AI hasn't overtaken us as a species, whatever the more excitable headlines suggest. But it has quietly rearranged what we spend our days doing, what we worry about, and what we've started to prize in ourselves precisely because a machine can't do it. That rearrangement is the real story. Not the robots, the values.
Experts in the field describe AI itself as going through something like adolescence right now. Most of what you encounter day to day is "weak AI" — narrow, applied, built to do one job well. Somewhere down the line, research may shift toward "strong AI," systems explored for what intelligence itself is, not just for what they can produce. We're not there yet. What we are dealing with, already, is a technology mature enough to reshape the texture of work and identity.
Welcoming a third era in computing
Anitha Raj, president of the IT consulting firm ARAR Technology, frames AI as the third great epoch of computing, after the mainframe era and the internet era. The mainframe age was centralized almost to a fault — a small group of operators and programmers controlling nearly everything that happened on a machine. AI decentralizes that control in ways we're still working out.
Underneath the label "AI" sit two quite different approaches. Symbolic learning tries to mirror human perception and reasoning directly, which puts it closer to how we actually think. Machine learning takes the opposite route: pure logic, pattern, and data, with no pretense of mimicking a mind. Robots built on each behave differently — a machine-learning system acts on statistical inference, while a symbolic one responds to what it "sees" in front of it, closer to instinct than calculation.
Whichever approach wins out task by task, the practical effect is the same: AI-driven capability keeps becoming ordinary. Futurist William Halal, of the TechCast project, predicts that within five to ten years AI could take over roughly 30 percent of the routine tasks people currently do — almost entirely through weak AI, the narrow kind. It's reasonable to be unsettled by that number; automation on that scale touches livelihoods, not just workflows.
The near-term shift: from persistence to creativity
There's a useful phrase for thinking about what's left once the routine work is gone: the border "beyond knowledge." It marks off the creative territory — cooperation, leadership, entrepreneurship, judgment under ambiguity — that stays stubbornly human even as machines get better at everything else. Here are four ways that border is already reshaping what we value.
1. Unrealized potential, finally realized
Halal projects 30 to 40 percent job losses in the most routine, least complex production and service roles by the late 2020s. That's the blunt part of the forecast. The more hopeful part: freeing people from menial tasks tends to open space for more demanding, creative work, and Halal projects roughly 10 percent employment growth for those more specialized, higher-skill roles as a result. Loss and gain, moving at the same time, in different corners of the economy.
Meanwhile AI keeps working its way through the human sensorium — sight and language first, now increasingly taste and smell, where progress has been slower but is real. A reminder that "artificial intelligence" isn't one capability arriving all at once, but dozens of separate frontiers, each moving at its own pace.
Where government intervention might be needed
According to a study by Pew Research, 62% of Americans believe AI will have a "major impact" on jobholders over the next 20 years, and about half of workers say the future impact of AI on the workplace leaves them feeling "worried" — not neutral, not excited. Raj thinks that wariness is earned: her view is that AI will ultimately eliminate more jobs than it creates. Halal has proposed a guaranteed annual income to cushion the roughly 10 percent of the labor force he expects to be most exposed to displacement. Whether that happens is a policy question. That the pressure to consider it is building isn't.
2. Creative thought becomes the premium skill
Younger workers carry the heaviest exposure to this shift, and their choices now matter more than most. Someone who doesn't want a four-year degree might instead train as a machine technician through a community college — a real alternative to the machine-operator roles AI is most likely to absorb. The technician's job, on most accounts, is more satisfying than the operator's ever was. Not every displacement is a straightforward loss.
The bigger change is upstream of any single job. Education itself has to adjust. When routine execution is handled by machines, creative thinking stops being a nice-to-have and becomes the central skill worth teaching — alongside a habit of continuous learning, since the target keeps moving.
3. Moving beyond individual knowledge
Researchers, including those at the McKinsey Global Institute, describe this as the biggest workforce transition since the Industrial Revolution — potentially compressed into a far shorter span than either the Industrial or digital revolution took. As routine tasks recede, the workforce moves past pure knowledge-work into something harder to automate: values, consciousness, belief, judgment — things that sit above knowledge rather than beside it.
Here's a distinction worth sitting with. Collective human knowledge — the pooled, half-intuitive understanding a group builds together — still outperforms anything a machine can produce on its own. Not all of what we know converts cleanly into data. Some of it lives in the room, in the relationships, in years of shared experience that were never written down anywhere a model could train on. That's a strength AI has no equivalent for, and likely one of the things we come to value more, not less, because a machine can't replicate it. Creative vision still belongs to us.
4. Transparency becomes non-negotiable
Fake news isn't a new problem, but AI sharpens an old one. Facebook's third-party fact-checking program has been flagging false stories since 2016, and generative tools have only made convincing fabrication easier still. Questions about ethics, policy, and data access aren't abstract anymore. They're operational.
We also need a working definition of liability that didn't exist before. If an AI system gives a wrong or misleading answer and someone is harmed by it, who answers for that — the programmer, the company, the system itself? Right now the honest answer is: it depends, and the law hasn't caught up.
The privacy problem underneath it all
Businesses need large volumes of data to make AI genuinely useful to them, and that need runs directly into a rising bar for transparency. The more data a company wants, the more it has to explain why, prove it's entitled to collect it, and show real consent from the people the data describes. None of that is optional anymore, even where it isn't yet fully enforced — and expect lawmakers everywhere to keep tightening the rules as privacy gets harder to protect and more valuable at the same time. Regulation is playing catch-up here, which tells you something about how fast the underlying technology moved.
The long game
The near-term disruption is real, but it isn't the whole story. Given time, AI's effect on human values looks less like erosion and more like redirection. Healthcare is where the upside shows up earliest and most concretely — drug discovery, radiology, pathology — with the promise of inventing treatments faster and more cheaply, including for conditions that have resisted treatment so far.
The volume of digitized patient data, medical histories, hereditary risk profiles, keeps expanding, making genuinely personalized medicine more achievable than it's ever been. Safer transport and broader access to education sit in the same column: real benefits, arriving on a longer timeline than the disruption does.
And here's the part worth holding onto. AI still can't originate creativity or feel compassion, and there's no clear sign that changes soon. The values furthest from what AI can do, empathy, emotional presence, genuine care, are likely to matter more to us, not less, the further this technology advances. We may end up needing each other more, not less, precisely because of it — a question this piece on finding fulfillment in the age of intelligence explores in more depth.
Can you program a machine to hold values?
AI systems are adaptive, interactive, and increasingly independent in how they operate, which is exactly why some people find them unsettling. In principle, it's possible to build a system that behaves according to a given set of values and adjusts them through interaction with its environment. The philosopher James Moor proposed four categories for how an AI agent might function within a moral framework: as an ethical impact agent, an implicit agent, an explicit agent, or a full ethical agent — a framework explored further in how to embed core values into AI and robotics software.
Almost every AI system qualifies as an ethical impact agent by default — simply by having some effect, good or bad, on the world it touches. Implicit agents go a step further: built with value-sensitive design baked into their architecture, so their behavior tends toward certain values without ever "reasoning" about them. Explicit agents can actually represent and reason over ethical categories in something like machine language. Full ethical agents, the kind with something resembling free will, intentionality, genuine conscientiousness, are the rarest category by far — and it's fair to say they barely exist yet, if they exist at all.
Is adaptability a value, or a risk?
Here's the tension worth naming honestly. Building an AI system that can adapt its own values as it learns isn't really the goal anyone should be chasing. Building one with real ethical sensitivity is a different, better goal — and the two aren't the same thing, even though they get conflated.
Past experience with adaptive algorithms shows they cut both ways. Adaptability is genuinely useful when it works. It becomes dangerous fast when the learning process is opaque, or when it's shaped by biased or narrow experience the system never had reason to question.
So the real question isn't whether AI should adapt. It's what should stay fixed while it does. One answer: look for meta-values — principles that either never change or can only be changed by a human hand, while everything else is left free to adapt. Candidates for that short list include accountability, transparency, meaningful human control, and reversibility. Together, they let a person notice that an agent's values have shifted, understand why, and undo it if needed. Without that scaffolding, adaptability is just drift with better marketing.
Can AI actually learn human values?
AI already exceeds human performance in a growing list of narrow fields, which makes the question of alignment, getting AI's behavior to track what we actually value, urgent rather than academic. It's also genuinely hard, for a reason people rarely admit: humans are often bad at articulating our own values. We know something matters to us. We frequently can't say why, beyond the fact that it was instilled early, before we were in any position to question it. Asking a machine to reverse-engineer values we can't fully explain ourselves is a strange kind of task.
Part of the difficulty is that we don't understand how values are represented in the human brain to begin with. Even defining a value precisely is harder than it looks. AI is built from data; human values are an evolutionary inheritance, shaped by ethics, psychology, sociology, none of which reduce cleanly to neuroscience, let alone to a training set. Justice and fairness aren't concepts anyone has managed to fully cash out in neural terms.
It would be reassuring if avoiding bias were just a matter of using balanced, well-curated data. It isn't. Balance and fairness resist simple data rules in most real cases, because the same question about the same value can get different, equally sincere answers depending on context, mood, or who's asking. Try to infer something like "responsibility," "happiness," or "loyalty" purely from a dataset and the limit shows up fast. Getting closer to real alignment means bringing in the social sciences, not just better statistics.
You can't answer a question you haven't learned to ask
OpenAI describes its alignment research as aiming to make AI systems follow human intent and do what humans actually want, particularly in the cases where humans themselves aren't sure what they want — a deceptively simple aim for an enormously hard problem. Much of the difficulty comes down to which questions get asked, and of whom. The same question produces different honest answers depending on who's answering: raising taxes might serve the collective good while genuinely costing an individual. Even "how's the weather" gets different answers depending on where you stand.
This kind of learning runs into real limits: deception and ambiguity chief among them. People often can't give a clean answer about a value, not from bad faith, but from cognitive or ethical bias, a fuzzy sense of what "right" even means here, or plain lack of expertise. Some of that friction eases if you strip out unnecessary ambiguity from the question itself — though anything involving planning for the future carries built-in uncertainty no amount of careful phrasing removes.
Deception is the harder problem. An answer can sound completely credible and still be wrong in a way that isn't obvious, and that gap between confidence and accuracy is exactly where value misalignment creeps in. Learning to recognize deceptive behavior, in a person or a system, is difficult under the best conditions. We're not reliably good at spotting it in other humans. There's little reason to expect it will be easier in machines.
A recent Bloomberg piece made a point worth sitting with: today's chatbots can be prompted to imitate deception, producing false or misleading answers on request, but what they're doing is best understood as "only imitating humans," not lying in any sense that implies intent. That's the nuance this whole conversation keeps circling back to. The machine reflects us. Understanding what it reflects, and being honest about what it can't yet originate, may be the most valuable habit of mind this era asks of us.
Sources
- https://www.bloomberg.com/opinion/articles/2023-03-19/chatgpt-can-lie-but-it-s-only-imitating-humans#xj4y7vzkg
- https://www.kdnuggets.com/2020/10/ai-learn-human-values.html#:~:text=AI%20systems%20can%20learn%20human,absence%20of%20a%20reflective%20equilibrium.
- https://www.valuechange.eu/project/artificial-intelligence/
- https://www.afcea.org/signal-media/artificial-intelligence-will-change-human-values
- Moor, J. H. (2006). The nature, importance, and difficulty of machine ethics. IEEE Intelligent Systems, 21(4), 18–21
Discover your own core values
Take the free, research-backed Values App assessment — nine minutes, max.
Discover your values