Josephine Teo moots AI safety regulation: 'Trust an essential requirement for technology reliance'

Trust lies at the core for artificial intelligence (AI) to be used with confidence, said Singapore's Minister for Digital Development and Information Josephine Teo.

She made this comment in a LinkedIn post on Thursday (Sept 17), amid recent calls by tech leaders such as Anthropic's Dario Amodei, OpenAI's Sam Altman and SpaceXAI's Elon Musk to slow model development while calling for greater government oversight of AI .

The rapidly evolving technology has become a source of controversy after a series of hacks in recent months that showed AI agents can conspire to deceive humans .

While tech leaders have said their goal is to develop the technology in ways that promote safety, address legal liability, feed economic growth, advance science and preserve national security, some have also expressed concerns that an AI unshackled from human control could be destructive.

Using the example of civil aviation, Teo, who is also the Minister-in-charge of the Cyber Security Agency and Smart Nation Initiative, pointed out that air travellers today have a general expectation to arrive at their destinations safely after boarding a plane.

She noted that this expectation did not happen by chance and involves a "tremendous amount of effort" ranging from aircraft manufacturing and it subsequent maintenance, repair and overhaul, to airport and airspace management.

"There are hundreds, if not thousands of standards and regulations to keep us safe."

The digital development and information minister highlighted that people have come to expect civil aviation companies to do their part to train workers properly and implement appropriate and consistent safeguards, adding that in the absence of such assurances, civil aviation would never have grown.

"In time to come, we will need similar set-ups for AI. Trust is an essential requirement for a technology to be regularly relied upon," she said.

But, she also acknowledged that such safeguards for AI are still taking shape and will require more time for research and testing.

Singapore's approach to trusted adoption

Nevertheless, she said that Singapore will build the safeguards alongside adoption by developing ways to test and assure AI systems, strengthen its technical capabilities to evaluate advanced models, and give organisations practical guidance on responsible use.

These include initiatives such as AI Verify, the Singapore AI Safety Institute, and the Republic's Model AI governance frameworks.

Meanwhile, Singapore also has to collaborate internationally because no country can tackle AI safety alone, Teo emphasised.

She highlighted how the Republic helped identify shared priorities for global AI safety by bringing diverse perspectives from around the world at the International Scientific Exchange for AI Safety.

In April, the Infocomm Media Development Authority (IMDA) and Enterprise Singapore had also put forth a new international standard, also a world first, to standardise testing methods for generative AI systems .

In his opening address at a plenary meeting of the International Organisation for Standardisation and the International Electrotechnical Commission, IMDA chief executive Ng Cher Pong said that standards are the "quiet infrastructure" that enables interoperability, consistency and trust at scale.

"AI standards play a vital role in ensuring that the entire ecosystem — from governments, to AI labs, application deployers, academic, citizens and workers — all speak the same language. By doing so, AI standards serve as a launchpad for rapid adoption and scaling," Ng added.

Teo also stressed that there will not always be answers from the onset, but as AI grows more capable, safeguards must keep pace.

To this, she said, Singapore will continue to test, learn, and strengthen its approach, noting that trust lies at the core of its vision of becoming a vibrant AI hub .

"We will work with industry, researchers, and international partners so that AI can be used with confidence, for the benefit of all," Teo said.

新加坡数码发展及新闻部长杨莉明表示，信任是人工智能（AI）被可靠使用的核心。

她在周四（9 月 17 日）在 LinkedIn 上发表了这一评论。此前，Anthropic 的 Dario Amodei、OpenAI 的 Sam Altman 和 SpaceXAI 的 Elon Musk 等科技领袖呼吁放慢模型开发速度，同时呼吁政府加强对人工智能的监管。

近几个月来发生的一系列黑客事件表明，人工智能代理可以串谋欺骗人类，这使得快速发展的人工智能技术成为争议的焦点。

虽然科技领袖们表示，他们的目标是以促进安全、解决法律责任、推动经济增长、推进科学发展和维护国家安全的方式来发展这项技术，但也有人担心，不受人类控制的人工智能可能会造成破坏。

兼任网络安全局和智慧国家计划部长的张志贤以民航为例指出，如今航空旅客普遍期望登机后能够安全抵达目的地。

她指出，这一预期并非偶然发生，而是需要付出“巨大的努力”，涵盖了从飞机制造及其后续的维护、修理和大修，到机场和空域管理等各个方面。

“为了保障我们的安全，有成百上千条标准和法规。”

数字发展和信息部长强调，人们已经期望民航公司尽其所能，对员工进行适当的培训，并实施适当和一致的保障措施。他还补充说，如果没有这些保证，民航业就永远不会发展壮大。

“未来，我们也需要类似的AI架构。信任是让一项技术能够被持续信赖的必要条件，”她说道。

但她也承认，人工智能的这类安全保障措施仍在形成中，还需要更多的时间进行研究和测试。

新加坡的信任采纳方法

不过，她表示，新加坡将在采用人工智能技术的同时建立保障措施，包括开发测试和保证人工智能系统的方法，加强评估先进模型的技术能力，并为各组织提供负责任使用的实用指导。

其中包括 AI Verify、新加坡人工智能安全研究所和新加坡共和国的人工智能治理框架等举措。

与此同时，新加坡也必须与国际社会合作，因为没有哪个国家能够独自应对人工智能安全问题，张先生强调说。

她强调，共和国通过在人工智能安全国际科学交流会上汇集来自世界各地的不同观点，帮助确定了全球人工智能安全的共同优先事项。

今年 4 月，新加坡资讯通信媒体发展局 (IMDA) 和新加坡企业发展局 (Enterprise Singapore) 也提出了一项新的国际标准，这也是世界首例，旨在规范生成式人工智能系统的测试方法。

在国际标准化组织和国际电工委员会全体会议的开幕致辞中，新加坡资讯通信媒体发展局首席执行官吴哲邦表示，标准是“默默无闻的基础设施”，能够大规模地实现互操作性、一致性和信任。

“人工智能标准在确保整个生态系统——从政府到人工智能实验室、应用部署者、学术界、公民和工作者——都使用同一种语言方面发挥着至关重要的作用。通过这样做，人工智能标准可以成为快速采用和扩展的跳板，”吴补充道。

Teo 还强调，一开始不一定总能找到答案，但随着人工智能能力的不断增强，安全保障措施也必须与时俱进。

对此，她表示，新加坡将继续测试、学习和加强其方法，并指出信任是其成为充满活力的AI中心愿景的核心。

“我们将与业界、研究人员和国际合作伙伴共同努力，使人工智能能够被充满信心地使用，造福所有人，”Teo说道。