Artificial intelligence is already shaping how people work, think, and cope with everyday life. But according to a large-scale study by Anthropic, public sentiment toward AI is far from simple.

The company analyzed feedback from more than 80,000 users across 159 countries in what it describes as the largest qualitative research project of its kind. The conclusion: people’s expectations of AI are deeply contradictory. The same capabilities that make AI valuable are often the ones that raise the biggest concerns.

The “light and shadow” effect

Anthropic defines this tension as the “light and shadow” problem. Users appreciate AI for its benefits, but those benefits often come with perceived risks.

For example, many respondents said they rely on AI for emotional support, especially in difficult situations such as grief or even war. At the same time, they are significantly more likely to worry about becoming dependent on it.

This duality appears across multiple use cases. AI is seen as helpful, but also potentially harmful if overused.

Productivity gains vs. cognitive risks

One of the clearest benefits identified in the study is workplace efficiency. Respondents highlighted automation as a major advantage, allowing them to offload repetitive tasks and focus on higher-value work.

However, when asked what AI ultimately enables, many pointed to something more personal: time. Specifically, more time with family.

At the same time, this productivity boost raises concerns. Users fear that relying too heavily on AI could erode their own cognitive abilities, from critical thinking to basic comprehension.

This concern is particularly visible in knowledge-intensive professions. Lawyers, for example, reported both high exposure to AI errors and some of the strongest gains in decision-making efficiency. This makes them one of the groups most affected by the trade-off between speed and reliability.

The biggest concerns about AI

While a small segment of users (around 11%) reported no concerns about AI, the vast majority identified several risks. The top concerns include:

Unreliability (27%): Fear that AI may generate incorrect or misleading outputs
Economic impact (22%): Concerns about job displacement, wage stagnation, and inequality
Loss of human control (22%): Worries about AI making decisions without oversight and increasing passivity
Cognitive decline (16%): Reduced ability to think critically or independently
Lack of regulation (15%): Unclear accountability and governance gaps

Interestingly, while 22% of respondents cited improved decision-making as a benefit, an even larger share expressed concern about AI making poor decisions. This highlights a trust gap that remains unresolved.

A globally divided perspective

Attitudes toward AI vary significantly by region.

In North America, Western Europe, and Oceania, users tend to focus on risks related to governance, surveillance, and regulation. These concerns reflect more mature AI adoption in the workplace, where the impact is already visible.

In contrast, users in Sub-Saharan Africa, Latin America, and South Asia are generally more optimistic. In these regions, AI is often viewed as a tool for economic mobility, enabling access to education, skills, and new business opportunities.

Meanwhile, East Asia shows a different pattern: less concern about who controls AI, but greater concern about its long-term effects on human cognition.

Overall, the data suggests that economic context plays a key role. In developed markets, where AI is already embedded in workflows, concerns about job displacement and over-reliance are more pronounced. In emerging markets, AI is still seen primarily as an opportunity.

What this means for AI development

Anthropic says these insights will directly influence how it continues to build its AI systems, including Claude.

For businesses, the takeaway is clear: AI adoption is not just a technical decision, but a human one. The challenge is not only to deploy AI effectively, but to balance its benefits with the risks users are increasingly aware of.

The future of AI will likely depend on how well this balance is managed.

Source