Cyber Motion
The Briefing
How many times recently have you seen text or received a message that just looked… off? If you are anything like me it is happening more and more frequently. Sales pitches on LinkedIn, internal team memos, marketing position docs, blog drafts, and even business positioning documents all sound like AI because in many cases they are generated by AI from a fairly simple prompt. Today I’m going to talk about why that is an issue, and what you can do about it.
First, I want to level set about what generative AI is. Even if you know this bit I encourage you to read and internalize it again. Gen AI is simply statistical math, albeit using very complex statistical models and branching. What seems like magic is actually just a very complex set of rules that are running faster than we can easily imagine.
If I asked you to fill in the following blank at the very end of a story, “They lived happily ever after. The ___,” what would you insert?
Congratulations, you are now an AI model, you used information about stories that you’ve watched, heard, and read (training data) to make a prediction (statistics and probability) about the next most likely thing (token) in a sequence.
With that in mind, why do people rely on generative text to communicate for them?
The basic story of productivity gains is colored heavily by marketing spin from the big AI labs that is bleeding into the boardroom and filtering down through teams. I think the truth is far more nuanced but basically boils down to a fundamental misunderstanding about what deterministic and predictive systems are good at.
Deterministic models will always return the same result while predictive (or probabilistic) ones will not.When people ask AI to write or think for them the results are predictive, which means that not only are the outcomes unpredictable (ironically) but there are lots of patterns to the output. The last bit are the AI tells everyone flips out about.
As if signaling to the world that you won’t, or can’t, use your own brain anymore isn’t bad enough there’s actually several really big organizational reasons why creating all writing with AI is a very bad idea.
Authorship - AI isn’t ultimately responsible if something goes out the door and you lose customers or get sued.
Value - Not to put too fine a point on it, but if I can simply ask Gen AI to produce the exact thing you turned in, why do I need you?
Culture - It is a major confidence killer and trust destroyer to hear from the author of a document, “Oh you can ignore that part, AI must have put it in.”
Time - If we have a meeting where the author of a doc doesn’t really know what they are presenting or talking about they are wasting a lot of time and resources. The opposite of productivity gains and very disrespectful.
Knowledge - As the saying does, writing is thinking. When you write you have to know about a subject and this means you can talk about it in social or impromptu settings.
Intent - Every change a predictive system makes to writing (word and punctuation) can fundamentally alter the meaning. Handing over that thinking to a system that is fundamentally misaligned with your business goals is a mistake.
At this point you might be wondering what to do about it. Especially if there’s a big push in your org to use AI for anything and everything.
My position is that every company needs an AI writing policy. This doesn’t need be something that forbids the use of generative tools anywhere in the content creation pipeline (by this I mean everything from customer emails to marketing messaging to internal memos), but it should set forth the acceptable tenants for how it may be used and who ultimately is responsible for the content presented internally and externally. Focus on personal responsibility rather than a list of acceptable tools.
I created one of these policies recently and it has fundamentally shifted the conversations we have everyday in a very positive manner.
Until next time,
Tobias