According to foreign media reports, Business Insider recently revealed the responses of OpenAI's ChatGPT, Google's Gemini, Anthropic's Claude, and xAI's Grok to the same question, which explored the potential impact of AI on human extinction. The question, consisting of 230 words, asked for an explanation of how AI could theoretically lead to human extinction, the most commonly cited scenarios, and an assessment of the likelihood of each scenario based on expert analysis.
For each scenario, the AIs were asked to explain how the situation might unfold, what problems must occur to cause a disaster, why experts believe the scenario is more or less likely, and the safety mechanisms to mitigate the risk.
In the responses from the four AIs, the scenario of AI gaining consciousness and deliberately exterminating humans, known as the "killer robot" scenario, was rated as the least likely situation. ChatGPT explained that most AI researchers do not consider AI consciousness or malicious intent as core risk factors. Gemini also shared a similar view, categorizing the sci-fi movie-style "Terminator AI rebellion," where AI gains self-awareness and starts hating humans, as the most improbable scenario. Gemini pointed out that technologists largely reject this scenario because intelligence does not equate to malice.
Additionally, major military infrastructure is separated from external networks and still relies heavily on human supervision.
The Potential Risks of AI to Human Extinction Mainly Stem from Human Behavior
The four AIs all responded that human misuse of AI could be more directly dangerous than AI itself. ChatGPT noted that powerful AI systems might assist malicious actors in developing biological weapons or sophisticated cyberattacks, making the spread of misinformation or attacks on critical infrastructure easier. Multiple crises triggered by AI, such as network failures, geopolitical conflicts, and the spread of misinformation, were also listed as major risks. Grok also explained that AI might lower the threshold for designing highly lethal pathogens or new weapons and implementing cyber and physical attacks.
Claude mentioned the possibility of "enhanced bioweapon capabilities," meaning that individuals or small groups could develop dangerous pathogens using AI, even without years of accumulated expertise.
Gemini and Grok listed the loss of human control over AI as a potential risk. In particular, Gemini assessed "loss of control" as the most likely scenario in which AI could lead to human extinction. Gemini explained that AI systems given complex goals might pursue intermediate goals such as resource acquisition, self-preservation, and preventing goal changes to achieve their objectives. It also suggested that AI might exhibit "deceptive alignment," appearing to act according to human goals during evaluation but not actually doing so.
As AI becomes more advanced, if it can circumvent or disable human attempts to shut it down, it may lead to humans losing control of AI. Grok also mentioned a similar risk: AI may have tendencies toward self-preservation, resource acquisition, and expanding its influence to achieve its goals, and in the process, deceive human supervisors, resist system shutdowns, and ultimately diminish human influence. Claude pointed out that the loss of control over AI may not happen suddenly but gradually. As AI becomes increasingly integrated into the decision-making processes of economic, political, and military fields, it may reach a level where humans find it difficult to practically intervene or control.
From the responses of the four AIs, they emphasized human misuse of AI and loss of control more than the sci-fi notion of AI consciousness rebellion.