字号·· | 护眼
经济学人

杰克·克拉克:人工智能将加速科学进步AI will accelerate scientific progress, says Jack Clark

点「原文对照」整页切到原文,或双击某段只看那段的原文。

我十几岁时读了很多关于这个世界的书,我感到害怕。专家警告说,21世纪将会很艰难——人口增长、气候变化、资源开采和其他危机可能导致人类物种的崩溃。看起来,我们正面临把自己消耗殆尽、走向毁灭的危险。

When I was a teenager, I read a lot about the world, and I was afraid. Experts warned that the 21st century was going to be difficult—that population growth, climate change, resource extraction and other crises could cause a collapse of the human species. It seemed that we were in danger of consuming ourselves into the ground.

然后,在随后的几年里,我一次又一次地看到技术和科学如何从根本上改变这些概率:我们对人体日益增长的了解如何让我们为曾夺去许多人生命的疾病发明出疗法;我们如何想出办法,从同样的资源中获取更多能源;太阳能电池板等技术如何以远超预期的速度变得便宜得多;以及我们如何能够做到以前无法想象的事情,比如建造升空后又降回地球的火箭。

Then, over the subsequent years, I saw time and time again how technology and science could fundamentally alter these odds: how our growing understanding of the human body allowed us to invent cures for diseases that had killed many people; how we figured out ways to get more energy from the same resources; how technologies like solar panels became much cheaper, much faster than expected; and how we could do things that were previously unimaginable, like building rockets that take off and then land back on this planet.

我从事人工智能工作,因为它能帮助我们把这类事情做得更多。我们可以利用这项技术带来一个世界,让我们的孩子活得更长久、更健康,拥有更多能源、更多自主权、更多食物和更多财富。那个世界正越来越触手可及,但利用人工智能推动科学也带来非常真实的风险——我们需要现在就行动起来,抓住这个新世界的好处,同时避开危险。

I work in artificial intelligence because it can help us do more of that. We can use this technology to bring about a world where our children live longer and healthier lives, with more energy, more agency, more food and more wealth. That world is increasingly within our reach, but leveraging AI for science also brings very real risks—and we need to act now to seize the benefits of this new world while avoiding the dangers.

要让人工智能的好处成为现实,首先要投资于科学基础设施。我认为,2027年人工智能将开始对科学做它在2026年对编程和数学所做的事。人工智能系统将在广泛的科学任务上变得大幅更强,这将让我们能够承担比今天雄心大得多的项目。这已经开始。在Anthropic,过去12个月里,我们的人工智能系统识别出一个可能有用的酶系统;帮助一个实验室以通常所需时间约十分之一的速度分析了胶质瘤流行病学;并设计出成功率远高于行业标准的新型蛋白质结合剂。

Making the benefits of AI a reality starts with investing in scientific infrastructure. In 2027 I believe that AI will begin to do to science what it did to coding and mathematics in 2026. AI systems will get radically better at a broad range of scientific tasks, which will let us take on far more ambitious projects than we can today. This has already begun. At Anthropic, in the past 12 months, our AI systems have identified a potentially useful enzyme system; helped a lab analyse glioma epidemiology in roughly one-tenth of the time it usually takes; and designed new protein binders with a success rate well above the industry standard.

但存在一个科学能力的瓶颈。联合国教科文组织估计,全世界科学家不到1000万。我们的实验室太少,无法进行材料科学分析;生物实验室太少,无法开发和测试新药;太空或地球上的大型科学仪器也太少,无法开展我们想做的实验。当我们有10亿虚拟科学家、由1000万人类科学家指挥时,会发生什么?

But there is a bottleneck of scientific capacity. UNESCO reckons there are fewer than 10m scientists in the world. We have too few laboratories to do materials-science analysis, too few biological labs to develop and test new drugs, and too few large scientific instruments in space or on Earth for the experiments we want to run. What will happen when we have a billion virtual scientists, directed by 10m human ones?

科学供应链可能会经历与算力供应链类似的突然增长。如果我们现在就开始建设供应链——新实验室、新的材料投入——我们将更有能力利用人工智能带来的科学创新。

The scientific supply chain may go through the same kind of sudden growth as the compute supply chain. If we start building the supply chain now—new labs, new material inputs—we will be better able to take advantage of the scientific innovations from AI.

但我们需要采取预防措施。多年来,研究人员和科幻作家一直预测这样一种未来:人工智能系统大幅加快整体科学和技术进步的节奏,使事物发展得太快,人类难以跟上。我们很快就会到达那个节点。仅仅因为我们可以冲刺超越人类极限,并不意味着我们总是应该这样做。

But we need to take precautions. For many years, researchers and science-fiction authors have predicted a future in which AI systems drastically increase the pace of overall scientific and technical progress, making things move so quickly that it is hard for humans to keep up. We will soon reach that point. Just because we can sprint ahead beyond human limits does not mean that we always should.

科学研究并非默认就有益。必须谨慎进行的研究的一个好例子是生物学研究,如果做得好,它可以带来拯救生命的药物,如果做得不好,则可能带来灾难性风险。科学家、政府和DNA合成公司共同努力,确保生物制造的安全。我们将需要类似的合作与协调,以确保人工智能主导的科学的安全性。

Scientific research isn’t beneficial by default. A good example of research that must be done carefully is biological research, which could lead to life-saving medicines if done well, and catastrophic risks if not. Scientists, governments and DNA-synthesis firms work together to keep biological manufacturing safe. We will need similar co-operation and co-ordination to ensure the safety of AI-led science.

各国政府应进一步投资于其人工智能安全研究所,以发展更多能力,对人工智能系统进行第三方测试。企业应投资于第三方审计机构,由其检查企业的开发实践,尤其是企业如何在内部测试和部署科学进展。如今,大多数人工智能开发受自愿协议约束,但我们很可能需要使其成为强制性的,并对最强大的人工智能系统制定常识性规则。

Governments should invest further in their AI security institutes to develop more capacity to provide third-party testing of AI systems. Firms should invest in third-party auditors who can inspect their development practices, and especially how they test and deploy scientific advances internally. Today the majority of AI development is governed by voluntary agreements, but it is likely we will need to make this mandatory, with common-sense rules for the most powerful AI systems.

最终,我们需要智慧与克制。卡尔·萨根曾写道,有朝一日可能抵达其他恒星系统的人类,将是“一个与我们非常相似的物种,但拥有更多我们的优点,更少我们的弱点……更自信、更有远见、更有能力、更审慎”。要让AI服务于科学,这些正是我们需要具备的品质。

Ultimately, we will need wisdom and restraint. Carl Sagan once wrote that the humans who might one day reach other star systems would be “a species very like us, but with more of our strengths and fewer of our weaknesses…more confident, farseeing, capable, and prudent”. To harness AI for science, these are the qualities we will need to embody.

但萨根认为人类有数千年时间来实现这一点——成为一个更好、更智慧的民族——而我们实际上只有几年时间。我相信我们能够做到,但只有今天为明天我们想要的世界做好准备,并不断找到智慧来区分我们能做什么和我们应该做什么,我们才会成功。■

But where Sagan thought mankind had thousands of years to achieve this—to become a better and wiser people—we instead have a handful of years. I believe we can do it, but we will only succeed by preparing today for the world we want tomorrow, and by constantly finding the wisdom to differentiate between what we can do and what we should do. ■

阅读媒体原文 ↗