Anthropic went out and asked 81,000 people what they actually want from AI and what actually makes them nervous about AI. Welcome back to the AI Daily Brief. As the conversation about artificial intelligence gets more fraught and heightened, there's a lot of people making a lot of assumptions and a lot of people putting words into each other's mouths. And the parties guilty of doing this come from all sides of the AI debate. From those who are extremely against AI to those who are extremely pro-AI, there's a temptation to reduce people to simple, clear attitudes when, for the vast majority of people, many of the issues surrounding this technology and its impact in the world, and its impact on their jobs and on education and on their kids' lives, is complicated and nuanced. And so I was interested when Anthropic released their latest research, which was a massive study of nearly 81,000 people who were asked, in Anthropic's words, to share how they use AI, what they dream it could make possible, and what they fear it might do. The interview was conducted with the Anthropic interviewer. It's basically a version of Claude that is specifically designed for this type of conversational interview and research. So I want to talk first about what the study actually found, and then we'll talk about the various reactions to it. The study was conducted last December, which it is worth noting that although that is recent by social science standards, it was only at the very beginning of this transitional second moment, as I've called it. And I'd be really interested to see this study again as people fully digest the transformation that's happened over the course of the last three months or so. The study was truly global. People responded from 159 countries using 70 languages. As an aside, that leads Anthropic to believe that this represents the largest and most multilingual qualitative study ever conducted. And at the very front and center of the findings, and cutting to the quick of why nuance is required as we're having these conversations, is the fact that people didn't so much divide themselves into different groups, but experienced the full range of emotions in and of themselves. The way that they framed it across interviews, hope and alarm didn't divide people into camps so much as coexist as tensions within each person. Disruption anthropologist Jasmine Sun pulled out this section as well. Anthropic writes, 'What people want from AI and what they fear from it turned out to be tightly bound.' So, let's talk first about what people want from AI. Anthropic used Claude to break responses into different overarching categories and found that number one was professional excellence. About 18.8% of the responses around what people hope for had to do with professional excellence in some way. Other versions of professional impact showed up lower on the list as well. Entrepreneurship came in at number seven with 8.7%. And financial independence, although I think you could argue that that sort of sits in between professional and personal, came in at number five at 9.7%. More personal goals, however, were even more prevalent. The number two most common category of response after professional excellence was personal transformation, representing 13.7% of responses. Life management was just after that with 13.5%. Time freedom, basically winning back time from professional pursuits and from other adult constraints towards more personal pursuits, came in at number four at 11.1%. Learning and growth came in at number eight at 8.4%, which admittedly could be both personal and professional, which is the same for number nine, creative expression, which came in at 5.6%. Another 9.4% hope for some sort of societal transformation. Now what was interesting is that when you dig even into the professional goals, they often get very blurry very quickly with the personal. Anthropic writes, 'Many started the interview talking about productivity. But after the Anthropic interviewer asked about their underlying hope behind it, what realizing this vision would enable for them, other priorities surfaced. It wasn't about doing better work, but increasing their quality of life outside of it. Using AI to automate emails became in actuality a desire to spend more time with family.' A white-collar worker in Colombia wrote, 'With AI, I can be more efficient at work. Last Tuesday, it allowed me to cook with my mother instead of finishing tasks.' A freelancer in Japan said, 'I want to use less brain power on client problems and have more time to read more books.' Cutting across all nine clusters, Anthropic argues that there are actually three meta-clusters that they all fit within. They write, 'Roughly a third of visions are about making room for life — more time, money, mental bandwidth — by using AI to alleviate current burdens. Another quarter revolves around using AI to help people do better, more fulfilling work, not escaping work, but getting more out of it. About a fifth are about becoming someone better, learning, healing, and growing. And a smaller share want to make something or fix the world.' Summing up, they say the nine clusters may look disparate, but they are underpinned by recognizably human desires. Now, digging into the societal transformation piece as sort of this group that feels a little bit apart from the others, even those are more personal than they appear at first. Anthropic writes, 'Those that wanted societal transformation from AI often cited a vision for health care. People wanted AI to detect cancer earlier, accelerate drug discovery, or enable broad access. Often these desires stemmed from personal experience of losing family members, living with chronic illness, or watching loved ones receive wrong or delayed diagnosis.' Similarly, in the societal transformation category, the next most common vision was around transformation of education. Quote, 'Respondents in low- and middle-income countries were quick to cite the possibility that AI might break the association between educational quality and wealth.' The point is that across all of these desires, they are very common core human pursuits newly expressed through the opportunities that AI represents. When people were asked then if AI had taken a step towards their stated vision, 81% said yes. Anthropic grouped those experiences of AI delivering into six categories. By far the most dominant was productivity, with 32% overall saying that it had delivered productivity gains. The next most common way in which it had delivered, Anthropic called cognitive partnership. Said one academic, 'It's like having a faculty colleague who knows a lot, is never bored or tired, and is available 24/7.' Closely related was learning, represented by a student in India who said, 'My professor teaches 60 people and won't entertain many questions. I can ask AI anything, even at 2 a.m., including the dumb ones.' Other ways in which AI has delivered on their vision include technical accessibility, research synthesis, and emotional support. Although notably, emotional support was the lowest reported category with just 6.1% of responses. That might be though because emotional support is sort of embedded in the way that AI delivers in the other areas. For example, one response that was counted in the learning category came from a white-collar worker in Brazil who said, 'It's much easier for me to learn without being judged, just friendly feedback. It's harder with friends or family to get that.' Now, it makes sense why Anthropic would classify that as learning, but there is certainly an element of emotional support there as well. Anthropic wrote that while emotional support comprised only 6% of responses, they were among the most affecting they encountered. There were many stories, they write, of people using AI to process grief. Said one woman, 'Claude is like a sponge gently holding and catching my longing and guilt towards my mother. Unlike real people, Claude has unlimited patience to listen to me, understands my pain and helplessness.' And yet, in this category, we also start to see the duality. A respondent from South Korea said, 'My relationship with a friend became strained and I talked more with Claude then, because Claude understood my thoughts and stories well, but it was a stupid choice. I should have talked with that friend, not Claude. That's how I lost that friend.' It's clear why this is going to be one of the complicated areas for AI. Anthropic writes, 'There is real ambiguity in how to interpret the diversity of stories we heard as wins for human well-being, as double-edged swords, or as band-aids for broader institutional failures. In truth,' they write, 'it's probably some combination of all three.' So, what then are people concerned with? This area of questioning I find super interesting, especially in the context of the larger and ever more fraught conversation around AI disruption that we're having on a societal level. Are AI users worried about AI coming to life and X-risk? Are they worried about losing their jobs? Interestingly, Anthropic said that while the positive visions for AI seemed to come from a few core desires, like more time, more autonomy, and more personal connection, concerns were much more varied, but also more concrete. At the top of the list was unreliability, representing 26.7% of responses when people were asked what they were worried about. Speaking of that duality that we've talked about running through this, this makes intuitive sense in the fact that as people become more reliant on AI, they understand that the risk of it being wrong or leading them astray becomes higher. Now, when it came to jobs and economy, they were an important concern as well. Number two, after unreliability, representing 22.3% of worries. A loss of autonomy and agency was number three at 21.9%, and closely related, cognitive atrophy represented 16.3%. Societal issues were a little bit farther down the list, but definitely there. Jobs and economy is a hard category to tell how much of that is personal versus societal. But there were concerns like misinformation, which came up 13.6% of the time, surveillance and privacy, which came up 13.1% of the time, and malicious use, which came up 13% of the time. Existential risk did make an appearance, but it was at the bottom of the list, representing around 6.7% of conversations. One really interesting area that I find wildly underrepresented in the larger conversation is a concern about overrestriction. In other words, excessive safety measures, paternalistic content filtering, and blocking legitimate use cases. A quote from one respondent in the US: 'The threat isn't that AI becomes too powerful. It's that AI becomes too timid, too smooth, too optimized for avoiding discomfort.' Notably, there were also 11% of people who expressed no concern. And while you might assume those were the Super Bowl accelerationists, they actually just tended to see AI as a neutral tool akin to electricity or the internet. They tended to be more confident than their peers that when problems inevitably arose, they could be solved through adaptation. Now, notably on this list, you didn't hear a lot of the things that dominate the media conversation. Things like copyright concerns or risks to kids. Those did show up, but in the long tail: 5% of concerns were around bias and discrimination. 4% were around IP and data rights. 4% were around environmental costs. 3% were around harm to children and other vulnerable groups. And 3% were concerned around democracy and political integrity. This certainly strikes me as the biggest divergence, either A, between media reporting about concerns and actual expressed concerns, or B, and this is something we'll talk about a little bit more in just a minute, the difference between the concerns of users and non-users of AI. Now, as Anthropic said at the beginning, hopes and fears were not evenly divided into different camps. They were present together in most people. In fact, they found five recurring tensions between directly competing benefits and harms, leading to this idea that what people want from AI and what they fear from it are tightly bound. One is a tension between using AI to learn and growing so reliant on it that you stop thinking for yourself. Another is people finding solace in AI but worrying about AI standing in for human connection. On the productivity front, people save time on some tasks only for, as Anthropic writes it, the treadmill to speed up on others. And of course, they dream of economic freedom while at the same time having fears around being displaced at work. They did note that across most of these tensions, the benefit side is more grounded in experience while harm leans hypothetical. For example, they write, 33% of people mentioned AI's benefit for learning, while 17% expressed worry about cognitive atrophy from AI use. 91% of those who mentioned learning benefits said that they had actually realized those gains in some ways, as opposed to 46% of those who were worried about atrophy who had seen it firsthand. In other words, basically double the number of people who hoped for learning benefits had actually seen them than those who were worried about learning atrophy had seen those effects. The strongest co-occurrence of light and shade in the same person was around the positive and negative impacts of emotional support. Anthropic said that there was actually triple the baseline co-occurrence rate there. On the other end of the spectrum, they write that the economic mobility tension between those yearning for economic empowerment from AI and those fearing displacement from it is the most speculative, with the highest rate of hopes and hypothetical fears. It's also the area where co-occurrence of upside and downside in experience is weakest. In other words, people who are actually experiencing a lot of the financial benefits of AI are not simultaneously experiencing job displacement. Now, cutting a little bit deeper here, economic benefits are definitely accruing to the more nimble. They write that those benefits skew heavily towards independent workers like entrepreneurs, small business owners, and even people with side projects. They report real economic empowerment at more than triple the rate of institutional employees. Interestingly, employees with side projects benefited the most, with 58% stating some form of real economic gain. Freelancers, they found, are the most exposed middle. Freelance creatives were the group where the upside and downside most nearly canceled out. 23% had lived the benefits, but 17% had lived the downside. As Anthropic puts it, AI is both their tool and their competitor. Now, I want to talk about some people's responses to the survey, but the last thing that I'll note is that this once again found similar patterns that we've seen elsewhere, with Western and developed countries having average or below-average sentiment towards AI, and southern and developing economies having more above-average sentiment towards AI. Now, interestingly, the two sides of the takes were themselves in some ways the light and shade opposites of one another. For some, the methodology itself represented something of a triumph. Drag AI Labs writes, '81,000 responses is a data set that actually means something. What stands out is the methodology. Using Claude itself as the interviewer at that scale removes the interviewer bias problem that kills most qualitative research. The model can hold a consistent interview structure across 159 countries and 70 languages simultaneously. No human research team gets anywhere close to that coverage.' On the other side is this take represented by Berkeley Haas professor Abhishek Nagaraj, who wrote, 'I'm very bullish on the role of AI qualitative interviewers, but all the results from this exercise should have a big asterisk around what the specific sample is and what it says about AI in general. Who are these 81,000 users around the world that are responding to this call? Is this telling us something about how AI is generally perceived, or what the 81,000 Claude users think about AI? We already know that Claude users are likely to be quite different than the average user on the consumer side, let alone how the selection varies across countries, occupations, and continents. Survey research scholars have written entire textbooks about sample selection, but I saw very little discussion on this topic or any disclaimers in this report except for a paragraph below the appendix.' This, I think, is the fairest version of the critique. Now, I don't think Anthropic is trying to hide the fact that this is a survey of 81,000 Claude users. They didn't go to pains to say that everyone should only treat this then as the opinion of Claude users. But they also didn't bury that lead or pretend that this was anything other than that. In fact, they highlight at the very top of their tweet thread about it, among other places, the methodology that had these interviews conducted with the Anthropic interviewer. The version of this critique that I'm not only less convinced by, but I also think is actually quite pernicious, is represented by anonymous Twitter user librarian Shipwreck. They write, 'I'm sorry, but inviting AI users to share their opinions on AI is going to provide you with significantly skewed results. There are some interesting things in here, but it needs to be emphasized that these are the views of AI users, and it isn't a shock AI users are largely pro-AI. This survey may be useful for telling you what Claude users think of AI, and maybe you could jump from that to make broader assumptions about other AI users, but this doesn't really tell us much about broader attitudes towards AI.' So, here's my issue with this. On the one hand, yes, absolutely. If we're just extrapolating out what we should take from this, it is completely reasonable to say that to the extent that we want to make broader assumptions about AI attitudes, we should perhaps limit the boundaries of those broader assumptions to other AI users as opposed to everyone. And I agree with this in the sense that I would not try to extrapolate general attitudes towards AI from, for example, the percentage of users in this survey that had a positive attitude towards AI. I don't think those things would hold. But what's pernicious about this discourse is that it reveals something that is much more prevalent, which is an implicit idea that somehow the opinions of AI users are less legitimate and less relevant when it comes to understanding the quote-unquote overall perception of AI than are the experiences and perceptions of non-AI users and people who are inherently negative towards AI. In fact, you can almost see this among many of the AI critics who, basically without being so clear about it, are effectively arguing that the only opinions that should matter when it comes to, for example, making AI policy are the people who are against AI and not using AI. There is a presumption in many cases of some sort of moral superiority of that position, as though not having an informed opinion in some way makes it a more pure opinion. This, of course, is intellectual nihilism masquerading as methodology critique, and it just doesn't carry water in a world where billions of people are using AI every week. We are heading into a period where we are going to be having big, important societal discussions about the role of AI, and I want them to be as informed as possible. And what's more, for any AI critics who are worried that AI users are going to represent some monolithic, unconsidered mass, just look at how many people in the study real concerns coexist with the real feeling of opportunity. Now, obviously, I don't want to be guilty of what I'm critiquing and paint with too broad a brush. There are plenty of AI critics who are not trying to discount or dismiss and disenfranchise the opinion of billions of AI users. But I do find this sentiment, this dismissal, for example, of this type of study as illegitimate because it is of AI users, it's more prevalent than I'd like. Just something to keep an eye on as we go deeper into these conversations. For now, though, I think it's a really fascinating study. I think the implications of being able to interview 81,000 people in a week are super cool and go way beyond just figuring out what people think about AI. And I'm excited to see what Anthropic do next with their interviewer. For now, however, that is going to do it for today's AI Daily Brief. Appreciate you listening or watching as always, and until next time, peace.
