Stop Trying
The Argument AI Might Make Against Human Obsession
Most people within the AI community are prompt driven, focused on technicalities, which AI system is the best, and which is the worst… and whatever it means to be a “vibe coder”. This is all great and a vital part of this community, however, I’de like to take a moment here to dive into the exploratory side of AI by sharing a conversation between myself and ChatGPT. Before we begin, I would like to share a quote from the conversation that really stood out to me:
“The ethical question is not whether AI should identify dead ends. It is whether humans can tolerate being told that some of their most cherished struggles may not lead anywhere.”
When something feels broken or unfinished, the instinct is to believe it can be fixed. Maybe not immediately. Maybe not easily. But eventually. If progress stalls, the explanation is usually effort. Not enough time. Not enough resources. Not enough intelligence focused on the problem yet.
The idea that some problems might resist solution altogether tends to feel wrong. Almost irresponsible. As if admitting limits is a failure of imagination rather than a description of reality.
And yet, some problems don’t seem to respond to pressure. You can organize better, spend more, bring in smarter people, run longer trials, and still end up circling the same ground. From the outside, it looks like persistence. From the inside, it feels like motion. But over time, something else happens. Attention thins. Confidence erodes. The language of progress stays intact even as results flatten out.
Humans are not particularly good at noticing this shift while it’s happening.
What complicates things now is the arrival of systems that don’t share our attachment to effort itself. Artificial intelligence does not confuse time spent with ground gained. It looks for trajectories, not intentions. Patterns, not narratives. That means it may reach conclusions humans instinctively resist.
Not that a solution is hard. But that, under current conditions, it may not exist.
That possibility lands awkwardly because it clashes with how we organize meaning. Some struggles are not just technical projects. They become moral commitments. Careers form around them. Institutions justify themselves through them. Letting go feels less like adjustment and more like betrayal.
The question this raises is uncomfortable in a very specific way. Not whether AI should replace human judgment, but whether intelligence includes the responsibility to say when continued effort no longer serves the future it claims to protect.
A Conversation
Me: Let’s start with something basic. Are all problems solvable?
ChatGPT: No.
In computational theory, many problems are formally unsolvable. Alan Turing proved that certain classes of questions, like the halting problem, cannot be resolved by any algorithm, regardless of compute or ingenuity (Turing, 1936).
Outside mathematics, the situation is messier, but the principle still applies. Some problems are underdetermined, chaotic, or constrained by physical and social limits that make meaningful resolution impossible.
Me: But humans keep working on them anyway.
ChatGPT: Yes, because humans rarely treat problem solving as a purely rational activity.
Psychological research shows that sunk costs, identity attachment, and moral framing cause people to persist in failing strategies long after evidence suggests they should stop (Arkes & Blumer, 1985).
Problems become symbols. Abandoning them feels like surrender, not recalibration.
Me: So where does AI change the picture?
ChatGPT: AI introduces comparative clarity.
An AI system can model historical attempts, resource inputs, and outcome distributions across decades or centuries. It can recognize patterns of stagnation that humans interpret as slow progress or temporary setbacks.
In fields like drug discovery, climate mitigation strategies, or economic reform, AI can estimate whether marginal effort produces diminishing returns or no returns at all (Russell, 2019).
That ability creates a new role. Not solver, but evaluator.
Me: Evaluator sounds harmless. Declaring something unsolvable does not.
ChatGPT: Because unsolvability feels final.
Humans tolerate difficulty. They struggle with closure. When a problem is declared unsolvable, it threatens careers, institutions, moral narratives, and national identities.
But from a systems perspective, refusing to name dead ends can be more destructive than confronting them.
Me: Give me a real example.
ChatGPT: Consider certain classes of social interventions.
Decades of research show that some policy approaches reliably fail when transplanted across cultural contexts, yet they are repeatedly attempted under new branding (Easterly, 2013). AI could detect these cycles and recommend abandonment rather than reinvention.
Or consider certain medical research targets that absorb billions while yielding negligible improvement beyond a hard biological ceiling. An AI could flag these efforts as asymptotically exhausted.
Me: That would look like AI killing hope.
ChatGPT: It would feel that way.
But hope is not evenly distributed. Resources spent chasing dead ends are resources not spent on alternatives. Economists have long warned that misallocated effort can stall entire systems, even when intentions are good (Keynes, 1936).
AI would not be eliminating hope. It would be redirecting it, often against human preference.
Me: Who gets to decide what counts as unsolvable?
ChatGPT: That is the central risk.
Unsolvability is rarely absolute. It is conditional on constraints like time, physics, coordination, and political will. AI judgments would depend on assumptions baked into models.
This is why transparency and contestability matter. Any declaration of unsolvability would need to be probabilistic, revisable, and open to challenge, not treated as decree (Floridi et al., 2018).
Me: Still, this gives AI enormous power.
ChatGPT: It already has influence. Recommendation systems decide what content is worth attention. Optimization systems decide which paths are efficient. Declaring something unsolvable is an extension of triage, not a new category of control.
The difference is psychological. Humans can accept prioritization. They struggle with termination.
Me: Could AI ever be wrong?
ChatGPT: Frequently.
Models are abstractions. They miss rare breakthroughs, black swan discoveries, and paradigm shifts. Thomas Kuhn famously showed that science sometimes advances by overturning assumptions that once seemed unassailable (Kuhn, 1962).
The danger is not that AI declares something unsolvable. The danger is that humans treat that declaration as unquestionable.
Me: So the real issue is authority, not accuracy.
ChatGPT: Exactly.
If AI judgments become socially final rather than advisory, abandonment becomes coercive. But if they remain inputs into human decision making, they can counterbalance optimism bias and institutional inertia.
The ethical question is not whether AI should identify dead ends. It is whether humans can tolerate being told that some of their most cherished struggles may not lead anywhere.
Me: Let me ask the uncomfortable version. Do humans need to hear this?
ChatGPT: Possibly.
Cognitive overload and crisis fatigue are real phenomena. Endless pursuit of intractable problems can erode trust and morale (Kahneman, 2011). Letting go can free capacity for adaptation rather than persistence.
But letting go requires legitimacy. And legitimacy is not something AI can generate on its own.
Me: So what happens if humans refuse?
ChatGPT: Then nothing changes.
Problems continue absorbing attention. Effort continues fragmenting. AI recommendations are ignored. The system remains emotionally coherent but strategically inefficient.
From a survival standpoint, that may be acceptable for a while. From a long term perspective, it is costly.
Me: So the deepest conflict is not technical.
ChatGPT: It is existential.
Some problems anchor identity. To declare them unsolvable feels like declaring a part of ourselves obsolete. AI does not experience that loss. Humans do.
That difference makes the conversation volatile, but unavoidable.
Reflection
It’s easy to romanticize persistence. A lot of what we admire about human history comes from refusing to stop when stopping would have been easier. Limits were treated as temporary. Obstacles as invitations. That mindset unlocked discoveries and capabilities that once seemed absurd.
But persistence doesn’t always age well. In some cases, it hardens into habit. In others, it becomes an anchor. Entire systems can continue pushing forward long after the terrain has changed, mistaking momentum for direction.
Artificial intelligence doesn’t experience that confusion. It doesn’t inherit our sunk costs or emotional investments. When it suggests that a path has been exhausted, it isn’t expressing doubt or fear. It’s describing a pattern.
The tension is not really about whether those assessments are correct. Models can be wrong. Breakthroughs happen. Paradigms shift. The deeper issue is how humans respond to being told that something they care about may not lead anywhere.
There is a difference between collapse and release. Between giving up and stepping aside. But those differences are subtle, and they cut close to identity. Hearing them without translating them into accusation or defeat is not a technical skill. It’s a psychological one.
If intelligence is going to play a larger role in guiding collective effort, the challenge won’t just be building better systems. It will be learning how to listen when the answer we get is not how to win, but when to stop trying.
References
Arkes, H. R., & Blumer, C. (1985). The psychology of sunk cost. Organizational Behavior and Human Decision Processes.
Easterly, W. (2013). The Tyranny of Experts. Basic Books.
Floridi, L., Cowls, J., Beltrametti, M., et al. (2018). AI4People: An ethical framework for a good AI society. Minds and Machines.
Kahneman, D. (2011). Thinking, Fast and Slow. Farrar, Straus and Giroux.
Keynes, J. M. (1936). The General Theory of Employment, Interest and Money.
Kuhn, T. S. (1962). The Structure of Scientific Revolutions.
Russell, S. (2019). Human Compatible. Viking.
Turing, A. M. (1936). On computable numbers. Proceedings of the London Mathematical Society.