The AI Scare Is Getting Louder. Does That Mean You Should Stop Using AI?
If you have spent any time online recently, you have probably noticed a change in the conversation around AI.
The optimism of the last few years is increasingly sharing space with something else: warnings about autonomous systems, misinformation, privacy, job displacement, psychological dependency, and increasingly capable AI agents behaving in ways their developers did not anticipate.
Some of these concerns deserve to be taken seriously.
But there is a problem with the way the conversation is evolving.
We are beginning to talk about "using AI" as though it were one single activity.
It isn't.
Asking an AI assistant to organise meeting notes is fundamentally different from allowing an autonomous agent to access your email, credentials and financial accounts. Using AI to reformat a document is different from asking it to make a medical decision. Asking it to summarise several sources is different from allowing it to become your primary source of truth about yourself or the world.
The question, then, should probably move beyond whether we should use AI.
A much more useful question is: What should we use it for, what should we share with it, and where should human judgement remain firmly in the loop?
AI Risk Is Not Binary
Every technology has a risk surface.
We intuitively understand this elsewhere.
Using Google Maps to find a restaurant and allowing an autonomous vehicle to drive you there are both applications of software and location data, but nobody would argue that they carry identical consequences when something goes wrong.
AI deserves the same distinction.
The risk of an AI interaction changes according to several variables: what information you provide, how much authority you give the system, whether its output can be independently verified, and what happens if the system is wrong.
That gives us a much more practical way to think about adoption.
A useful rule is:
The greater the consequence of an error, the greater the need for human verification and control.
This sounds obvious. Yet much of the current debate collapses everything from grammar correction to autonomous agents into the same category.
They are not the same category.
Start Where AI Is Boring
Ironically, some of the most useful applications of AI are also the least spectacular.
Research is one.
Instead of opening 20 tabs and manually extracting information, AI can help identify relevant material, compare sources and organise what you find. The important distinction is that the sources should remain visible and verifiable. AI can accelerate the research process without becoming the evidence itself.
Fact-checking is another useful application, provided it is grounded in external sources rather than asking a model, "Is this true?" and accepting the answer. A good AI workflow helps you locate the original study, government document, company filing or reputable reporting, then allows you to make the judgement.
There are even simpler applications: formatting information, restructuring a document, turning rough notes into categories, summarising material you have already read, comparing options against criteria you defined, or helping organise a collection of thoughts into something coherent.
These use cases are almost mundane.
That is precisely why they matter.
You do not have to hand AI the steering wheel to benefit from the engine.
The Skill We Need Is Discernment
For years, digital literacy meant understanding that you should not click suspicious links, reuse passwords or believe everything you read online.
AI literacy will require its own set of instincts.
One of them is understanding what should never be casually shared.
Passwords, authentication credentials, confidential company information, proprietary datasets, sensitive personal records and information you do not have permission to disclose should trigger the same hesitation they would if you were entering them into any other third-party service.
Another is understanding the difference between assistance and authority.
AI can help you investigate a question.
It should not automatically determine what you believe.
AI can help you structure a decision.
It should not necessarily make the decision.
AI can surface possibilities you had not considered.
It should not become a substitute for your own judgement.
This distinction becomes increasingly important as AI systems become more conversational and persuasive. The interface feels human, but the mechanism underneath it is still a computational system capable of producing confident errors.
Verification Is Part of the Workflow
There is an odd expectation emerging around AI: if it occasionally gets something wrong, it cannot be trusted.
That is the wrong standard.
Humans get things wrong. Search results contain misinformation. Wikipedia has errors. Consultants make bad recommendations. News organisations publish corrections.
The useful question is whether the system allows you to verify what matters.
For low-stakes tasks, verification may barely be necessary. If AI reorganises your brainstorming notes incorrectly, you fix the category.
For research, verification means opening the sources.
For financial modelling, it means checking assumptions and formulas.
For legal, medical or other consequential decisions, it means involving qualified professionals and authoritative information.
AI literacy is partly the ability to scale scrutiny with consequence.
This Matters Even More for Entrepreneurs
For founders, rejecting AI entirely would mean ignoring one of the largest reductions in the cost of starting a business in decades.
A solo founder can now use AI to research a market, explore business models, organise customer feedback, structure an initial go-to-market plan, draft operating documents and interrogate assumptions before spending significant money.
None of this requires surrendering control.
In fact, the best use of AI may be the opposite: using it to improve the quality of the questions you ask before you make a decision.
You can use a tool like our Business Model Picker to explore possible models, then research and challenge the assumptions behind them yourself.
The AI does not need to be the founder.
It can simply make the founder better equipped to think.
The Next Phase of AI Will Require Better Humans, Too
The current wave of concern around AI is probably healthy.
Powerful technologies should be questioned. Developers should be held accountable. Safety failures should be investigated rather than dismissed, and users deserve to understand where their information goes and what authority they are giving automated systems.
But caution and adoption can coexist.
We do not need to choose between unquestioning enthusiasm and abandoning the technology altogether.
There is a much larger middle ground: use AI where the downside is understood, keep humans involved where judgement matters, verify claims that matter, and think carefully before sharing information or delegating authority.
The most important skill of the AI today, in light of safety, will be discernment. That is: Knowing when to ask AI vs when something should remain human.