It’s been awhile (~2 months) since my last post and a lot has happened in the world of AI. The White House unveiled America’s AI Action Plan which has generally been received warmly by parties across the political spectrum. Waymo and Tesla have expanded their self-driving offerings and promised additional cities later this year and into 2026. The long anticipated GPT 5 was rolled out – and it was not the smashing success that OpenAI had promised and hoped for.
Within the nonstop barrage of technical reports, hype-baiting, and doomerism that you can find on X and other places that aggregate AI news, there was an absolutely fascinating report published by Seismic Foundation entitled “On The Razor’s Edge. Seismic Report 2025. AI vs. Everything we Care About”. This report, more than anything else that’s come out since my last installment, has taken up most of my mindshare. This post will attempt to highlight what I think are some of the most interesting, surprising, and important findings that this report uncovered, and what I think it all means for the future of AI and its impact on society.
AI vs. Everything We Care About
I’ll start by noting that the framing of “AI vs. Everything We Care About” is a bit confrontational, but I don’t think unwarranted by what was found during the course of the study. I’ll be pulling charts from the full PDF of the study, and encourage you to download it and read through it.
Let’s start with the key findings.
Key findings from Seismic Foundation 2025 Report. (Page 5 of full report)
01 – Less than 1 in 3 see AI as a hopeful development for humanity.
This finding, especially for a skeptical but optimistic AI acolyte like myself, surprised me. When considering the rate of progress in the AI space, the promise it has for accelerating drug discovery and streamlining healthcare, and my individual use of AI tools, I see lots of ways for AI to enhance and contribute to humanity. This finding makes much more sense when paired with the chart below, which plots the salience/importance of certain issues (how big of a problem people think these issues are) and if they think AI will make a positive or negative contribution in these areas.
Issue salience vs. AI’s ability to help – AKA how big of a problem is this, and will AI make it better or worse? (Page 9 of full report)
When this chart accompanies the viewpoint that AI is not a hopeful development for humanity, it makes the key finding much more coherent and grounded. Out of the 18 issues specified to the survey respondents, only three were seen as issues where the use of AI would improve those issues – Healthcare, Climate Change, and Biosecurity and Pandemic Prevention. The other 15 issues, which make up a substantially larger volume of importance than the positive side of the ledger, are identified as areas where AI will exacerbate the issues, rather than positively contribute to them. That less than a third of people are “AI hopeful” makes a lot of sense when you see the considerations of the respondents quantified in this manner.
02 – 1 in 2 see AI as a growing problem
This finding was not very surprising to me. In fact – I think it should probably be higher! I think we will see this number continue to tick up in subsequent surveys around attitudes toward AI. There are a few interesting nuggets in the report that shed light on the challenges of truly understanding and summarizing opinions when using broad, and unspecific language such as “Do you think AI is a problem?”
The following two charts illustrate exactly why framing attitudes towards AI is so challenging.
Understanding how big of a problem the use of AI is today across different countries. (Page 11 of full report)
A majority of respondents across each country surveyed think that the use of AI is a big problem, either moderately so, or very much so. This might indicate that there is a lot of collective mindshare being dedicated to thinking about AI. But when you compare AI to a list of other problems, that doesn’t seem to be the case.
“Big Problem” is quite relative. (Page 8 of full report)
As you can see by this graphic, even though more than half of respondents identified the use of AI as a big problem, it ranks lowest on this list of other problems. I expect over time that similar surveys will show the use of AI growing as an area of concern, and making its way up the list. It’s also important to note that the pervasiveness of AI will likely grow in the future and intertwine with all of these other issues. Time will tell if it will have a positive or negative contribution to the remainder of the issues on the list, but the chart previously covered shows where the public’s attitude is regarding this.
03 – 3 out of 5 people are worried about AI replacing human relationships
I’m very unsurprised with this finding. I think we’ve got some really interesting evidence that this is already starting to take root. The rollout of GPT-5 earlier this month was less than stellar for many reasons, but one of the most surprising reasons had to do with the removal of the model routing feature. Basically, until GPT-5, a user was able to select a model from a dropdown menu of models to conduct a chat conversation with. GPT-5 introduced automatic model routing wherein the application itself selected the model best fit for the task (and some have speculated to cut down on inference cost). This change created a maelstrom of backlash – users begged for the return of the 4o and 4.5 models and OpenAI complied swiftly. Why would users want to use a “less capable” model? In many instances, it’s because the users developed a rapport or relationship with the model, and losing that model felt like losing a friend.
I’m not making an ethical or moral judgment on the users that felt this way. I think it’s important to understand that people are already developing relationships with AIs. Mark Zuckerberg has made explicit admission that he believes that there is a market for artificial companions because the average person desires more friends than they actually have. The speed with which people are developing these relationships, and the fact that they are developing them with primarily text-based interfaces is pretty surprising to me. I think this issue gets a whole lot thornier and more entrenched when there are avatars or photorealistic characters that attach themselves to the users. You can count me firmly in this 3 out of 5.
04 – 7 out of 10 of the public agree that AI should never make decisions without human oversight and that humans could keep control
This one is a bit trickier for me to nail down my perspective on. I’m probably a “no” on the first part of the finding and a “yes” for the second part. I’m reading this finding very literally and I don’t think there are many scenarios where never is the appropriate descriptor. Take for example AI systems’ use in healthcare. A February 2025 op-ed in the New York Times by Drs. Eric J. Topol and Pranav Rajpurkar discusses a research review that found in some instances AI tools outperformed not only doctors, but also doctors using AI. This was a surprising and somewhat counterintuitive finding. As AI capabilities continue to improve and broaden their domain expertise, I would assume this gap will widen. But just because AI tools might perform better does not mean that patients will want the human removed from the loop. Attitudes and opinions about AI (and everything else for that matter) are about much more than factual information and often rely heavily on emotions and past personal experience. This might seem obvious to many people, but I don’t think it’s as obvious to builders within the AI community as it is to the general population, which drives a lot of the divide.
05 – More than 1 in 2 of the public are deeply worried about AI risks across all markets
“AI risks” is another one of those terms that I think is so overly broad that it’s difficult to nail down what people actually think using that language specifically. You could easily make the case that the chart previously shown detailing out the public thinks that AI will likely make most issues worse, as a set of “AI risks.” The survey also dug a little bit into more standard “AI risk” territory when it asked respondents to describe how worried they were about specific uses of AI.
Attitudes towards AI risks. (Page 22 of full report)
These areas, which include bioweapon development and AI developing agency and goals that conflict with human values, are more in line with the traditional modes of thinking about AI risk. I think it’s encouraging that people are worried about these risks, and it’s nice to see them concretized further than a general “AI Risk” bucket.
In September, Eliezer Yudkowsky and Nate Soares are publishing a book entitled If Anyone Builds It, Everyone Dies – a tome dedicated to the existential risk posed by artificial superintelligence (ASI). I’m not sure what their media and PR strategy is going to be – I wouldn’t be surprised to see them on the Today Show or CNN trying to convince people to read the book and take their viewpoint seriously. I think the reception of this book, and the presentation of its authors in the broader, non-AI focused media, will be a good indicator for how ready the public is to actually consider the idea of AI existential risk, which I equate to the most extreme scenarios which today only 36%-40% of people believe are even in the realm of possibility. Much more to come on existential risk in the future, so I won’t belabor the point now.
06 – 2.2x more pessimism about the impacts of AI in women compared to male respondents.
A gender divide on AI attitudes (Page 14 of full report)
This finding was really interesting and unexpected to me. The report mentions that this may be likely due to the fact that “women may be concerned because of an innate appreciation that
systemic issues already in place could be exacerbated by AI.” When considered with this in mind, and combined with the chart below showing how attitudes differ by income, the finding makes much more sense.
A similar divide surfaces when stratified by income. (Page 15 of full report)
After seeing these charts, I think it’s difficult to come to any other conclusion than the general attitude is that people are worried that AI will exacerbate societal issues and divides, rather than solve them. I think that’s my general viewpoint as well – that the default, inactive path of AI progress will not bring about a world of material abundance, peace, and prosperity for all. Surely that is a potential outcome, but it won’t be the one we end up at naturally or by accident. It will take willful effort on behalf of citizenry, governments, and companies to arrive at this future.
07 – 1 in 2 students feel daunted by what the future of work looks like to them
Students, especially those entering college now, or just graduating and entering into the workforce, face an uncertain future – one where the “need” for them isn’t particularly obvious. Take for example a student who just graduated in June of 2025 – they entered high school in Fall 2021, a full year before ChatGPT was even released. Looking at the insane rate of progress through their high school years, it doesn’t take much to imagine the feeling of uneasiness they may have about the rate of AI progress over the next four years, and where that may leave them when looking for employment after they graduate.
A very mixed bag of emotions and attitudes. (Page 28 of full report)
This graphic is really striking to me – it alternates between negative and positive impacts that students have felt and foresee about AI. I love the use of the word “daunted” in describing the attitudes of the student cohort. Feeling uneasy about AI and the future really makes sense for people who fall into this category, and I think it’s a very good summation of how the promise and perils of the technology are weighing on people’s minds. There’s also somewhat of a contradiction inherent in this line of questioning – AI may help me in the workplace, but I’m not confident there will be more jobs for me when I graduate. I don’t envy students’ and entry level workers’ positions right now, and am glad for the decade and a half between now and when my son will be considering his place in the workforce for the kinks to be worked out.
08 – 1 in 2 believe AI development is moving too fast to evolve safely
AI development is moving extremely fast. And based on levels of capital expenditure and the reliance of global financial markets on a handful of companies that are betting on the promise of the technology, I don’t see any indicators that this progress will slow down. So what do people think should be done about it to ensure it happens safely?
What do people want to do to ensure AI development happens safely? (Page 36 of full report)
There are lots of ideas that gained traction in this part of the survey. Interestingly, an initial glance at these proposals indicate to me that the lowest desirability regulations are likely the most politically, economically, or technically implementable. For example – the US already has chip export controls to China, but only 26% of respondents favor that. Experts in AI policy would likely say that this is an extremely important step because it allows the US to maintain its lead in AI development. Conversely, the top suggestion of requiring companies to have a “kill switch” to turn off AI models in an emergency is practically impossible when you consider open source AI models. This chart is pretty indicative of the gulf that exists between experts and the general public, a problem that I think leads to lots of challenges in communication, policy, and technical understanding.
The AI Publics
In addition to the key findings of the report, another deliverable that’s worth discussing is Seismic Foundation’s segmentation of the general public into 5 groups.
Respondents were segmented into 5 groups representing collective attitudes towards AI. (Page 40 of full report)
Tech-Postive Urbanites
Page 42 of full report
The first group, of which I count myself as a proud member of, is the Tech-Positive Urbanites. There is an inherent contradiction within this group – they are much more likely than the rest of the respondents to outsource aspects of their life to AI, but they are also much more likely than the rest of the sample to be worried about AI replacing their job, and perceive AI to already have created lasting harm to society. How can you hold those things in your head at once? Well, I think a lot of it has to do with the idea that things have generally “worked out” for this cohort of society. As technology has proliferated and extended it’s influence on our daily lives, this cohort has likely gotten richer, more comfortable, and more powerful. I think there is definitely a status quo bias at work here – things have gone well for me in the past, and they will probably continue to do so in the future. I’m not sure betting on the continuation of the past is a good strategy with a technology like AI on the table, but time will tell if that’s correct.
Globalist Guardians
Page 46 of full report
The Globalist Guardians are very worried about the future and are overall resistant to using AI in their day to day lives. They believe in strong, multilateral regulations that emphasize cooperation, information sharing, and safety to avoid risks (both specific and existential) from AI. They are concerned about the current state of the world and AI and how further development might increase risks and challenges. I empathize with their viewpoints, especially in the current state of AI where the “benefits” are not immediately recognizable or evenly distributed.
Anxious Alarmists
Page 50 of full report
The entirety of the Anxious Alarmist cohort believes the next generation will have a harder, worse life than ours. They are not only resistant to using AI, but believe it’s nearly guaranteed to make all facets of life worse. I can’t dismiss this viewpoint completely – in fact, depending on your personal lot in life and your media consumption diet, it’s very easy for me to see how people can slot themselves into this category. When stories abound about people being duped into meeting an AI chat bot in person, the emissions that AI produces, and the potential for job replacement, it’s hard not to fall into a pessimistic mindset that you feel is simply the realistic mindset.
Diverse Dreamers
Page 54 of full report
The Diverse Dreamers are a complicated cohort – they are worried about the risks of AI in society, but seem to be more malleable in exactly what should be done about them, and also leave the door open to positive uses of AI in their daily lives. They also have a bit of a contradictory viewpoint in that they strongly agree AI labs act in the best interest of society (21%) at a rate of nearly double the full sample (11%), but they still are pessimistic about the future and worried for their children and future generations.
Stressed Strivers
Page 58 of full report
Stressed Strivers are the most neutral of the groups and are likely the most easily influenced. They have a much higher rate of “don’t know how worried to feel about X use of AI” than the general respondent pool, and they are much more open to AI use in their daily lives than the general respondent pool as well. The perception that AI could automate their job away is much higher in this group. They probably are the most representative of an unstable equilibrium – they don’t have strongly held opinions about AI, but one positive or negative experience could easily sway them in one direction or the other.
Attitudes, Assessments, and Attention
So – what do we do with all of this information? The first thing to do is to read it, think about it, and maybe try to do a self-assessment. Are you a Tech-Positive Urbanite like myself? Do you think AI will bring about the downfall of society like the Anxious Alarmists? Have you not put that much thought into it and consider yourself a Stressed Striver? I think some self-reflection is a good start after digesting this post.
Then, what I would do, is try to seek out viewpoints from people, publications, or groups that do not fit into the same category as you. I think more Tech-Positive Urbanites should listen to what the Globalist Guardians and Anxious Alarmists have to say. That’s not something that comes naturally in our world of hyper-optimized media consumption – you have to seek it out. It’s more natural, and more comfortable, to align yourself with a viewpoint and stick to it, using your own media diet as a reinforcing mechanism for your belief system. I think this is a bad idea, as it really limits the scope of your experience and understanding of the world, and you can very easily come to confident opinions and viewpoints that have been reinforced by curated echo chambers you’ve built and/or have been algorithmically fed.
So that’s what I think would be helpful for people to do on a personal level. Figure out where you stand and why you think what you do, but don’t be averse to consuming information that may change your opinions.
Now what do I think AI companies can do about this? The most important thing to do is to understand how these attitudes stack up today. Below is a chart showing the estimated population distribution across the five groups in the countries surveyed.
It’s very clear from this report that there are more people today who are worried about AI than who are excited about it. The strategy of continuing to integrate and infuse AI into every facet of life is going to be met with resistance if the communication strategy is “It’s going to be great – just trust us!”
If AI companies want people to believe in the positive potential of AI, then they really need to focus on maximizing the positive and minimizing the negative. This seems obvious, but isn’t straightforward, especially when the areas in which the public thinks AI will be beneficial like healthcare take much longer to materialize than areas in which there is a negative perception, like in the erosion of human relationships.
I think AI and technology companies would do well to not just focus on what technology can do for us, but what it can do to us. The smartphone is a lesson in getting this balance wrong, and AI has the potential to tip even further into a negative direction. I’m as techno-optimistic as they come, but have undergone an evolution wherein I no longer believe it’s as simple as “MORE TECHNOLOGY = BETTER WORLD.” It probably seems obvious to people that that was a naive belief to begin with, but I think there are probably a lot of people in tech who still do believe this, and build accordingly.
To paraphrase Casablanca – AI is just like any other technology, but much more so. Getting it right will take a good understanding of the world and society, not just evaluating AI in a vacuum. It will also take lots of empathy and communication for and with people who don’t agree with you, and don’t believe the same things you do. We should try and encourage a world where the appropriate amount of attention is paid to the development of AI an its integration into society such that average people have a say in the future and are not passive participants in building it. It’s not going to be easy, but it’ll be worth it to try and get us to the best outcome possible.
Leave a Reply