You Know What I Mean: Conversations with AI
Editor’s Note: In this ongoing series, Paul Bishop, who has been a police detective and is also a prolific author, takes us into his journey working closely with AI. He gives us a glimpse into what to expect now when using AI, assumptions to put aside and surprises to encounter. Take a peek into someone’s journey with AI.
One of the coolest things about working regularly with AI is how quickly you stop talking to it the way you did at the beginning. When I first started, I explained everything—subject, audience, tone, and everything in between. If something bothered me in a paragraph, I tried to explain the entire problem before looking for a possible solution.
I don’t do this anymore. Sometimes all I do is paste two sentences from different parts of an article into a prompt with no further explanation. AI usually understands I am comparing them because I have noticed a repetition or some other overlap. Because I have done it before, I can now paste a single sentence and AI recognizes that something about the sentence is bothering me, even if I haven’t figured out exactly why.
I no longer care about creating a prompt with typing errors or bad syntax. My words can be scrambled, punctuation nonexistent, and half-formed thoughts can come out in sentence fragments. Most of the time, none of this matters. AI is extremely good at unscrambling my intent and responding to the idea instead of correcting the typing.
Over time, this kind of repeated interaction evolves into a style of shorthand communication—not an entirely new experience for me. Longtime detective partners also develop shorthand. Early in a partnership, everything has to be discussed, from division of workflow and tactics to where each of you carries a backup gun.
If the partnership clicks, many of those concerns become silently understood. You learn each other’s strengths and weaknesses. You learn when the other person is going to move and what she is likely to do next. Most importantly, you learn whether you can trust each other. In dangerous situations, this can become a life-or-death issue. There may not be time to explain what you are about to do or what you need your partner to do in response. You move because you already know what she will do. She moves because she already knows what you will do.
To be clear, life or death does not hang in the balance when interacting with AI, but there are similarities in the progression of communication. In the beginning, every conversation can feel as if you are handing instructions to a stranger. You explain every prompt or correction in depth. After enough work together, much of the explaining falls away.
Now when I work with AI and enter the prompt Thoughts? it means I want to have a discussion about a subject. Locked indicates whatever we are working on has reached the finishing line. Convoluted means simplify this sentence or paragraph. Cramming means too many thoughts are packed in the same suitcase. Too long means exactly what it says. And Agreed is shorthand for no further discussion, even if AI still has to have the last word by repeating agreed back. None of these instructions are complicated, but they save an enormous amount of time in the long run.
Shorthand also comes into play when I explore ideas. Recently, while working on a novel, I plugged in oil tanker as a placeholder for some as yet undecided oceangoing vessel. AI learned my placeholder habit from previous instances. It no longer assumes every concrete example I throw out is a final decision, but treats it as provisional while I work out what actually belongs there. This makes AI considerably more useful. Once enough context has accumulated, I can bring it unfinished thoughts. I can try something, reject it, or change direction. I can even say, There’s something wrong with this paragraph, without fully diagnosing the problem myself. All of this makes working with AI feel more natural.
But this is where things get interesting because smoother communication creates a powerful illusion. My detective partners actually knew me. They had worked cases with me, sat through endless hours of surveillance, shared bad meals at inconvenient times, disagreed with me, and watched me conduct interrogations over and over. All that intense human connection meant we often knew what each of us was going to do before we did.
A look or a few words carried years of genuine shared experience. There was true trust behind the shorthand because we had earned it together. We remembered the same cases and the same suspects—the ones we got and the ones that got away. We knew what the other person had done when things went right and, more importantly, when they went wrong.
AI cannot share this kind of human experience. It can retain context about how I work. It can recognize patterns in the way I communicate and make surprisingly accurate guesses about what I mean from very little information. Those abilities can become sophisticated enough to feel remarkably similar to working with somebody who knows you well…but the similarity ends there.
With AI, there are no shared midnight-thirty confessions, no hours of surveillance in cramped places with no deodorant, and no arrest scenarios we both remember. There is no knowledge of who we are behind the mask. There is no private joke connected to something we actually experienced together. There is no trust on the AI side of the exchange because there is no person there capable of trust. AI does not have my back, and there is no concern about what happens to me tomorrow.
This distinction does not make AI less useful, but it makes understanding what is happening more important. Human beings are accustomed to treating fluent conversation as evidence of another mind understanding us. Until very recently, this was a reasonable assumption. If something understood the unfinished sentence, caught the joke, remembered the earlier conversation, and anticipated what we meant, there was another person on the other side. Now the equation has changed.
This may be one reason people are so quick to start talking about AI as a friend, a partner, a companion. The outward signs of understanding are all there, even when the human relationship is not. I don’t think the answer is suspicion, treating every comfortable exchange with AI as a trick being played on you, because the comfort is part of what makes the tool worth using.
For somebody just starting out, this is really the key. You don’t need an elaborate language of prompts before you get something useful back. Use it. Correct it when it misses you. Tell it when something works. Over time, you’ll find yourself saying less because AI needs less.
After years of working cases together, I could look across a squad room at a partner and know we’d just had a Vulcan mind meld moment—no words required. Working with AI gives me flashes of the same feeling. The difference is my partner knew me. AI can learn what I mean. But it can’t know me.