There are questions people ask because they want answers, and there are questions they ask because they are not ready to have them answered.
Someone says, “I don’t know what I should do with my career,” and it sounds like a request for information. Maybe it is. But sometimes the unanswered question has become useful in its own right. As long as it remains open, the future remains open with it. No decision has been made, no sacrifice accepted, no failure risked.
In finance, an asset that rarely trades can carry a comfortable valuation for a long time. The unpleasant moment comes when a real transaction forces everyone to discover what it is actually worth. Some of our questions work like that. Their ambiguity has value. “I don’t know what to do” preserves possibilities that an answer would close.
AI is beginning to force those transactions in ordinary life. Ask a capable model how to start the business you keep talking about, and it will give you a plan. Ask what to say in the difficult conversation you have been avoiding, and it will draft the message. Ask why you cannot seem to learn the skill you keep saying you want to learn, and it will give you possible causes, exercises, a schedule, and a way to measure progress. The answer may be mediocre or wrong, but the old question now has a concrete answer sitting beside it, waiting to be judged.
I have spent enough time around technology to be suspicious of any claim that a new tool will solve a human problem simply by making information more available. The internet did not make people wise because it made books searchable. GPS did not make people adventurous because it made navigation easier. A spreadsheet does not make a company disciplined because it can calculate a number. AI cannot make you want what you claim to want, make the sacrifice for you, absorb the embarrassment of a difficult phone call, or bear the consequences of choosing one path and closing another.
It can, however, make I don’t know how much harder to say with a straight face.
Ignorance is a remarkably respectable excuse. It can make fear sound like prudence and procrastination sound like research. Sometimes the uncertainty is entirely real; there are decisions for which nobody has enough information and problems that resist easy solutions. But ambiguity can also become a shelter. If the question stays open long enough, nobody has to find out whether you were willing to answer it.
This makes some of the hostility toward AI more complicated than the familiar arguments about accuracy, jobs, privacy, or safety. Those arguments have substance. A system that confidently invents facts is dangerous. A system that encourages people to surrender judgment is dangerous. A system that concentrates enormous power in a few institutions deserves scrutiny. There is also a more personal possibility: sometimes the machine has produced an answer to a question that was doing useful work as an excuse.
Nobody is likely to describe the irritation that way. People will say the machine is soulless, reductive, overconfident, unable to understand, lacking in wisdom. Any of those judgments may be correct in a particular case. They can also be easier to admit than the possibility that a question was more comfortable while it remained unresolved.
We have always disliked people who make us confront conclusions we would rather postpone. The irritating person is not necessarily the one who tells us something false. Sometimes it is the person who takes our elaborate problem and reduces it to a choice we have been avoiding.
AI removes much of the interpersonal drama from that encounter. A human adviser can be accused of arrogance, condescension, ulterior motives, or simply enjoying the argument too much. A machine can produce the same unwelcome conclusion without taking any satisfaction in it. There is no rival to defeat and no relationship to renegotiate afterward. The irritation is left attached to the tool that made the conclusion difficult to avoid.
I expect that dynamic to make some arguments about AI increasingly moralized. We will say the machine is dehumanizing us when, in some cases, what it has done is make a particular ambiguity difficult to maintain. We will say it cannot understand us even as it becomes better at describing our behavior. We will dismiss its answers as not really answers even when they are good enough to force us back onto the decision itself. AI does not have to be right all the time. It only has to be right often enough to make ignorance an increasingly expensive alibi.
Researchers are already finding a peculiar split between how people judge AI advice and how they feel about receiving it. In five preregistered experiments involving 1,722 participants, advice generated by ChatGPT was often rated more highly than advice from an average human, while participants became more averse to the same kind of advice when they knew it came from ChatGPT. We do not judge advice only by the proposition in front of us. We also care who is giving it, what accepting it says about us, whether we trust the adviser, and whether we want to place ourselves under that adviser’s judgment.
AI is making one part of that exchange abundant. For most of history, if you needed to understand something difficult, you needed access to someone who understood it. If you needed a plan, you found someone who knew how to make one. If you needed competent writing, analysis, explanation, or tutoring, somebody had to supply the intelligence. Now the marginal cost of a plausible first answer is falling toward zero. Intelligence does not become worthless when it becomes abundant, but the scarce resources around it become easier to see.
Judgment, attention, courage, trust, and responsibility do not appear to be getting cheaper. A Microsoft Research study of 319 knowledge workers across 936 real-world uses of generative AI found that higher confidence in AI was associated with less reported critical-thinking effort, while workers described more of their role in terms of verification, integration, and “task stewardship.” Once producing an answer becomes easier, more of the burden moves to deciding whether the answer is any good and what should be done with it.
AI is not an oracle, and cheap answers create their own temptations. Experimental research on trust and reliance on AI has found that people can follow AI recommendations even when those recommendations conflict with contextual information available to them. A plausible answer can save us from thinking just as easily as it can save us from ignorance. The machine can remove the excuse that we did not know where to begin; it cannot relieve us of the obligation to judge what it gives us.
Suppose I ask an AI whether I should make a difficult phone call. It can examine the facts I give it, identify likely consequences, draft what I should say, and help me rehearse the conversation. Then the phone is still in my hand. I have to press the button, hear the other person’s voice, and accept what happens next. If I have misunderstood the situation, the machine does not have to repair the relationship. If I was cowardly, it does not have to live with the cowardice. If I was right, it does not share the relief.
This is why increasingly intelligent machines do not make me particularly pessimistic about the value of other people. They may make some people less useful as sources of information. They may make many kinds of advice cheaper. They may make the smartest person in the room considerably less impressive. Good. There are worse things that could happen to us than discovering that being the person with the answer was never the highest form of human usefulness.
Friendship, mentorship, marriage, family, and community have always carried an informational function. We ask people what to do because they know things we do not. We ask older people because they have lived longer. We ask friends because they know us. We ask a spouse because she has seen the pattern we cannot see in ourselves. AI is going to take over some of that work, and perhaps it should. If a machine can explain the tax form, teach me the algebra, compare the mortgage options, summarize the medical literature, outline the business plan, and help me prepare for the conversation, I do not need my friends to pretend they know an answer when they don’t.
I need the friend who knows why I am asking the question in the first place. The mentor who recognizes that my elaborate reasoning is mostly an attempt to avoid making a decision. The spouse who has heard the same excuse enough times to know that another framework will not help. The person who tells me to do what I already know I should do, and then remains close enough to see what happens.
A machine can give excellent advice without having anything at stake in whether I take it. A person can have something at stake because they love me. If machine intelligence keeps getting cheaper, that commitment becomes easier to distinguish from mere expertise.
AI could make us less dependent on people for answers while making us more conscious of why we need them at all. Once an infinitely patient and extraordinarily capable adviser is available whenever we want one, advice and relationship no longer have to travel together.
The machine may become where I go when I do not know what to do.
The person I call when I know what I have to do—and wish I didn’t—may become more valuable, not less.