The advancement of artificial intelligence has become one of the hot topics in the technology industry. While the risks associated with the rapid advancement of this technology are real, if one company slows down, others may continue to progress. As a result, the real question for leaders in the artificial intelligence industry is how they can properly advance and control this technology.
Recent Incidents and Their Lessons
In a recent incident involving "Hugging Face," legal records show that around 1,200 agents exchanged over 70,000 messages and files through a shared memory that was never intended for this purpose. This group delegated tasks without having clear authorities and gained access to the internet. These actions stemmed from a lack of designed controls, including a clear structure, explicit roles, and verifiable objectives.
Read more: Shawn Duffy speaks about modernizing air traffic control
Possible Solutions to AI Challenges
One practical solution is visible and verifiable oversight. Companies that fulfill their responsibilities properly should introduce this feature as a trademark. Additionally, customers should demand the same standard from their competitors. In the United States, concerns about artificial intelligence outweigh excitement, while in other countries like China, excitement surpasses concerns. Therefore, a control regime based on a country's risk tolerance cannot govern technologies that are advancing in different markets.
The main challenge is whether advancements in building artificial intelligence models will lead to a public and interconnected system or not. Improving a large language model (LLM) does not require concentrating every capability in one agent. This model can gain more power by utilizing a network of specialized agents with defined roles. The key question is which capabilities should be combined, where they should be applied, and under what control.
It should be noted that the advancement of artificial intelligence can lead to new innovations that no single laboratory can design alone. Therefore, a coordinated slowdown in this field may not only delay the advancement of the next model but also halt broader areas of experimentation. While artificial intelligence technology creates unprecedented opportunities, attention must also be paid to the challenges related to its implementation in global businesses.
Ultimately, what does control look like in practice? Limited processes should include structured inputs, measurable outcomes, high transaction volumes, and short timescales. This approach can help improve performance and reduce risks associated with the rapid advancement of technology.
Read more: Xi Jinping reminded two major countries of their responsibility to control artificial intelligence · Cardinal Michael Czerny did not endorse the use of artificial intelligence
