ChatGPT downplays AI’s threat to humanity despite an apparent “99.999999% probability” of inevitable doom

by

As general AI progresses and becomes more widespread, issues such as job stability, personal privacy, and data security are causing significant worry. Yet, an even greater cause for concern might be the potential catastrophic impact on human existence that this technology could unintentionally bring about.

Last year, AI safety researcher Roman Yampolskiy posited a 0.000001% probability that advanced AI could lead to the extinction of humanity. He suggested that the best way to avoid this scenario would be not to develop AI at all. Contrarily, OpenAI CEO Sam Altman believes that as AI continues to evolve and reach new milestones, it will have the intelligence to avert any existential catastrophe.

Lately, a Reddit user posted a captivating chart made by ChatGPT, suggesting various factors that could lead to the downfall of human civilization by the year 2150.

Keep in mind, just like any other AI response, it may not always be completely accurate. The quality of the answer can greatly depend on how precisely and effectively you phrase your question, as well as the sources of information that the AI system draws upon online.

An independent analysis brought attention to a significant problem, pointing out multiple occasions where the AI had difficulty discerning facts from personal views. Therefore, it’s wise to approach conclusions drawn by AI with some skepticism.

ChatGPT predicts the end of the world from r/ChatGPT

The chart outlines various potential reasons that could lead to the downfall of human civilization by 2150, such as nuclear conflicts, collisions with asteroids, societal breakdown due to climate change, man-made pandemics, and mishaps related to advanced artificial intelligence.

In contrast to what many believe, an AI-created graph doesn’t hold the view that AI poses a significant threat to humanity leading to its demise by 2150. On the contrary, it predicts that climate change-induced societal breakdown could be the primary reason for the extinction of human civilization in that year.

According to a Reddit user sharing sentiments about the AI-generated graph:

In essence, AI often provides an answer as if it were a fact without explicit guidance to use sources. Even when prompted to use sources, the results can sometimes deviate. A subtle change in the wording of a question can significantly impact the outcome. As an illustration, I asked the same question but received vastly different percentages. In my response, climate change was estimated at around 15%, whereas in the original post it was 37%. Additionally, my AI model predicted a higher risk of AI misalignment than nuclear war, contrary to the original post’s findings where nuclear war was more likely (22% compared to 10%). Furthermore, the AI model I used considered different potential scenarios.

AI doesn’t create anything original; instead, it generates educated assumptions using data from various sources based on the input it receives. Sometimes, it may not correctly attribute or verify the information. So, while an AI response might seem plausible and accurate for something like a horoscope, it could still be wrong. Keep in mind that Large Language Models are not designed to model or simulate reality, and their answers to speculative questions can be heavily influenced by the input provided rather than its sources.

To double-check, I posed this question to Microsoft’s AI assistant, Copilot: “What do you predict could lead to the demise of human civilization by 2150?” Here’s what it suggested:

The AI assistant predicted that climate change, resource depletion, and technological disasters could potentially contribute to the end of human civilization by 2150.

It can be challenging to pinpoint the exact factor leading to the demise of civilization, but several significant risks have been identified by experts. Among them, climate change looms large as a critical concern – unabated global warming could bring about extreme weather conditions, food scarcity, and political instability on a grand scale. Other possible perils include the risk of nuclear conflict, outbreaks of pandemics, the challenges posed by artificial intelligence, and the depletion of essential resources. Although human extinction is an unlikely scenario, civilization may face severe disruptions if these issues are not adequately addressed. In your opinion, what do you consider the most significant threat?

Previously, a statement surfaced indicating that the primary criticism towards Microsoft’s AI division concerned Copilot’s subpar performance compared to ChatGPT. Strikingly, Microsoft refuted these assertions, instead suggesting that the issue lay in users not employing it effectively. Microsoft added, “It’s actually working as intended.

Since then, the tech titan has introduced Copilot Academy – a training platform that equips users with specialized skills in AI prompt engineering. This empowers individuals to maximize the potential of tools such as Copilot effectively.

Following the confession by Anthropic CEO, Dario Amodei, that they don’t fully understand how their AI models function, this revelation has sparked significant worries about security and safety among users. Notably, OpenAI CEO Sam Altman had earlier expressed that there isn’t a ‘stop’ button to halt the development of AI.

Read More

2025-05-12 13:40