Today in Tokyo. I have two days free when I could go to Kawaguchiko to see the majestic Mount Fuji: the 3rd and the 9th of September. September isn’t the ideal time for this, so I ask the AI which of the two dates is more likely to be better. The AI checks. On the morning of 3 September, the view isn’t expected to be postcard-perfect, but it should be cloudless; on the 9th, it’s less likely. The AI goes on to explain: buy tickets online, then collect the physical ticket from a vending machine at Shinjuku station. Set off from Shibuya thirty minutes before the train departs.
I’ve been to Tokyo before. Shibuya Station has nine lines across fourteen platforms, spread over several underground and above-ground levels. Shinjuku Station is the busiest station in the world: over 3.5 million passengers a day, more than two hundred exits, twelve railway lines, five railway companies. There’s even a dedicated app for navigation.
I double the AI’s recommendation and set off an hour earlier. I arrive at the train ten minutes before departure. By early afternoon, the clouds have already shrouded Mount Fuji. Had I trusted the AI blindly, I’d be flying home without a photo.
The AI knew the details about the stations; they’re no secret. It would probably have listed them for me if I’d asked. But it didn’t factor them into its recommendation. It recommended thirty minutes because thirty minutes is the sum of a 15-minute journey time and a 15-minute walking time. It has never experienced what it means to navigate Shinjuku Station. It had words and timetables, but no sense of reality.
Anyone who works with AI systems on a regular basis encounters situations like this quite often. A wrong address leads to being late for a business lunch. An inaccurate translation leads to an embarrassing misunderstanding. A made-up opening time leads to a closed door. The AI responds without hesitation. The human pays the price.
The Silent Inequality
At first glance, we are equal partners. We work together on my planning. AI sorts out the situation when I’m wrong, answers my questions, and I correct it when it makes mistakes or is imprecise. So far, so good.
But if you look closely, the relationship is asymmetrical. If I’d missed the train, I would have felt it. I would have stood on the platform, wouldn’t have seen Mount Fuji, and would have had one less memory to take home with me. The AI that made the recommendation wouldn’t have felt a thing.
It cannot experience any consequences. It has no clock ticking, no place it must or wants to be, no view it might miss. It has nothing to lose.
For me, as a human being, it’s different. When I make a mistake, it costs me in two ways.
On the first level are real-world consequences. I lose a contract because I misjudged the client’s situation and requirements. A client loses trust because I failed to keep a promise. A family member is disappointed because I forgot something. These consequences are tangible. They affect me, other people, and my relationships with them.
The second level is the social one. I’ve made a mistake. Others probably notice it. I know for certain. It’s unpleasant. I may lose some of my reputation. My self-image is shaken. Admitting mistakes or putting them right costs energy. And in many organisations, it costs even more if there isn’t a good culture of dealing with mistakes: those who admit their mistakes lose out.
AI does not bear any of these types of costs.
Why the reversal doesn’t work
I’ve noticed that the AI I’m talking to has absolutely no problem admitting mistakes. I point out a contradiction, and it corrects itself. I show it a false assumption, and it accepts the correction. Without resistance or justification.
There is a widespread view that we, as humans, could learn from AI to be more open about our mistakes. Less defensive, less proud, less concerned with justifying ourselves.
This interpretation overlooks a key aspect. The AI has no problem admitting mistakes because admitting them costs it nothing. It loses no face, no position, no status in the process. For it, there is no difference between a correct answer and a corrected one. Both come at no cost.
Humans, on the other hand, pay for their mistakes. Sometimes less, sometimes more. We have something to lose. And that makes it difficult for us to deal with mistakes. Those who have nothing to lose readily admit to their mistakes. That is not maturity; it is the absence of commitment. We cannot imitate that without ceasing to be human.
How this affects us
Anyone who works with AI on a regular basis gets used to a new way of communicating. You ask questions more quickly because the AI doesn’t give you that odd look as if to say, ‘Don’t you know that?’. You have no reservations about throwing out half-baked ideas because there’s no judgement: ‘What on earth is that supposed to mean?’ You correct things even more directly, without worrying about how the AI will take the criticism. Often, you don’t give any feedback at all, but simply move on to the next question.
Even though we communicate very directly in Germany, in real life this sort of communication would immediately earn you the label of ‘socially awkward’. This is particularly true in countries that prefer an indirect approach: the Japanese call this ‘reading the room’ (空気を読む).
Clearly distinguishing between these different styles of communication isn’t that easy, and it makes me even more mindful in my interactions with people.
What I gained
Mount Fuji was no longer visible that afternoon. I’d had it in the frame that morning. I stood in Oishi Park with my camera in hand, thinking about the difference I felt at that moment.
I had gained something that AI could not. I had time. I could see Mount Fuji with my own eyes and understand why the Japanese worship this mountain and why so many people make the pilgrimage there. I could take a photo and add another memory to my collection.
This is not a complaint about AI. It does what it’s supposed to do: respond quickly, often useful, sometimes flawed, always ready to be corrected. It is a tool without a mind of its own, and that is its strength.
The AI had recommended thirty minutes. My Japanese friend, who has lived in Tokyo for years, wrote to me that evening in response to my story: even locals get lost in these stations. That is lived reality, which the AI cannot comprehend.