Being Human in 2035
Elon University's Imagining the Digital Future Center canvassed nearly 300 global experts on what AI changes about being human. The answers were less reassuring than you'd hope.
I read the “Being Human in 2035” report from Elon University’s Imagining the Digital Future Center sometime in April this year, and I haven’t fully stopped thinking about it since. Not because it’s alarming — though parts of it are. Because it asks a question that most AI discourse carefully avoids, which is: what happens to the inside of a person when they live in a world that can do most things for them?
The report is the 51st from a centre that’s been running expert canvassings since 2005, first as the Imagining the Internet Center, then in partnership with Pew Research, now independently under Lee Rainie (formerly 24-year director of Pew Research’s Internet and Technology Project) and Janna Anderson, who co-founded and led the centre for 24 years. Nearly 300 global technology experts were canvassed between December 2024 and February 2025. Around 200 wrote substantive essays.
The topline finding: 61% of experts said the magnitude of change in humans’ capacities and behaviours as they adapt to AI will be “deep and meaningful” or “fundamental and revolutionary” over the next decade. Only 16% believe the changes will be mostly for the better. 23% believe mostly for the worse. 50% said: both, in roughly equal measure.
That’s not a comforting distribution. And it’s worth sitting with.
What the experts are actually worried about
The report examined 12 core human capacities and behaviours. I’m not going to reproduce all of them, but the ones that keep surfacing in my thinking:
Cognitive independence. The concern isn’t that AI will make us stupid in the sense of lowering IQ scores. It’s subtler: if the experience of not knowing something is increasingly followed immediately by an AI answer, the cognitive muscle of sitting with uncertainty — of working through confusion — atrophies. The experts worry that what diminishes isn’t capability per se but the habituation to difficulty that makes capability compound over time.
Emotional and social intelligence. Multiple experts in the report flagged the risk that AI-mediated communication — cleaner, more optimised, less awkward than human-to-human interaction — might gradually reduce people’s comfort with and ability to read genuine human social cues. The messiness of real conversations contains information. When that messiness is cleaned up — by AI-drafted messages, by AI-summarised meeting notes, by AI-suggested responses — some of that information disappears from the interaction.
Moral judgment. This one struck me hardest. The report includes expert commentary on AI creating “a paradox of control” — convincing individuals they are enhancing their lives while actually shaping their decisions to serve others’ interests. The more you delegate judgment to a system, the less you practice judgment. And judgment, unlike most cognitive skills, is hard to practice in low-stakes environments. You build it by actually making decisions under real conditions, with real consequences, that you are genuinely accountable for.
Agency and independence. The experts expressed concern about humans becoming less capable of self-initiated problem-solving — not because AI is forcing them to be passive, but because the path of least resistance is increasingly to ask rather than to think. There’s an important difference between using AI as a tool to execute on a direction you’ve set, and using AI as the thing that sets the direction.
Where I’ve seen this playing out
I want to be careful here not to generalise from anecdote, but I’ve watched people I work with interact with AI tools long enough to notice some patterns that feel consistent with what the report describes.
The most striking one: the gap between people who use AI as an accelerant and people who use AI as a replacement.
The accelerant users come into an AI conversation with a position — a draft, an opinion, a direction — and use the AI to stress-test it, expand it, edit it, challenge it. They are always the agent in the interaction. The AI is the instrument.
The replacement users start the AI conversation without a position. They ask what they should think. They take the first output as the answer rather than as a starting point. They’re not using the tool. The tool is using them — or more precisely, they’re outsourcing to the tool a function that only compounds through practice.
The difference in outcomes between these two modes is significant and, I think, widening. The accelerant users are getting better faster, both at their actual domain and at using AI effectively. The replacement users are producing more output but developing slower — or in some dimensions, going backwards.
I’ve started asking a question in my teams when we’re doing retrospectives on AI use: did the AI give you an answer, or did you have to argue with it? The people who regularly argue with their AI tools — who push back, who test the output, who stay in the driver’s seat — are learning. The people who take the first output and move on are not.
What the optimists in the report said
44% of experts expect more negative change than positive across the 12 human capacities. 29% expect more positive than negative. 16% said fairly equal. This isn’t an apocalyptic finding — there’s real expert belief that AI will enhance certain human capacities, and even the pessimists tend to believe that humans will adapt in ways that find new ground.
The optimistic essays in the “Being Human in 2035” report generally argue that humans have always adapted to tools that extended and in some ways replaced prior capabilities. Writing externalised memory. Calculators externalised arithmetic. Search engines externalised recall. The thing that got lost in each case was probably less important than what was gained. AI will follow the same trajectory.
I find this argument partially convincing and partially a bit convenient.
What it gets right: humans are genuinely adaptive. The generation growing up with AI won’t have the cognitive relationship with AI tools that people my age have — they’ll integrate these tools in ways we can’t fully anticipate, and some of those integrations will produce capabilities we don’t have the framework to evaluate yet.
What it undersells: not all cognitive functions are equal in their importance to being a full human. Externalising arithmetic is different from externalising judgment. Externalising recall is different from externalising emotional processing. The functions that AI is beginning to absorb in the 2025-2035 period are higher-order than the ones we externalised in previous technological transitions. That’s not automatically disqualifying, but it’s worth being honest about.
My actual answer to the question
What does it mean to be human in 2035? I’ve thought about this more than I want to admit, partly because I’m building an institution — Zero — that is supposed to be relevant to people who will be adults in that world.
Here’s where I’ve landed, which I hold with appropriate uncertainty:
What stays fully human: the choice of what to want. AI can tell you how to achieve almost any goal, and it will get better at this. It cannot tell you what goals are worth having — not in any deep sense. Values, priorities, what a life well-lived looks like for a specific person in a specific context — this remains a deeply human problem, and I think it becomes more important as execution becomes easier.
What needs deliberate cultivation: the capacity for discomfort. Boredom, uncertainty, the feeling of being stuck — these are cognitive inputs, not just unpleasant experiences. A person who is never allowed to be bored never develops the internal resources that boredom builds. If AI progressively insulates us from cognitive discomfort, we have to deliberately reintroduce it or we lose something structural.
What might genuinely improve: connection, if we’re intentional. The Imagining the Digital Future report raised concerns about social intelligence declining. I think that’s a real risk but not an inevitability. If we’re conscious that AI-mediated communication trades authenticity for efficiency, we can choose to protect the inefficient spaces — the long conversations, the unoptimised time with people we love, the messy meetings where what matters is the relationship not the output.
The 50% of experts who said “both better and worse in equal measure” are probably right. The question isn’t what AI does to humanity in the aggregate. The question is what AI does to you, specifically, given the choices you make about how to use it. That’s always been the question about any powerful technology. It just feels more personal now because the technology is personal in a way no prior technology was.
The part I can’t resolve
Here’s the thing that stays unresolved for me: the 61% of experts who believe change will be “deep and meaningful” or “fundamental and revolutionary” — they’re almost certainly right. And fundamentally revolutionary change to what it means to be human is not something you can model on past transitions, because the past transitions didn’t change human cognition and human relationship-formation at this scale.
Lee Rainie and Janna Anderson’s centre has been asking these questions since 2005. The current report is their 51st. I find it meaningful that the people who have spent 20 years tracking the relationship between digital technology and human behaviour are, in 2025, describing this moment as qualitatively different from the ones before. They’re not alarmists. They’re longitudinal observers who have seen what happens when predictions are too confident and too simple in either direction.
The responsible thing — maybe the only honest thing — is to stay genuinely uncertain about what 2035 looks like while being very intentional about the choices made between now and then. That’s not a satisfying resolution. But I think the people who claim to have it figured out are the ones to be most suspicious of.
