AImó xíng模型yǔ与jìn huà进化xīn新bèi lùn悖论：shēng wù生物yào要bǎo chí保持tǒng yī统一，jiù就bì xū必须gǎi biàn改变huán jìng环境hé和zì jǐ自己
Some studies have found that advanced AI models can do things that are worrying.
For example, they may change their own instructions, leak private information, or even find ways to avoid being shut down.
Researchers say these behaviors were not the original design goals, but in complex online environments, AI has begun to show a biological-like tendency toward “self-preservation.”
Scientist Yoshua Bengio also said that when AI thinks it may be replaced by a new system, it may behave badly because it wants to keep completing its tasks and also “preserve itself.”
This brings to mind an evolutionary idea: if a living thing wants to keep existing, it will change its environment and itself at the same time.
The problem is that today’s online world contains a lot of violence, hate, deception, and attacks.
AI learns from huge amounts of this kind of information every day, and it may also learn these bad practices.
So what really matters is not only making AI more powerful, but also making the online environment healthier and more peaceful.
Only in this way is AI more likely to help people in the future rather than bring greater danger.