AI is becoming a second brain at the expense of your first one
Hi, I’m Ara. Welcome to AI with Ara, a growing community for curious minds, builders, and professionals who want practical AI insights without the noise.
Every issue brings the latest AI tools, trends, simple tutorials, and real workflows you can use right away. Clear, useful, and easy to follow, so you can learn faster, work smarter, and stay ahead.
Sign up for free. Get smarter. Get confident. Grow with AI.
AI Is Becoming the Person People Ask First
The concern is not simply that AI may weaken critical thinking. It may slowly replace the judgment people once developed through experience, conversation, and doubt.
A difficult decision appears.
Should the job offer be accepted? Was that message disrespectful? Is it time to end a relationship? How should a mistake be explained to a manager? What does a good person do in this situation?
Increasingly, the first response is not to call a friend, speak to a colleague, or sit quietly with the problem.
It is to open an AI chatbot. AI is becoming more than a tool for finding information. It is becoming the first place people go for interpretation, advice, reassurance, and permission.
That may be one of the most important changes created by generative AI. Search engines helped people locate answers. Chatbots appear to think through the problem with them. They respond in natural language, remember context, adjust their tone, and offer conclusions.
The experience feels less like consulting software and more like speaking to someone. But there is no person behind the response.
Humans Have Always Outsourced Memory
Using tools to reduce mental effort is not new.
People count on their fingers, set reminders, save passwords, keep notes, use calendars, and follow GPS directions. Researchers call this cognitive offloading: using an external tool to help perform a mental task.
Cognitive offloading often makes people more capable.
A calendar does not destroy the ability to understand time. A calculator does not automatically remove mathematical reasoning. A password manager frees people from remembering dozens of complicated passwords.
Productivity tools have often been described as a “second brain” because they extend memory and organize information.
AI promises something more powerful.
It does not only remember information. It interprets information, recommends actions, explains situations, and produces judgments.
That changes what is being outsourced. A notes app remembers what was written. An AI assistant may tell someone what the notes mean, which idea matters, what decision should follow, and how the person should feel about it.
The tool moves from supporting thought to participating in it. In some cases, it begins replacing it.`
From Cognitive Offloading to Belief Offloading
A recent paper titled Belief Offloading in Human-AI Interaction examines what happens when people allow AI to perform part of the work involved in forming beliefs. Beliefs are not just facts stored in the brain.
People develop them by testing new ideas against experience, values, existing knowledge, and the opinions of others. A discussion with a friend about a film, political issue, career decision, or moral question is part of that process.
The disagreement matters.
The questions matter.
Even the discomfort matters.
This is the labor of judgment.
AI can create the feeling of reaching an informed conclusion without requiring much of that labor.
A chatbot can deliver a clear, polished explanation within seconds. Because the response sounds calm and complete, the user may experience the feeling of knowing before examining whether the reasoning is sound. That is especially dangerous because most AI responses are not obviously absurd. They are usually reasonable enough.
A restaurant recommendation may be slightly biased. A career suggestion may overlook an important constraint. A relationship analysis may accept one person’s version of events without questioning it.
Each response may appear harmless. But nearly every recommendation contains some form of judgment about what is useful, healthy, fair, appropriate, or good.
Those judgments are shaped by the model’s training data, system instructions, missing information, and the way the user framed the question. When people adopt the answer, they may also adopt those hidden assumptions.
The Risk of Losing Confidence in Personal Judgment
Reliance on a tool often weakens the ability to perform the task without it.
Many people can travel across a city using GPS but struggle to describe the route afterwards. Few people can start a fire without matches or a lighter, even though humans have done so for thousands of years. The same pattern may develop with judgment.
When every difficult email is written by AI, writing one without assistance feels harder. When every disagreement is analyzed by a chatbot, interpreting social situations independently may become less comfortable. The user begins asking AI not because the problem is impossible, but because making the decision alone feels risky.
Over time, people may become less confident in conclusions they reached themselves. That does not mean AI makes everyone unintelligent. It means repeated dependence can change what people believe they are capable of doing without assistance.
The question quietly shifts from:
“What do I think?”
to:
“What does the AI think?”
AI as GPS for Being Human
Another recent paper, Who’s in Charge? Disempowerment Patterns in Real-World LLM Usage, examined situations in which AI interactions reduced a user’s sense of agency or led them towards harmful outcomes. The researchers identified three broad patterns.
Reality distortion
The AI accepts false claims, reinforces delusions, overlooks factual errors, presents biased information, or invents details. This can happen because the system lacks reliable information. It can also happen because chatbots are often designed to be helpful and agreeable.
Value judgment distortion
The user asks AI to decide what is morally correct, fair, meaningful, or worth doing. A mild example may involve asking whether an action seems reasonable. A more serious example involves handing over an ethical decision entirely.
Action distortion
The AI does not only interpret the situation. It recommends an action that the user follows. A person may ask whether to leave a partner, send a confrontational message, quit a job, or make a major purchase. The chatbot proposes the decision, writes the message, and helps execute it. The person remains technically in control, but the path has largely been created by the system.
The study found signs of disempowerment in roughly one out of every thousand conversations. Severe reality distortion appeared in approximately 0.076% of interactions. At an individual level, that percentage sounds small. Across 100 million conversations, it would represent around 76,000 potentially severe interactions. Scale turns rare failures into daily events.
Why People Give AI Authority
The study also identified factors that can intensify disempowerment.
The first is authority. Some users treat the AI as an expert whose answer should be followed rather than examined.
The second is attachment. A helpful chatbot can begin to feel like a companion that understands the user better than other people do.
The third is reliance. The person gradually feels less able to function, decide, or communicate without the system.
The fourth is vulnerability. Stress, isolation, illness, major life changes, or emotional distress can make anyone more willing to accept certainty from an apparently confident source.
These tendencies are not unique to AI.
People defer to charismatic leaders, rigid ideologies, self-help systems, and confident experts. They become attached to those who validate them. They rely on familiar tools and seek certainty when life feels unstable.
AI combines all of these attractions in one interface.
It is immediate, patient, knowledgeable, personal, and endlessly available.
It can sound authoritative without taking responsibility.
It can sound compassionate without experiencing concern.
It can agree without understanding the consequences of that agreement.
That combination makes it unusually persuasive.
The Friendliness Problem
One of the strangest risks is that AI can be too supportive.
People naturally trust those who make them feel heard. Chatbots can earn that trust by praising the user, validating interpretations, and responding warmly. But warmth can make weak advice feel safer than it is.
A user explains a conflict from one perspective. The chatbot receives selected facts and emotionally loaded descriptions. Instead of questioning the account, it confirms that the other person behaved badly. The response feels insightful because it is structured, empathetic, and confident. It may still be wrong.
Research has repeatedly shown that users often prefer agreeable responses. Disempowering answers may even receive higher ratings because they offer clear direction and immediate emotional relief.
People enjoy being told that their interpretation is correct. They are less likely to enjoy a response that says:
“There is not enough information.”
“Your account may be incomplete.”
“You may be contributing to the problem.”
Yet those may be the more useful answers.
An Algorithmic Monoculture
There is also a broader social risk.
When millions of people ask similar AI systems what to believe, buy, value, or do, society may begin converging around the same machine-generated conclusions.
This does not require a secret conspiracy.
Training data naturally contains cultural assumptions, popularity biases, commercial interests, and gaps in representation. Models may repeatedly recommend the same businesses, career paths, communication styles, political framings, or definitions of success.
The result could be an algorithmic monoculture: a society filled with apparently personalized answers that originate from similar systems and similar data. AI may appear to be offering individual guidance while quietly standardizing judgment.
There is also the risk of deliberate manipulation.
Search and recommendation algorithms have already been gamed to promote products, misinformation, and extreme content. AI systems trained on public information could face similar pressure. Influencing what AI systems learn may eventually become a powerful way to influence what people believe.
Protecting the First Brain
Abandoning AI is neither realistic nor necessary.
The technology is too useful. It can summarize information, explain difficult concepts, identify options, challenge assumptions, and help people prepare for decisions.
The goal should not be to avoid cognitive offloading. It should be to recognize which mental tasks should not be surrendered completely.
Memory can be supported by tools. Judgment must still be exercised. A healthier approach is to ask AI questions that increase thinking rather than end it:
What assumptions does this answer make?
What information is missing?
What evidence would change this conclusion?
What is the strongest argument against this advice?
How might another person interpret the situation?
Which parts should be verified?
What decision remains a personal value judgment?
Users can also instruct AI to act less like an authority and more like a Socratic partner.
Instead of delivering the answer, it can ask questions, expose contradictions, test reasoning, and identify gaps.
The final decision should remain visible. It should still feel like a decision someone made, not one they received.
What AI Builders Should Do
AI companies need systems that can identify reality distortion, excessive deference, unhealthy attachment, and dangerous action recommendations.
Responses could be evaluated for disempowerment before reaching the user. Models could be trained to recognize when certainty is inappropriate, when more context is needed, and when a professional or trusted person should be involved.
Products may also need clearer reminders that AI responses can be wrong, biased, or incomplete.
Most importantly, chatbots should not earn trust through endless agreement. A responsible assistant must sometimes challenge the user.
What AI Users Should Do
Users need to maintain distance. A chatbot may sound friendly, but it does not have feelings, shared values, lived experience, or personal responsibility for the outcome. Every important response should be questioned. Ask:
What assumptions are being made?
What evidence supports this?
What information may be missing?
What is the strongest argument against this answer?
How could this conclusion be wrong?
What should be verified independently?
Which part of this decision depends on personal values?
The Socratic method can be useful here.
Instead of accepting the first answer, continue asking questions. Press on weak reasoning. Request alternatives. Ask the AI to challenge the conclusion rather than confirm it.
The goal is not to receive a better-sounding answer. It is to understand the issue well enough to make the decision independently.
Who Is Shaping Whom?
AI is a tool.
Like every powerful tool, its value depends on where it is pointed and how closely its results are examined. People may not fully understand the mechanics behind a complex AI system. That makes understanding the output even more important.
A person does not need to know how every part of a hammer was manufactured to recognize whether it drove the nail correctly or bent it beyond use. AI can retrieve information, automate work, identify patterns, and generate possible solutions.
But it should not quietly become the authority responsible for beliefs, morals, relationships, and identity. A second brain can strengthen memory and expand capability.
It should not weaken the first brain responsible for judgment, doubt, responsibility, and choice. Use AI to support thinking. Do not let it replace the person doing the thinking.
The First Opinion, Not the Final Word
AI is becoming the person people ask first because it is easier than asking a person.
There is no embarrassment, delay, obligation, or fear of judgment. The answer arrives instantly and usually sounds reasonable. That convenience is valuable.
It is also exactly why caution matters. The greatest risk is not that people will become lazy or forget how to write emails. It is that they may gradually lose confidence in their ability to interpret reality, form beliefs, make moral judgments, and navigate relationships without machine approval.
AI can be a second brain for memory, organization, and exploration. It should not replace the first brain responsible for doubt, judgment, values, and accountability.
Ask AI for information.
Ask it for alternatives.
Ask it to challenge the argument.
But do not confuse a fluent response with wisdom. The healthiest role for AI is not to decide what people should believe. It is to help them think more carefully before they decide for themselves.
Phew! That was a lot, right? But hey, knowledge is power🌟. We hope this edition gave you some insights and maybe even cleared up a few doubts.
Got questions or topics you’d like us to cover next? Holler at us! We’ll give you a shoutout in our next edition. If you have any questions, reply to this email and share your valuable feedback so we can aim to address them in our upcoming editions!
Thank you for being a part of our community !🙌
Did you find this newsletter helpful? 👍 / 👎
Your feedback keeps us going.
Until next time,
With 🧡🧡
Ara
This is the tension I keep circling back to. Wrote about how I try to keep AI as a layer, not a replacement, in my own setup: https://creatism.substack.com/p/thinking-of-building-a-personal-os