in 1970.

他在爱丁堡大学继续深造，1978年获得人工智能博士学位。曾先后任职于苏塞克斯大学、加州大学圣地亚哥分校，并担任伦敦大学学院盖茨比慈善基金会计算神经科学单元的创始主任。杰弗里被誉为'人工智能教父'，他在神经网络与深度学习领域的开创性研究为ChatGPT等系统奠定了基础。去年从谷歌离职时，他曾警告人工智能可能带来的风险。

He continued his studies at the University of Edinburgh where he was awarded a PhD in artificial intelligence in 1978. He also worked at the University of Sussex, the University of California San Diego. He was the founding director of the Gatsby Charitable Foundation Computational Neuroscience Unit at University College London. Geoffrey is frequently described as the godfather of AI and his pioneering research on neural networks and deep learning has paved the way for systems that are familiar to many of us now, like ChatGPT. He warned of the dangers posed by AI when he resigned from Google last year.

我在剑桥读本科时，就对大脑运作机制产生了浓厚兴趣。研究大脑运作有两种途径：通过实验观测，或是建立计算机模型。当时用计算机模拟进行科学研究还算新兴方法，而这正是我理解大脑学习机制的理想途径。于是我专注于编写模拟脑细胞网络的程序，试图解答：脑细胞间的连接应该如何改变，才能让这个网络学会完成复杂任务——比如识别图像中的物体。

When I was an undergraduate at Cambridge, I got very interested in how the brain works. And there's two different ways you could study how the brain works. You could do experiments on the brain, or you could try and make computer models of how it works. So at the time, the idea of doing science by simulating things on computers was fairly new, and it seemed just like the right approach to trying to understand how the brain learns. So I spent my time writing computer programs that pretended to be networks of brain cells, and trying to answer the question, how should the connections between brain cells change so the collection of brain cells hooked up in a network can learn to do complicated things, like for example, recognize an object in an image.

时至今日，我们是否真正理解其运作原理？显然你一直在研究我们完全不了解的事物并试图建模。我们是已经成功了，还是仍在黑暗中摸索？

Do we have a clear idea even today of how that works? Because obviously you were working towards something we had no idea about and trying to model it. Have we got there, or are we still in the dark?

两者都不是。我们尚未完全成功，但理解已深入许多。现在我们有运行良好的计算机模型。你可以在这些大型语言模型中看到，也能从手机现在能识别物体这一事实中体会到。所以我们知道如何让这类事物运作，也明白大脑与这些系统颇为相似。

Neither of those. We haven't fully got there, but we have a much better understanding. So we now have computer models that work really well. You see that in these large language models, and in the fact that your cell phone can recognize objects now. So we understand how to make things like that work, and we understand that the brain's quite like many of those things.

我们尚不完全确定大脑具体如何学习，但对学习内容已有更清晰认知。它学习像这些大型神经网络一样运作。

We're not quite sure exactly how the brain learns, but we have a much better idea of what it is that it learns. It learns to behave like one of these big neural networks.

如果归根结底是脑细胞之间的交流，而这些只是庞大的连接群体，用计算机建模不是相对容易吗？卡在哪里？为什么这很难实现？

If it's down to the fact that we've got brain cells talking to brain cells, and they're just big populations of connections, is that not relatively easy to model with a computer? What's the holdup? Why is it hard to do this?

关键在于如何制定规则，使连接强度能根据网络获得的经验而变化。例如早在1940年代或1950年代初，心理学家赫布提出：如果两个神经元同时激活，它们之间的连接就会增强。但用计算机模拟时你会发现，所有连接都会变得过强导致系统崩溃。必须同时有使其弱化的机制。所以问题是：能否使其运作足够精准，以完成复杂任务，比如识别图像中的物体，或早期的手写数字识别？

Well, the tricky thing is coming up with the rule about how the strength of a connection should change as a result of the experience the network gets. So for example, very early on in the 1940s, or maybe early 1950s, a psychologist called Hebb had the idea that if two neurons, two brain cells, fire at the same time, then the connection between them will get stronger. If you try and simulate that on a computer, you discover that all the connections get much too strong and the whole thing blows up. You have to have some way of making them weaker too. So the problem is, can you get it to work well enough so that it can do complicated things, like recognize an object in an image, or in the old days, recognize something like a handwritten digit?

你需要收集大量数字2、3等的样本，测试系统能否区分二者。事实证明这相当棘手。你会尝试各种学习规则来发现哪些有效，从而更深入理解机制优劣。

So you take lots of examples of twos and threes and so on, and you see if you can make it recognize which is a two and which is a three. And it turns out that's quite tricky. And you try various different learning rules to discover which ones work, and then you learn a lot more about what works and what doesn't work.

哪些方法行不通？为什么？

What doesn't, doesn't work, and why?

好的，我来告诉你一个确实有效的方法，因为这显然更有趣。你有一层神经元模拟像素。所以图像由一大堆像素组成，像素有不同的亮度，这就是图像的本质。它只是表示每个像素亮度的数字。这些就是输入神经元。

Okay, I'll tell you something that does work, because that's obviously more interesting. You have a layer of neurons that pretend to be the pixels. So an image consists of a whole bunch of pixels, and the pixels have different brightnesses, and that's what an image is. It's just numbers that say how bright each pixel is. And so that's the input neurons.

它们告诉你像素的亮度。然后你还有输出神经元。如果你在识别数字，可能有10个输出神经元，它们告诉你这是哪个数字。通常网络一开始并不确定，所以会分散风险，比如会说这可能是2，也可能是3，但肯定不是4。它会通过让代表2的输出单元较为活跃来体现这一点。

They're telling you the brightness of pixels. And then you have output neurons. If you're recognizing digits, you might have 10 output neurons, and they're telling you which digit it is. And typically, the network, at least to begin with, wouldn't be sure, so it hedge its bets and it'd say, it's probably a two, it might just be a three, it's certainly not a four. And it would represent that by the output unit for a two would be fairly active.

代表3的输出单元会稍微活跃，而代表4的输出单元则完全沉默。现在问题是，你如何让这些像素作为输入，在输出端产生这些活动？现在所有大型神经网络都使用一种方法，叫做反向传播，原理是这样的：在输入和输出之间有几层神经元。代表像素强度的神经元会先连接到第一个隐藏层，然后是第二个隐藏层，接着第三个隐藏层，最后到输出层。

The output unit for a three would be a little bit active, and the output unit for a four would be completely silent. And now the question is, how do you get those pixels as inputs to cause those activities in the outputs? And here's a way to do it that all the big neural networks now use, and it's called back propagation, and it works like this. You have some layers of neurons between the inputs and the outputs. So the neurons that represent the pixel intensities have connections to the first hidden layer, and then the second hidden layer, then the third hidden layer, and finally to the outputs.

之所以叫隐藏层，是因为一开始你并不知道它们该做什么。这些网络最初只有随机连接，所以显然做不出任何合理的事情。当你输入一个数字图像时，它通常会对所有10个数字都分散风险，说它们可能性都差不多，因为它完全搞不清楚状况。然后你会问下面这个问题。

So they're called hidden layers because you don't know to begin with what they should be doing. And you start off with just random connections in these networks. So the network obviously doesn't do anything sensible. And when you put in an image of a digit, it will typically hedge its bets across all the possible 10 digits and say they're all more or less equally likely, because it hasn't got a clue what's going on. And then you ask the following question.

我该如何调整某一层神经元与另一层神经元之间连接的强度，让它能稍微更准确地得到正确答案？假设你只想区分2和3。一开始你给它一个2，它说有0.5概率是2，0.5概率是3——它在分散风险。然后你问：该怎么调整连接强度才能让它说51%是2，49%是3？

How could I change one of the strengths of the connections between a neuron in one layer and a neuron in another layer so that it gets a little bit better at getting the right answer? So suppose you're just trying to tell the difference between twos and threes. To begin with, you give it a two, and it says, with a probability 0.5, it's a two, with a probability 0.5, it's a three. It's hedging its bets. And you ask, well, how could I change connection strength so that it would say fifty one percent two and forty nine percent three?

你可以想象通过微调连接来实现。你可以选择网络中某个连接强度，稍微加强它，看看网络表现是变好还是变差。如果变差，显然就该减弱这个连接。这有点像进化过程：你选取一个底层变量（连接强度），然后思考如何微调它才能让系统表现更好？

And you can imagine doing that by just tinkering with the connections. You could choose one of the connection strengths in the network, and you can make it a little bit stronger, and see if that makes the network work better or work worse. If it makes it work worse, obviously you make that connection a little bit weaker. And that's sort of a bit like evolution. You're taking one of these underlying variables, a connection strength, and you're saying, if I change it a little bit, how can I change it to make things work better?

并保存这些改变。这样做最终肯定能成功，但会耗费大量时间。早期我们使用的网络只有几千个连接，而现在这些大型聊天机器人有数万亿个连接。用这种方式训练会永远都完不成。

And save those changes. You could do that, and it's obvious that in the end that will work, but it would take huge amounts of time. So in the early days, we would use networks that had thousands of connections. Now these big chatbots have trillions of connections. It would just take forever to train it that way.

但通过这种称为反向传播的算法，你几乎可以实现同样的效果。具体操作是输入一张图片，假设是数字2，初始权重是随机的，即连接上的权重。信息会通过网络前向传播，网络可能会说50%是2，50%是3。然后你向网络反向发送一个信息，这个信息实际上是在说：我希望你增加识别为2的概率，降低识别为3的概率。

But you can achieve pretty much the same thing by this algorithm called backpropagation. So what you do is you put in an image, let's say it's a two, weights are initially random, weights on the connections. So information will flow forward through the network, and it'll say 50% is a 250% is a three. And now you send a message back through the network. And the message you send back is really saying, I'd like you to make it more likely to be a two and less likely to be a three.

所以我希望你提高数字2的百分比，降低数字3的百分比。如果你以正确的方式发送这个反馈信息，就能同时计算出所有连接应该如何微调，使得结果更准确一些。这就是所谓的反向传播。它运用了微积分，但本质上是在调整连接强度——这种调整原本需要通过进化过程逐个改变，而反向传播算法能同时计算出所有连接应该如何微调才能让系统表现更好。因此如果你有万亿个连接，这种方法就比逐个调整高效万亿倍。

So I'd like you to raise the percentages on two and lower the percentages on three. And if you sent the message back in the right way, you can figure out for all the connections at the same time how to change them a little bit so the answer is a little bit more correct. That's called back propagation. It uses calculus, but it's essentially doing this tinkering with connection strengths that evolution would do by just changing one at a time, but the back propagation algorithm can figure out for all of them at the same time how to change each one a tiny bit to make things work better. And so if you have a trillion connections, that's a trillion times more efficient than just changing one and seeing what happens.

但底层网络如何知道上层会做出什么调整，以确保自己获得的输入是正确的？这样它所做的调整及其产生的概率才能更优化，从而避免自身调整后产生的信号前向传播到网络中改变了其他部分，结果反而使自己的配置变得不再最优——你能明白我的意思吗？

But how does the layer at the bottom know what's gonna be changed above it to make sure that the input that it then gets is the right one, so that the change it's just made to it and its probability ends up being even better, so that you don't end up changing yourself, then that feeds forward back up the network, changes something else, but then it becomes less optimal for you, if you get what I'm saying.

我完全理解你的意思。这是个非常好的问题。本质上，网络中较早的连接其实是在做一种假设：假设其他所有连接保持不变，改变我这个连接的强度会如何改善结果？所有连接都在进行这样的计算。

I get just what you're saying. It's a very good question. And essentially what's happening is, if you take a connection early in the network, it's kind of making an assumption. It's saying, suppose all the other connections stayed the same, how would changing my connection strength make things better? And they're all doing that.

如果你大幅改变连接强度，情况确实可能变得更糟。因为虽然单独调整某个连接可能改善结果，但当所有连接同时调整时反而会恶化。但事实证明，只要调整幅度非常小，这个问题就会消失。当调整幅度很小时，我计算出如何改变某个连接强度，而其他连接的调整幅度也都非常小，这样就不太可能把有益的调整变成有害的——比如把原本改善结果的调整变成损害结果的调整。

So if you change the connection strengths by a lot, things could actually get worse, because you could choose a way to change each connection strength. But if you did that change alone, would make things better. But when you do all the changes at the same time, it makes things worse. But it turns out, if you make the changes very small, that problem goes away. If you make the changes very small, then I figure out how to change one connection strength, and because the changes in all the other connection strengths are very small, it's very unlikely they'll turn, for example, a change that helps into a change that hurts.

不过这些独立网络层能否用'引号'告诉我们它们的想法？因为研究人员常向我反映的问题是：他们非常希望知道，当构建这类系统时，系统是如何得出结论的——即所谓的可解释性。比如当经过癌症识别训练的系统看到一张图片时，它应该能解释图片中哪些具体特征让它判定这些细胞是癌变的。有些模型能做到这点，但有些不能。

Can those individual layers tell us what they think in inverted commas though? Because one of the problems that researchers, when I go and talk to them, say to me is that they would very much like to know how when they build these sorts of systems, it's arriving at its conclusion. It's so called explainable. So when it sees a picture of cancer having been trained to recognize cancers, it can explain what particular features of the picture it saw singled out those cells as cancerous. And some models do this, but others don't.

那么那些能做到可解释性的模型，是通过网络层能告诉你它们做了哪些改变才得到最终输出的这种方式实现的吗？

Now is the way that they do do it by those things being able to tell you what they changed in order to make the output that they got?

与其说是告诉你它们改变了什么，不如说是解释它们如何运作。举例来说，假设我们试图区分数字2和3，在接收PIC细胞输入的神经元层中，你可能会发现其中某个神经元正在寻找图像底部附近的一排明亮像素，其上方和下方各有一排暗像素。它通过与该排需亮像素建立强正连接权重，与需暗像素建立强负连接权重来实现这一点。如果它通过自我学习形成这样的连接模式，就会非常擅长检测水平线。这个特征可能是区分2和3的有效方法，因为数字2底部通常有水平线而3没有。

It's not so much tell you what they changed, but tell you how they work. So for example, if you take the layer of neurons that receives input from the PIC cells, let's suppose we're trying to tell the difference between a two and a three, you might discover that one of those neurons in that layer is looking for a row of bright pixels that's horizontal near the bottom of the image with a row of dark pixels underneath it and a row of dark pixels above it. And it does that by having big positive connection strengths to the row of pixels that it wants to be bright, and big negative connection strengths to the row of pixels it wants to be dark. And if you wind it up like that, or rather if it had learned to wire itself up like that, then it would be very good at detecting a horizontal line. That feature might be a very good way to tell the difference between a two and a three, because twos tend to have a horizontal line at the bottom and threes don't.

这对于第一隐藏层（即第一层特征检测器）是可行的。但一旦网络层级加深，要理解其实际运作机制就变得极其困难。虽然相关研究很多，但在我看来，要合理解释这些具有多层结构的深度网络为何做出特定决策，将会面临极大挑战。

So that's fine for the first hidden layer, the first layer of feature detectors. But once you start getting to deeper in the network, it's very, very hard to figure out how it's actually working. And there's a lot of research on this, but in my opinion, it's going to be very, very difficult to ever give a realistic explanation of why one of these deep networks with lots of layers makes the decisions it makes.

你刚才解释的工作原理是否具有普适性？也就是说，如果我观察任何这类模型，它们很可能都以类似方式运作。如果是这样的话——

Is the explanation you've given me for how this works pretty generic? So if I took any of these models, they're probably working in a similar sort of way. And if so

是的。

Yes.

当有人说'我正在研究人工智能'时，鉴于我们已经拥有这类基础平台，他们实际在研究什么？我们如何尝试从你描述的这个核心基础算法出发，对人工智能进行改变、改进或开发？

When someone says, I'm working on AI, given that we have that sort of platform, what are they actually working on? How are we trying to change, improve or develop AI away from that main principle, that core fundamental operating algorithm that you've described for us?

总体而言，我们并非要脱离这个算法进行开发（尽管确实有人这么做）。我们主要致力于设计能充分发挥该算法效能的架构。以自然语言理解为例：大约在2014年，神经网络突然在语言翻译方面表现出色。

On the whole, we're not trying to develop it away from that algorithm. Some people do. But on the whole, what we're trying to do is design architectures that can use that algorithm to work very well. So let me give you an example in natural language understanding. In about 2014, neural networks suddenly became quite good at translating from one language to another.

比如输入一串英语单词，要求输出对应的法语单词序列。具体而言，给定英语单词串后，网络需要生成句子的首个法语单词；然后结合英语单词串和已生成的首个法语单词，预测第二个法语单词——它始终在尝试预测下一个词。通过大量英法语句对进行训练来实现这一点。

So you'd give them as inputs a string of English words, and you'd want them as outputs to produce a string of French words. In particular, given some string of English words, you'd like them to produce the first French word in the sentence. And then given a string of English words plus the first French word in the sentence, you'd like them to produce the second French word in the sentence. So they're always trying to predict the next word. And you train them up on lots of pairs of English and French sentences.

首先，在2014年刚开始时，当你试图预测下一个单词时，会受到之前所有单词的影响。后来人们发现，与其让所有前文单词产生同等影响，不如重点关注与你相似的词汇并赋予更大权重。我们并非要摒弃基础算法或绕过它，而是通过引入注意力机制等要素来增强其性能。

And to begin with in twenty fourteen, when you're trying to figure out the next word, you'd have influences from all the previous words. And then people discovered a bit later on that rather than letting all the previous words influence you equally, what you should do is look at previous words that are quite similar to you, and let them influence you more. And so you're not trying to get rid of the basic algorithm or circumvent it, you're trying to figure out how to supplement it by wiring in certain things like attention that make it work better.

当我们听说大型语言模型存在'幻觉'问题（这种现象如今愈发明显），即它们会编造不存在的内容并以极高权威性输出时——这种行为从何而来？它们为何能生成这些虚假内容？

When we hear that one problem with the large language models that we're seeing manifest very much now is that they can hallucinate, where does that behavior come from? How do they generate these spurious things that don't exist, but they're said with enormous authority the outputs from these sorts of engines? Where does that come from?

首先我要纠正一点：这应该被称为'虚构'而非'幻觉'。在语言学范畴我们称之为虚构。这种现象在1930年代就被广泛研究过。首先要明白，这使它们更像人类而非相反。

So first, let me make a correction. It ought to be called confabulation, not hallucination. When you do it with language, it's called confabulation. And this was studied a lot in people in the nineteen thirties. And the first thing to realize is that this makes them more like people, not less like people.

如果你让一个人回忆很久以前的事，他们会信心十足地说出大量错误细节。这正是人类记忆的典型特征，毫不例外。人类记忆本就如此。所以当你担任陪审员时，必须对人们的记忆保持高度怀疑。

So if you take a person and you ask them to remember something that happened quite a long time ago, they will with great confidence tell you a lot of details that are just wrong. And that's very typical of human memory. That's not exceptional at all. That's how human memory is. And that's why if you're ever on a jury, you should be very suspicious when people remember things.

人们的记忆经常出错。在这方面，大型聊天机器人与人类如出一辙。它们和人类之所以如此，是因为信息并非被字面存储。我们习惯了计算机内存的工作方式——你可以存储一串词语，之后能准确检索原样内容。但大型聊天机器人并非如此运作。

They often remember things wrong. So the big chatbots are just like people in that respect. And the reason they're like that and the reason people are like that is you don't actually store things literally. We're used to a computer memory, where you could take, for example, a string of words, then you could store it in the computer memory, and later you could go and retrieve that string of words, and you get exactly the right string of words. That's not what happens in these big chatbots.

大型聊天机器人会分析词语序列，通过调整网络权重来预测下一个单词。目前它们在虚构方面还不如人类，但正在不断进步。

What the big chatbots do is they look at strings of words and they're trying to change the weights in the network so that they can predict the next word. Now the chatbots are worse than people at confabulating, but they're getting better.

令我担忧的是：我们对人言总会持保留态度（对某些人更甚），却对机器抱有极大信任，因为它们在我们心中是完美运行的。而现在我们使用的机器在某些方面表现得越来越像人类，也具有人类的缺陷——正如你刚才所述。这是否意味着未来我们需要教育人们：不要认为机器是绝对可靠的？

The worry to me is that we regard what people say with a pinch of salt, some more than others. But we tend to have this enormous trust that we place in machines because they behave in a perfect way to our mind. And we're now using machines that behave more like people and have people's flaws in some respects, as you've just been outlining to us. So are we gonna have to educate people not to think about machines as quite so reliable in future?

是的。我们研发的这些大型聊天机器人就像是一个新物种，既与我们非常相似，又与普通计算机程序截然不同。我们必须学会不能像对待老式计算机程序那样依赖它们。你不能这样做。

Yes. What we produced in these big chatbots is it's like a new species that's very like us and very unlike a normal computer program. We have to learn not to treat the chatbots like you would have treated an old fashioned computer program where you could rely on it. You can't.

我们之前交谈时，你提到最初使用计算机是为了理解大脑运作方式。但现在看来，计算机和你描述的事物正在向我们展示自然界的运作规律，这就像是一个循环正在闭合。

When we were talking earlier, you said you started using computers to understand how, say, a brain worked, but it strikes me that we're now at a position where computers and things like you've been describing are showing us how nature works. It's almost like the the loop is closing.

没错。我认为通过研发这些大型聊天机器人，我们对语言有了更深入的理解。过去像乔姆斯基这样的人认为语言是与生俱来的，不是后天习得的。但现在这种观点变得不太可信，因为这些聊天机器人最初只是随机权重，通过观察英语字符串就能学会说一口流利的英语。这让我们对人类自身的运作方式有了很多新认识。

Yes. I mean, I think we've understood a lot more about language from producing these big chatbots. So in the old days, people like Chomsky said that language was innate, it wasn't learned. Well that's become a lot less plausible because these chatbots just start off with random weights, and they learn to speak very good English just by looking at strings of English and learning. It's told us a lot about how we work.

我们的工作方式与它们非常相似，所以不应该比信任人类更信任它们。实际上，我们应该比对人类的信任更少些。

We work very like them, so we shouldn't be trusting them any more than you trust a person. We should probably trust them less than you trust a person.

你是何时获得'人工智能教父'这个称号的？因为我们的对话直接切入到了最艰深的部分。我们是如何走到今天这个位置的？很多人突然觉得AI是现在才出现的，但你获得博士学位时我才刚出生。过去四十多年发生了什么？幕后有哪些故事？你在这个过程中扮演了什么关键角色？

When did you get this name of being the godfather of AI? Because we jumped straight into the hard stuff with our conversation. How did we get to the position we're in today? Because a lot of people suddenly think AI has arrived on the scene here and now, but you got your PhD in it not long after I was born. So what has happened in the last forty something years, and what's been going on in the background, and what was your role in being so instrumental in it?

让我打个比方，因为科学史上发生过一件非常相似的事。在1910或1920年代，一位叫魏格纳的气候研究者提出大陆漂移说，认为南美洲的凸起部分与非洲的凹陷如此吻合并非巧合，它们曾经是一体的。此后约五十年间，地质学家们都认为这是无稽之谈。魏格纳生前未能看到自己的理论被证实。

Well, let me give you an analogy, because there's another thing that happened in science that's actually quite similar. So in the nineteen tens or nineteen twenties, someone who studied climate called Wegener decided that the continents had drifted around, and that it wasn't just a coincidence that that bulge on South America fitted nicely into the armpit of Africa. They actually had been together and they come apart. And for about fifty years, geologists said, this is nonsense, continents can't drift around, it's complete rubbish. And Wegener didn't live long enough to see his theory vindicated.

到了1960年代左右，人们在大西洋中部发现海底扩张现象，大陆确实在分离并形成新的地壳。突然间地质学家们改口说魏格纳一直都是对的。神经网络的发展也经历了类似过程。早期关于智能系统的理论分为两派：一派认为可以通过具有随机连接的神经元网络从数据中学习连接强度，但当时没人知道具体方法；另一派则认为应该像逻辑推理那样运作。

But in, I think, the 1960s or sometime like that, they discovered in the middle of the Atlantic, there's this stuff bubbling up where the continents are moving apart and it's creating new stuff, and suddenly the geologists switched and said, oh, he was right all along. Now with neural nets, something similar has happened. So back in the early days of neural nets, there were two kinds of theories of how you could get an intelligent system. One was you could have a big network of neurons with random connections in, and it could learn the connection strengths from data, and nobody quite knew how to do that. And the other was it was like logic.

你曾有一种内部语言，类似于精炼过的英语，还有一套规则来操作这种精炼英语中的表达，通过这些规则可以从前提推导出新结论。比如我说'苏格拉底是人'，又说'人皆有一死'，就能推断'苏格拉底会死'。这就是逻辑，大多数从事AI研究的人——实际上不久后几乎所有人——都认为这是智能的好模型。结果证明并非如此。神经网络才是更好的智能模型，只是对大多数人来说这简直难以置信。

You had some internal language, sort of like cleaned up English, and you had rules for how you manipulated expressions in cleaned up English, and you could derive new conclusions from premises by applying these rules. So if I say, Socrates is a man, and I say all men are mortal, I can infer that Socrates is mortal. That's logic, and most people doing AI, in fact almost everybody after a while, thought that that's a good model of intelligence. It turned out it wasn't. The neural nets was a much better model of intelligence, but it was just wildly implausible to most people.

所以如果在20年前问别人：'能否让一个初始连接随机的神经网络通过海量数据学习说出地道的英语？'人们会说：'不，你完全疯了，这绝不可能。它必须要有先天知识，必须内置某种逻辑。'然而他们大错特错。

So if you'd asked somebody, even 20 ago, if you'd asked them, could you take a neural network with random initial connections and just show it lots and lots of data and have it learn to speak really good English? People would have said, no, you're completely crazy. That's never going to happen. It has to have innate knowledge and it has to have some kind of built in logic. Well, they were just wrong.

当我们听说人们给AI设置防护措施时，这是否意味着存在某种它会触碰到的边界？就像你解释的那样存在自由与控制连接的平衡。但当我们想对它说'不，我不希望你虚构黑人纳粹'——这是之前出现的问题，它生成各种图像时，不是有例子显示那些图像完全不符合历史或荒诞不经吗？现在据说已经修复了。你如何让系统避免再犯这类低级错误？

When we hear that people put safeguards around AI then, is that where you've got a sort of barrier that it rubs up against? As in you've got the freedom and the control connections in the way that you've been explaining. But when we want to say to it, no, I don't want you to invent black Nazis, which is the problem we had before, it coming up with all kinds of generated images, wasn't there as an example shown that the images were completely historically inappropriate or implausible, and now that's been fixed or allegedly has been fixed. How do you then lean on your system so that it doesn't make silly mistakes like that?

首先用大量数据训练系统。除非数据经过严格清洗，否则难免包含不良内容。人们随后尝试通过训练消除这些偏见，有时会矫枉过正。具体做法是雇佣人员与聊天机器人互动，当机器人出错时进行标注，或者让人类在机器人生成的多个回复中选择更合适的那个。

You first train up a system on a lot of data, And unless you've cleaned the data very carefully, the data contains unfortunate things. People then try and train it to overcome those biases. Sometimes they get a bit over enthusiastic. And one way to do that is you hire a bunch of people who get your chatbot to do things, and then the people tell you when the chatbot does something wrong. Or the chatbot maybe makes two different responses and people tell you which is the preferable response.

接着对聊天机器人进行更多训练，使其做出优选回复而避免不良回复。这被称为人类强化学习。但遗憾的是，这种额外训练往往容易被绕过。如果向公众发布神经网络的权重参数，人们就能训练系统突破所有人类强化学习，重新表现出种族主义倾向。

And now you train the chatbot a bit more, so it makes the preferable responses and doesn't make the other responses. And that's called human reinforcement learning. Unfortunately, it's often easy to get around that extra training. If you release to the public the weights of the neural network, then people can train it to overcome all that human reinforcement learning and start behaving in a racist way again.

但为什么系统不够'聪明'？（我谨慎地使用这个带引号的词）它为何不能意识到错误并自我纠正？比如发现'这里展示的群体历史上并不存在，那肯定是错的，我要修正'？为什么缺乏这种自校正能力？

But why is the system not bright enough? I'm using that word carefully and in inverted commas, to know that it's getting it wrong. Why does it not then go, well, hang on a minute. There there weren't a particular group represented that I'm showing here, so that must be wrong, I'll correct for that. Why does it not self correct?

最初在谷歌投入大量工作减少偏见前，系统本不会生成黑人纳粹图像。但谷歌为减少偏见所做的努力反而导致系统开始生成这类图像。这很不幸。但要记住：当系统生成纳粹图像时，它并非在回忆某个具体纳粹分子，只是在呈现它认为合理的内容。

So probably initially, before Google put a lot of work into getting it to be less biased, it wouldn't have produced black Nazis. But Google put lots of work into making it what it thought was less prejudiced, and as a result it started producing black Nazis. That's unfortunate, But you have to remember, when it's producing a picture of a Nazi, it's not actually remembering a particular Nazi, it's just saying what it finds plausible.

你后来加入了谷歌。是什么让你决定说，实际上，我要离开？

You went on to join Google. What led to you saying, actually, I'm gonna leave?

我当时在研究如何制造模拟计算机，它们会比我们用于这些聊天机器人的数字计算机消耗更少的能量。在研究过程中，我意识到数字计算机实际上比我们拥有的模拟计算机（比如大脑）更好。所以我一生大部分时间都在用数字计算机建立模型来试图理解大脑。我曾以为越接近大脑的模型效果越好。但后来我意识到，这些模型能做到大脑做不到的事，这让它们非常强大。

I was working on how to make analog computers that would use a lot less energy than the digital computers we use for these chatbots. And while I was doing that, I came to the realisation that the digital computers were actually just better than the analog computers we have like the brain. So for most of my life, I was making these models on digital computers to try to understand the brain. And I'd assumed that as you made things more like the brain, they would work better. But there came a point when I realized that actually, these things can do something the brain can't do, and that makes them very powerful.

它们能做到而大脑做不到的是：你可以拥有同一个模型的多个完全相同的副本。在不同计算机上模拟完全相同的神经网络。由于是数字模拟，你可以让它们行为完全一致。然后你制作大量副本，一个副本学习互联网的一部分内容，另一个副本学习另一部分，每个副本都会自主决定如何调整权重以更好理解其对应的互联网内容。当所有副本都确定好权重调整方案后，你可以让它们统一按照所有副本的平均值来调整。

And the thing that it can do that the brain can't do is you can have many identical copies of the same model. So on different computers, you simulate exactly the same neural network. And because it's a digital simulation, you can make it behave in exactly the same way. And now what you do is you make many, many different copies, and one copy you show one bit of the Internet, another copy you show another bit of the Internet, and each of the copies begins to learn for itself on its bit of the Internet, and decides how it would like to change its weights, so that it gets better at understanding that bit of the Internet. But now, once they've all figured out how they'd like to change their weights, you can tell all of them just to change their weights by the average of what all of them want to do.

通过这种方式，每个副本都能获知其他所有副本学到的知识。这样同一个模型的数千个副本可以同时学习互联网的不同部分，每个副本都能从其他副本的学习成果中受益。这比我们人类的学习方式高效得多。人类需要产出句子，然后我得琢磨如何改变神经连接强度才可能产出那些句子——这就是我们称之为教育的缓慢而痛苦的过程。

By doing that, you allow each of them to know what all the others learned. So now you could have thousands of different copies of the same model, which could look at thousands of different bits of the Internet at the same time, and every copy could benefit from what all the other copies learned. So that's much better than what we can do. What you have to do is produce sentences, and I have to figure out how to change my connection strengths so I would have likely produced those sentences. And it's a slow, painful business called education.

这些模型不需要那种意义上的教育。它们能以我们无法企及的高效率共享知识，带宽远超人类。

These things don't need education in that sense. These things can share knowledge incredibly efficiently with much higher bandwidth than we can.

那么是什么让你认定是时候从谷歌离职了呢？

So what led to you deciding that it was time to call time at Google then?

人们的理解有些偏差。媒体喜欢编造美好故事。一个美好的故事可能是：我对AI的危险性深感忧虑因此离开谷歌。但事实并非如此。我当时已经75岁了。

People have sort of the wrong story. The media loves to make a nice story. And a nice story would have been: I got very upset about the dangers of AI and that's why I left Google. It wasn't really that. I was 75.

是时候退休了。我不再像从前那样擅长做研究。我想放松下来，看很多网飞剧。但我认为应该借此机会警示人工智能的危险。于是我与《纽约时报》记者交谈，警告了AI的风险，然后一切就乱套了。

It was time to retire. I wasn't as good at doing research as I had been. I wanted to take things easy and watch a lot of Netflix. But I thought I'd take the opportunity just to warn about the dangers of AI. And so I talked to a New York Times journalist and warned about the dangers of AI, and then all hell broke loose.

我对引发的巨大反响感到非常惊讶。

I was very surprised at how big a reaction it was.

真的吗？

Were you really?

是的。我没预料到会有如此巨大的反应。我想情况是这样的——你知道当大浪来临时，总有一群冲浪者想抓住浪头。而某个冲浪者恰好在正确时机划水，所以抓住了浪。但若问为何偏偏是他？

Yes. I didn't expect there to be this enormous reaction. And I think what happened is, you know how when the huge wave comes, there's a whole bunch of surfers out there who'd like to catch the wave. And one particular surfer just happens to be paddling just the right time so he catches the wave. But if you ask why was it that surfer?

只是运气使然。其实很多人早就警告过AI的危险，但我恰好在它成为热点时发声。加上我因过往研究积累的良好声誉，就成了压垮骆驼的最后一根稻草——虽然之前已有很多稻草堆在那里。

It was just luck. And I think lots of people have warned of the dangers of AI, but I happened to warn of it at just the time it became something of intense interest. And I happened to have a good reputation from all the research I'd done. And so I was kind of the straw that broke the camel's back, but there were a whole lot of other straws there.

你认为主要担忧有哪些？

What do you think those main concerns are?

好的，存在一系列不同担忧。我公开谈论的是所谓'生存威胁'。我选择公开是因为很多人说这是愚蠢的科幻情节，永远不会发生。这种威胁指AI将比人类更聪明并接管世界。我想指出这些AI与人类极其相似。

Okay, so there's a whole bunch of different concerns. And what I went public with was what's called the existential threat. And I went public with that because many people were saying this is just silly science fiction, it's never going to happen, it's stupid, it's science fiction. And that's the threat that these things will get more intelligent than us and take over. I wanted to point out these things are very like us.

一旦它们比我们更聪明，我们不知道会发生什么。但我们应该认真思考它们是否会接管，并尽我们所能防止这种情况发生。现在还有各种更紧迫的风险。最直接的风险是今年选举中可能发生的事，因为我们现在有了能制作逼真假视频、假声音和假图像的生成式AI，这可能会严重破坏民主。

And once they get smarter than us, we don't know what's going to happen. But we should think hard about whether they might take over, and we should do what we can to prevent that happening. Now there's all sorts of other risks that are more immediate. The most immediate risk is what's going to happen in elections this year, because we've got all this generative AI now that can make up very good fake videos and fake voices and fake images, and it could really corrupt democracy.

感谢杰弗里·辛顿。下周来到银石赛道的科学巨匠恰如其分地是英国一级方程式工程师丹·费洛斯。如果

And my thanks to Geoffrey Hinton. And as f one comes to Silverstone, next week's titans of science is appropriately enough the British Formula One engineer Dan Fellows. If

你

you'd

想参与讨论，在此期间，请给我们发邮件至5lifescience@bbc.co.uk。我是克里斯·史密斯，感谢收听。下次再见，再见。

like to join in the conversation, in the meantime, do drop us a line to 5lifescience@bbc.co.uk. In the meantime, from me, Chris Smith, thank you for listening. And until we're back together again, goodbye.

BBC广播五台。

BBC Radio five live.

我们正在推出马丁·刘易斯播客的全新分支，令人困惑的是它叫'非马丁·刘易斯播客'。我

We're launching a brand new off shoot of the Martin Lewis podcast confusingly called not the Martin Lewis podcast. I

有账单。

got bills.

我将提出一些关于我通常不涉及的专题的专业问题。其中一个是关于租房者及其权利的。我们将讨论遗产税，然后是关于如何为养老金储蓄的专题。最后，还有关于所得税和国民保险的所有复杂繁琐的规定。这些重要信息，都可以在‘非马丁·刘易斯播客’部分的马丁·刘易斯播客中找到。

I'm gonna be asking specialist questions about subjects I don't normally cover. There's one for renters and their rights. We'll be talking inheritance tax, then a special on how to save for your pension. And finally, all the complex and convoluted rules about income tax and national insurance. Serious information, all available in the not the Martin Lewis podcast part of the Martin Lewis podcast.

明白了吗

Get it

在BBC Sounds上收听。嗨。

on BBC Sounds. Hi.