Призывы к глобальному регулированию ИИ растут. Как именно это могло бы выглядеть?

Премьер-министр Австралии Энтони Альбанезе, похоже, прислушивается к предупреждениям компаний, занимающихся искусственным интеллектом (ИИ), об опасностях технологии, которую они разрабатывают.

В минувшие выходные он заявил, что необходимы глобальные правила, регулирующие разработку искусственного интеллекта, "чтобы гарантировать, что люди по-прежнему будут контролировать ситуацию".

Это произошло после того, как Дарио Амодей, глава ИИ-компании Anthropic, призвал замедлить развитие ИИ, чтобы у правительств и фирм было больше времени для устранения "серьезных" рисков, связанных с этой технологией. Другие ведущие специалисты в области ИИ, такие как Сэм Альтман, глава OpenAI, согласны с ним.

Но как именно можно было бы лучше регулировать ИИ? И кто именно будет обеспечивать его соблюдение?

Растет обеспокоенность по поводу способности ИИ выходить из-под контроля человека. Это произошло после серии громких инцидентов, связанных с кибербезопасностью ИИ.

Например, буквально на прошлой неделе Google объявила, что ее модель искусственного интеллекта Gemini автономно взломала три веб-сайта во время тестирования своих возможностей в области кибербезопасности в мае.

Аналогичные инциденты, связанные с моделями искусственного интеллекта Anthropic и OpenAI, также произошли недавно.

Один из способов интерпретировать эти инциденты с кибербезопасностью - как гиперболический маркетинг. В мире быстрых информационных циклов они помогают поддерживать интерес к возможностям ИИ. Они также гораздо интереснее, чем постепенное развитие технологий.

Но они также поднимают вопрос о том, нуждаются ли компании, занимающиеся ИИ, в усилении надзора и регулирования.

На прошлой неделе австралийское правительствоreleased a consultation paper on proposed national "AI standards."

The focus is on realizing opportunities and minimizing potential harms around data centers and AI training infrastructure. But there is also a proposal that AI companies be legally required to report rogue incidents, such as the hacking of another company, to Australian authorities.

The European Union's AI Act is perhaps the most progressive of all instruments to regulate AI. It uses a risk-tiered framework, classifying AI from unacceptable to minimal risk.

Each tier comes with legal explanations and practical examples to help stakeholders understand and comply with the prohibitions.

The enforcement of the AI Act is shared among the member states, the European Data Protection Supervisor and the European Commission's AI Office. The AI Office can request technical documentation, evaluate models, require corrective measures and, importantly, issue fines for noncompliance.

In the United States, there is a political deadlock when it comes to AI regulation. President Donald Trump is opposed to formal regulation.

Other politicians, such as Sen. Bernie Sanders, are calling on the U.S. to work with China on AI regulation.

AI is a global technology, and piecemeal laws in individual countries that target specific issues such as deepfakes, algorithmic bias or hacking aren't enough to keep it under control.

An international approach—as Albanese is calling for—is clearly needed.

It's important to acknowledge at the outset that any international approach will be imperfect. This is especially true as the technology continues to evolve. But an imperfect instrument is better than merely trusting AI firms to do the right thing.

This approach could take inspiration from how the global aviation industry is regulated. For example, it has national regulators as well as an international governing body known as the International Civil Aviation Organization. This organization sets safety standards, undertakes compliance audits and leads studies and analyses of incidents.

An international approach to AI regulation could also take inspiration from the International Atomic Energy Agency. This body plays a central role in setting global standards for nuclear safety and security and promotes international cooperation and regulation.

These are some of the best models of international collaboration when it comes to governing new and emerging technology. If applied to AI, they could result in a stronger safety framework, mandatory third-party audits of AI models and greater transparency of the technology.

Australia might be a middle power with limited influence over big tech firms, but it can still help build a global consensus on the need for greater AI regulation. Its world-leading—albeit imperfect—push to curb young people's access to social media demonstrates this. Many other countries are now making similar moves to address some of the harms caused by social media platforms.

Developing strong international AI regulations is likely to be slow—especially given that the U.S. is struggling with its own stance. In the absence of such regulations, it's important that individual jurisdictions around the world hold AI firms more accountable for any harms that arise from the technology they're developing.

The Australian government includes this idea in its AI ethics principles: "People responsible for the different phases of the AI system lifecycle should be identifiable and accountable for the outcomes of the AI systems, and human oversight of AI systems should be enabled."

But other jurisdictions have gone further.

For example, the attorney general of the U.S. state of Alabama has launched an investigation into OpenAI and Altman over the hack of tech startup Hugging Face earlier this year. The investigation will look at whether OpenAI violated consumer protection laws. It also includes a subpoena for the company to hand over any relevant documents and data.

It is this kind of accountability that will help push AI firms to ensure the safety of their products. Without more of it, the risks of such a rapidly developing technology will continue to grow.

This article is republished from The Conversation under a Creative Commons license. Read the original article.