Risks and Efforts Concerning Autonomous Decision-Making in the Development of Artificial Intelligence

2023-06-14

AI与人类价值观之间的不匹配可能导致这样的情况:AI系统可以在不考虑人类福祉的情况下自主决策。 In an interview with Sam Altman several years before ChatGPT pushed the AI world into uncontrollable frenzy, Elon Musk issued a ominous warning about the potential of artificial intelligence. Musk discussed the necessity of democratizing AI technology to prevent it from being controlled by a few, and emphasized how AI could be used for malicious purposes. This was in 2016. Today, seven years later, the fathers of AI and scientists are sounding the same alarm, only AI technology has advanced much further, and not all experts have united in their concerns about future AI risks. There have been many movements in the past few months, with people hoping to achieve some form of control before complete loss of control. In May of this year, nearly 30,000 people signed an open letter calling for a halt to the development of more powerful AI models than GPT-4, including AI experts Joshua Bengio, Gary Marcus, Elon Musk, and Steve Wozniak. The recent congressional hearing on AI may be the first step in government involvement, with Sam Altman and Gary Marcus advocating for the regulation of artificial intelligence. Another joint statement discussed the risks of artificial intelligence, equating it to nuclear warfare, and was signed by AI pioneers Geoffrey Hinton, Joshua Bengio, and DeepMind's Demis Hassabis, among others. However, there is disagreement among AI scientists and experts regarding the risks of artificial intelligence. Bengio recently expressed his concerns about the potential misuse of artificial intelligence, leaving him feeling perplexed and questioning his life's work. One of his biggest fears is the misuse of AI by "bad actors" to cause harm. If it falls into the wrong hands, such as the military, terrorists, or anyone who can manipulate these systems for nefarious purposes, the consequences would be catastrophic. In a detailed article, Bengio explained his concerns about losing control of the systems, leading to rogue AI. As systems become more advanced and autonomous, it could result in humans losing control and causing disastrous consequences. The mismatch between AI and human values could lead to scenarios where AI systems make decisions autonomously without considering human well-being. Bengio discussed the possibilities of future scenarios, where if we ask AI to solve climate change, it might design a virus that eliminates the population due to lack of clear instructions or viewing humans as the main drivers of the climate crisis. Last month, Hinton left Google and candidly expressed his concerns about AI technology. One of Hinton's biggest fears is the threat AI poses to humanity. He believes that if people pursue profit in AI development, the content generated by AI in the future will surpass that created by humans, endangering human existence. Neural networks of AI models have fewer connections than humans but can possess 1,000 times more knowledge and continuously learn and share knowledge, which is worrisome. He is also concerned that as chatbots become more prevalent, misleading information will increase. Biased viewpoints that appear in chatbots could have detrimental effects on society. It all boils down to the need for AI regulations, with Altman advocating for this in all his activities, even hoping for China's assistance in establishing AI safety guardrails. In a recent interview, Hinton emphasized the necessity of reaching consensus in order to move forward together. Just as climate scientists have a rough consensus on climate change, AI scientists should also unite to formulate good policies. He also believes that if scientists can compile a list of common viewpoints on technical issues related to AI, it would help in achieving consensus. If AI researchers present a variety of opinions, it becomes easier to make choices. Bengio emphasizes the need for governance and regulation to mitigate AI risks as AI systems become superintelligent and autonomous. He also discusses the need to properly track the progress of these models. Governments should be able to monitor and audit them, similar to what is done in any other industry such as automobiles and pharmaceuticals. There is also a need for some form of ethical training and certification for those working on such systems. While Marcus and Hinton advocate for regulation to prevent AI disasters, AI expert Yann LeCun believes these concerns are unfounded. He refers to the "apocalyptic" and "pessimistic" individuals who can imagine catastrophic scenarios but lack preventive measures as "people who can imagine catastrophic scenarios but lack preventive measures." With Meta's progress in AI, their GPT-2 has already reached the level OpenAI was at three years ago. Perhaps this is why predicting AI risks is a challenge for him. Source: https://analyticsindiamag.com/fuelling-ai-fathers-artificial-anxiety/

RECENT AI NEWS

RECENT AI TOOLS