I remember discovering the internet in 1996. You could chat with people around the world. People who might never have had a platform suddenly had a voice.
Maybe it gave me a voice, too.
That discovery changed my career. I left my PhD because this new world was something I wanted to be part of. It wasn’t something I had learned in school. I was learning as I went, and I loved it.
AI gives me some of that same feeling today. The possibilities are amazing. Then it makes something up, misunderstands a request or sounds completely certain about an answer that deserves a second look.
And you think: can I really trust this thing?
I think that’s a fair question.
There are good reasons to hesitate
We shouldn’t assume that people who distrust AI simply don’t understand it. Dismissing hesitation as a lack of technical knowledge misses the point.
In a June 2026 survey, Pew Research Center found that 52% of U.S. adults felt more concerned than excited about AI’s growing use in daily life. Just 9% felt more excited than concerned. These are American findings, but they give us a useful reminder: enthusiasm isn’t universal.
For a small business owner, hesitation isn’t abstract. The questions are practical. Will it give a customer the wrong answer? Will it send something before I’ve approved it? What happens to the information I put into it?
For an employee, the question might be much more personal: are you asking me to help build something that will replace my job?
Those concerns deserve answers before another impressive demonstration.
“The smartest six-year-old kid ever”
That’s how I’ve described AI.
It can feel like the smartest person in the room. Ask it to explain something complicated, turn a jumble of notes into an outline or help work through a problem, and the response can be remarkable.
Then you discover the gap between a clever answer and dependable judgement.
“I even described AI as the smartest six-year-old kid ever.”
It’s a metaphor, of course. AI isn’t a child. It doesn’t mature through a human childhood, and treating it as innocent can make us too forgiving of serious mistakes. Being able to discuss almost anything doesn’t make it the best judge of everything.
The comparison captures my own experience of being impressed while still needing to supervise the result. It shouldn’t become an excuse for the companies building these systems, or the businesses using them, to avoid responsibility.
It can make things up and sound convincing
This is a real limitation. The technical label you’ll hear is hallucination: an answer that sounds plausible but contains invented or incorrect information.
The U.S. National Institute of Standards and Technology describes this risk in its generative AI guidance. It also warns that the explanation or citation offered to support an answer can be wrong.
That last part matters. Asking, “Are you sure?” isn’t much of a fact check.
If I’m preparing an article, I want the source. I want to see the passage behind the claim and understand how it was interpreted. Then I need to open that source and check whether it actually supports what we’re saying.
Imagine a retailer using AI to draft a product description. A polished paragraph is useful. An invented warranty could create a customer problem. The person reviewing it needs the real product information beside the draft.
Trust depends on what we’re asking it to do and how easily we can catch a mistake.
We worry that it will do something we didn’t want
There’s a real difference between an AI that suggests an email draft and one that has the authority to hit send.
Once a system can use tools, edit files or work across applications, a misunderstanding can become an action. That concern deserves more than “just write a better prompt.”
Anthropic’s guidance on trustworthy agents discusses both misread instructions and malicious content that tries to redirect an agent. It describes permission controls while acknowledging that safeguards aren’t a guarantee.
For your business, ask a few direct questions:
- What can this system read?
- What can it change or send?
- Which actions require my approval?
- Can I see what happened and stop further work?
Those boundaries need to exist in the actual setup. A friendly promise in a chat window isn’t a safety strategy.
This is part of planning an AI-assisted workflow: deciding how much authority makes sense for the job.
My Tesla ride brought that feeling home
I recently rode in a Tesla. There were moments when the stopping and hesitation made me uncomfortable, particularly around traffic. I found myself wondering how people become comfortable trusting a car to drive.
That was my reaction as a passenger, not a technical assessment of the vehicle or its software.
It did remind me how different it feels when technology moves from helping us think to acting in the real world. Tesla says its Full Self-Driving (Supervised) features require active driver supervision and do not make the vehicle autonomous.
The consequences of a driving mistake and an awkward marketing draft are obviously different. We should be just as clear about the level of responsibility we’re handing over in each setting.
Sometimes the fear is about who benefits
We can answer every question about accuracy and still leave someone uneasy.
If an employee believes AI will be used to reduce headcount, a lesson on better prompts won’t settle that concern. The employee may understand the technology perfectly well. They’re questioning the employer’s plans.
I’ve written separately about AI, jobs and the tasks that are changing. Here, the important point is simple: trust in a tool and trust in the people deploying it are connected.
Owners need to explain what they’re trying to improve, how staff will be involved and what remains uncertain. Promising that nothing will change is difficult to reconcile with introducing technology because it can change so much.
Do headlines and Hollywood affect the conversation?
I suspect they do. If the picture in your head is The Terminator, a conversation about an AI assistant starts with a fairly large disadvantage.
But I wouldn’t blame the media for everyone’s concerns. That would be too easy. Reporting can expose problems we need to understand, and marketing can also encourage unrealistic expectations.
A headline about a system going wrong doesn’t tell you whether AI could help organize your meeting notes. A dazzling demonstration doesn’t prove it should handle your customer complaints without review.
We need a sensible middle ground where we can examine the actual task.
You can be curious and cautious at the same time
Some people enjoy trying technology early. I’m one of them. That curiosity changed my life when I discovered the internet.
It doesn’t mean every experiment works, or that everyone should move at my pace.
If you’re a business owner who feels hesitant, you don’t have to change your whole operation overnight. I’d start with something familiar, low-risk and contained: turn your own rough notes into a meeting agenda, for example. Use information you’re allowed to share. Keep the result private. Read it and correct it.
That gives you a chance to learn how it responds without handing over the keys to the business. Our article on building a useful AI working habit explores that first step in more detail.
I believe AI is here to stay. We still have choices about how we use it, what we permit and who remains accountable.
At the end of the day, you don’t have to trust AI with everything to find something useful to do with it.
Written from Stephen Lau's experience and point of view, with AI-assisted research and editorial production.