ChatGPT turned one on November 30. A few weeks earlier, at OpenAI's first developer conference, Sam Altman reported about 100 million weekly users. In mid-November, OpenAI's board dismissed him, and he returned within a few days. For a week, the entire industry followed the news as closely as an election night.
Those were the most visible events of the year. A quieter change deserves more attention. In the course of 2023, asking an AI went from novelty to reflex. Students use it for homework. Office workers use it for emails and reports. At home, people ask it about their symptoms, their taxes, their children's science projects and their disputes with neighbours.
This December, our main concern is how quickly users are handing over the thinking itself.
How has a year of ChatGPT changed the way we think?
Schools were affected first, because the issue reached them first. Some banned ChatGPT at the start of the year. Many are now looking for ways to teach with it instead. We teach a business launch program for new entrepreneurs, so we also observe this from the teaching side, and no one yet has every answer.
In September, UNESCO published its guidance for generative AI in education and research, the first global guidance of its kind. It recommends a minimum age of 13 for using these tools in the classroom and places strong emphasis on protecting human agency, meaning that the learner remains the one doing the thinking.
The notion of human agency is particularly relevant. It applies as much to a business owner as to a teenager.
One pattern stands out in how people discuss AI this year. The first answer is treated as the definitive answer. Users rarely ask a second time, and rarely ask where the information came from. If a response sounds confident and is well formatted, it is accepted. The more useful the tool becomes, the less it is questioned, which is the opposite of what good practice requires.
People are no less capable than before. The tools have simply made it very easy to skip the step where we apply our own reasoning.
Which habits help you use AI while keeping your judgement?
Three habits make the difference. None of them is complicated, and we recommend them to every member of a team.
How should you handle the sources an AI provides?
Ask the AI for sources, then verify them. Requesting sources is a good practice; accepting them without checking is not. Since the spring, it has been established that these tools can invent references that appear entirely genuine, down to the page numbers. Two New York lawyers learned this in June, before a judge.
The habit therefore has two steps. First, ask where the information came from. Then open the link, locate the document and read the relevant paragraph. In most cases, the source is real. In some cases, it says something different from what the AI stated. In others, it does not exist. Only a direct check will tell you which.
Tools that search the web while answering, such as Bing's chat, are more reliable at providing real pages. The habit still applies: a real link to a page that does not support the summary is another form of error.
How can AI help you test your own ideas?
Use it to argue against your own position. In our experience, this is one of the most valuable uses of these tools, and very few people do it by default.
Most users present their idea and ask the AI to improve it. The AI will do so, because it tends to be agreeable. It will refine your plan, tidy your email and make your pitch sound compelling. The result is pleasant but of little help in making a decision.
We recommend reversing the request: "Here's my plan to open a second location in the spring. Give me the five strongest reasons it could fail." "Argue the other side of this." "What would a sceptical banker ask me?" The tool then works for you. You do not need to agree with its answers; you only need to address them.
When should you bring AI into a decision?
After you have formed your own view. If you ask the AI what to do before considering the question yourself, its answer becomes your starting point and anchors your thinking. Everything that follows becomes an adjustment of its idea rather than the development of your own.
Begin by writing down what you think, even in two lines: what you would do, and why. Then use the AI to identify weaknesses, fill gaps and point out what you may have missed. You will get more value from the tool, and the decision remains yours.
This matters more for small businesses than is often assumed. Your knowledge of your customers, your town and your market is precisely where you know more than any model. Handing it over first means giving up your most significant advantage.
What should businesses expect as AI spreads into everyday tools?
The trend points toward greater difficulty, not less. Two weeks ago, Google integrated its new Gemini model into Bard. Microsoft renamed Bing Chat to Copilot and is selling Copilot within Word and Outlook to business customers. Google is doing the same with Docs and Gmail. Increasingly, users will no longer choose to consult an AI; it will offer suggestions before they have finished thinking.
As the models improve, obvious mistakes will become rarer, and that is the main risk. Fewer visible errors lead users to stop checking, and the errors that remain are those that appear correct.
We suggest a simple exercise for the new year. After a few weeks of relying on an assistant for a given task, review the decisions you made during that period. Count how many began with your own idea and how many began with the AI's first answer. If the second group is larger, you will have gained a useful insight into your own practices at no cost.
Other articles on this subject are available on our blog. We use AI every day, and we do not advocate keeping it out of the work. Our recommendation is that you remain fully involved and that the final decision stays with you.