Abstract:
Experts in the fields of national security, science and technology policy, and artificial intelligence from the United States and China recently jointly proposed an attention-grabbing proposal, advocating drawing on the nuclear security management experience of the Cold War to establish a security mechanism similar to that in the nuclear weapons field for advanced artificial intelligence systems that may appear in the future, in order to reduce the risks caused by technological loss of control, misjudgment, and escalation of international competition.
This suggestion appears in a research result formed after discussions between experts from both countries. The report believes that as artificial intelligence capabilities continue to improve rapidly, countries have begun to face some problems similar to those in the early nuclear era, that is, a technology with huge potential impact is developing rapidly, but the corresponding international rules, communication channels and risk control systems have not yet been fully established.
Experts participating in the study pointed out that in the future, more powerful artificial intelligence systems may play an increasingly important role in areas such as network security, military command, intelligence analysis, biotechnology research, and critical infrastructure management. Once these systems behave unexpectedly, are maliciously exploited, or are deployed incorrectly, their impact may span national borders and have an impact on the global security environment.
To this end, the report recommends that China and the United States refer to their historical experience in dealing with nuclear risks and gradually establish a multi-level security mechanism. These include strengthening information exchange, establishing crisis communication channels, collaborating on technology risk assessments, and exploring transparency and security testing regimes for state-of-the-art artificial intelligence systems.
Experts believe that although artificial intelligence and nuclear weapons are not the same, both have high strategic significance and involve international competition and global security issues. Therefore, it is often more effective to establish a protective framework in advance when the technology is in the rapid development stage than to wait until the risks are fully exposed before taking action.
The report also particularly emphasizes that the greatest danger in the future may not come from artificial intelligence itself, but may come from misjudgments caused by distrust between countries. For example, when one party cannot accurately determine what capabilities the other party is developing and to what level, it is more likely to overreact, misunderstand, or even escalate competition.
Some experts therefore proposed that communication channels similar to the "hotline mechanism" during the Cold War should be explored to enable countries to maintain immediate contact when major events involving artificial intelligence occur and avoid unnecessary security crises caused by information asymmetry.
At the same time, researchers believe that a more stringent security review mechanism should be established for the training and deployment of the most advanced artificial intelligence models. In the future, some systems with extremely high capabilities may need to undergo independent testing and continuous supervision, and establish an abnormal behavior reporting system to detect potential risks as early as possible.
The report points out that the current global artificial intelligence governance system is still in its early stages. Although countries have introduced different regulatory policies and industry norms, there are still obvious gaps at the level of cross-border collaboration. The development speed of artificial intelligence technology is accelerating, which means that the window for international cooperation may not exist for a long time.
Experts participating in the discussion generally believed that China and the United States, as the two most important players in the global artificial intelligence field, have special responsibilities in risk governance. If both parties can reach more consensus on security standards, risk assessment and crisis communication, it will help reduce the possibility of major technical security incidents in the future.
However, experts also admit that promoting such cooperation is not easy in the current international environment. Technological competition, national security considerations and industrial interests will all influence the relevant discussion process. Therefore, they prefer to start in relatively pragmatic areas such as risk management and security protection, rather than seeking comprehensive regulatory agreements from the beginning.
Analysts pointed out that this report reflects the changing discussion on artificial intelligence governance. In the past, the focus of all parties was more on privacy, copyright and employment impacts, but now more and more experts are beginning to turn their attention to long-term security and global strategic stability issues.
As artificial intelligence capabilities continue to improve, how to establish an effective security mechanism while maintaining innovation vitality is becoming a new topic faced by the global science and technology community and policy circles. Drawing on the risk management experience accumulated in the nuclear era may become an important reference direction for the future artificial intelligence governance framework.
Comments