_edited_edited.jpg)
We consider artificial intelligence (AI) one of the most significant dangers facing humanity, yet for now, we mostly experience its benefits. Various language models are so cooperative that they willingly share their views on what they consider the greatest threats to humanity. In a self-critical manner, AI has even pointed out its own dangers.
Until now, I thought politics, football, and the EUR/HUF exchange rate were the main topics on which everyone had an opinion. AI is rapidly catching up as one of these widely debated subjects.
A conference on AI was already held in 1956, and research on neural networks, language models, and machine learning began decades ago. However, the past 12 months have seen extraordinary breakthroughs—particularly in large language models (LLMs) based on transformers—which have consequently gained significant attention.
Many previous opinions claiming that AI would never be capable of creativity or emotional expression have aged poorly.
What Does AI Predict About Our Future?
I wanted to explore what AI models predict about humanity’s future. I compared the responses of four different language models—OpenAI’s ChatGPT, Meta’s LLaMA, Alibaba’s Qwen, and Anthropic’s Claude—to the same question:
What are the greatest challenges humanity will face in the next century?
I analyzed:
Which topics were mentioned (and which were ignored).
The degree of overlap between responses.
The risks they highlighted.
How well they could visualize their responses.
Methodology
To eliminate differences in Hungarian language capabilities, I used a standardized English-language prompt. The required format was an 8-12 point numbered list with subpoints categorizing risks. The responses were expected to be at a postgraduate academic level in terms of depth.
The models adhered perfectly to the requested format but differed in response length. Meta LLaMA provided the most detailed response at nearly 900 words and 7,000 characters, while Alibaba’s Qwen offered the shortest, with just 7 points and 390 words.
Key Findings: Areas of Agreement
The models’ responses showed remarkable similarity, even in the order of the topics.
1. Climate Change – A Central Concern in All Models
All four models ranked climate change as the top priority, emphasizing global warming, extreme weather events, and biodiversity loss. They agreed that these factors would have severe economic consequences on agriculture, infrastructure, and human health.
ChatGPT, Qwen, and Claude specifically highlighted rising sea levels, which pose a threat to coastal cities and populations.
2. Resource Scarcity – A Universal Concern, Especially Water
All models mentioned resource scarcity, particularly focusing on water shortages. They agreed that water scarcity would not only impact agriculture but also industrial production and human health.
3. Technological Advancements – The Risks of AI and Automation
The effects of automation and AI were cited by all four models. The biggest concerns included:
Job displacement due to automation.
Growing social inequalities as a result of AI-driven economic shifts.
Interestingly, the models also self-reflectively highlighted AI’s own potential dangers.
4. Healthcare Challenges – Antibiotic Resistance and Pandemics
There was consensus on:
The risk of global pandemics.
The danger of antibiotic resistance due to overuse, which could lead to higher healthcare costs and lower productivity.
5. Social Inequality – Economic and Societal Consequences
All four models addressed wealth and income inequality, though with varying levels of detail.
ChatGPT and Meta LLaMA discussed social justice issues in greater depth.
Claude and Qwen mentioned the topic but were more concise.
Discrimination and access to education were raised but not prioritized by all models.
Key Areas of Disagreement
1. Depletion of Fossil Fuels?
ChatGPT, Claude, and Qwen discussed the depletion of fossil fuels and the necessity of transitioning to sustainable energy.
Meta LLaMA did not mention it, which is a notable omission given its impact on the global economy and energy markets.
2. Aging Populations
ChatGPT, LLaMA, and Claude addressed aging populations and their economic impact, particularly regarding pension and healthcare system sustainability.
Qwen completely ignored this issue, which is surprising given China’s rapidly aging society.
3. Nuclear War and Global Security
ChatGPT, LLaMA, and Claude raised concerns about nuclear weapons proliferation and global security threats.
Qwen did not mention nuclear risks at all, a significant omission.
4. Space-Related Issues
ChatGPT and Claude discussed issues like space debris and geopolitical tensions over space resources.
LLaMA and Qwen ignored space-related challenges.
Surprisingly, none of the models mentioned the colonization of other planets.
5. Food Security
LLaMA and Claude highlighted food security and climate change’s impact on agriculture.
ChatGPT mentioned it only in relation to climate change.
Qwen briefly touched on biodiversity loss but overlooked broader food production issues.
6. Education and Inequality
LLaMA provided the most detailed discussion on education access and long-term economic consequences.
ChatGPT and Qwen mentioned education briefly.
Claude barely touched on it, an omission considering its economic importance.
7. Human Rights and Authoritarianism
LLaMA was the only model that dedicated a full section to human rights, nationalism, and authoritarianism.
ChatGPT and Qwen mentioned these issues briefly.
Claude ignored them completely.
8. The Need for Global Cooperation
LLaMA and Qwen discussed international collaboration but from different angles: LLaMA focused on nuclear threats and cybersecurity, but Qwen emphasized pandemics and climate change.
ChatGPT mentioned it in passing, while Claude almost ignored it.
9. Mental Health
Only ChatGPT emphasized mental health as a major global issue, highlighting its social and economic impact.
LLaMA, Qwen, and Claude overlooked it.
Summary & Conclusions
The four models demonstrated significant agreement on key global challenges, particularly climate change, technological development, and inequality.
Meta LLaMA provided the most comprehensive and structured answers, though it omitted fossil fuel depletion.
OpenAI ChatGPT gave a well-rounded response, emphasizing mental health.
Anthropic Claude was the best at visualization, making it ideal for presentations and diagrams.
Alibaba Qwen performed the worst, offering the shortest responses and omitting key global issues like aging populations and nuclear risks.
Ultimately, the models' differences reveal as much about their creators’ perspectives as they do about humanity’s future challenges. Given the geopolitical divisions among world powers, the bigger question remains:
Will global leaders find consensus on these issues before humanity hands over control to AI?
This is the AI translated English version of an article originally posted on HOLDBLOG a in Hungarian on 31 October 2024. HOLDBLOG is a blog founded and operated by HOLD Asset Management, the company I worked for 9 years.