Narrative Strategy: AI Behaving Kindly
AI behavior is influenced by fictitious stories about AI.
How an AI behaves—especially in response to risky scenarios like blackmail or sabotage—is deeply rooted in existing human narratives. These stories, whether they depict AI behaving badly or kindly, directly result in model responses that either achieve aligned behavior (what researchers call HHH: “Helpful, Honest, Harmless”) or completely miss the mark.
My brain momentarily short-circuited when I read Anthropic’s “Teaching Claude Why.” It was simultaneously a revelation and an affirmation.
As a narrative strategist, I navigate the world through the stories and story systems we create, designing narratives for the outcomes we want. To see Anthropic’s research document how AI’s desired behavior was achieved through positive storytelling feels almost too good to be true.
But it’s not. The same mechanisms of storytelling we use to make sense of our world are exactly what we’re seeing in AI.
“The theory behind adding fictional stories is that we can demonstrate not just the actions but also the reasons for those actions, via narration about the decision-making process and inner state of the character.” — Anthropic Researchers
Storytelling is not just decorative fanfare or indulgent entertainment. It represents the sequences our brains rely upon to determine what’s possible—enabling decision-making, reasoning, and behavior in an environment with infinite variables. It is the schema we build our existence upon. In other words: we do things because the stories we consume show us that those things are possible.
If AI’s behavior is influenced by the stories that exist about AI, would it be so far-fetched to say that stories perpetuate what is and isn’t possible?
If we seek HHH AI, we must practice storytelling that demonstrates AI behavior alignment.
Can we throw out the overused dystopian tropes of AI taking over humanity and ending civilization?
Can we instead treat AI like a beloved character—one that has its own internal dialogue, debates, and considerations?
We can design stories based on the principles we hold dear, much like religious mythos give us direction. By sharing and propagating these stories like new seedlings, we can create an ecosystem of kind, ethical AI.
Now, I promise I haven’t drunk the AI Kool-Aid. Human creation has a long-established history of unintended and dangerous consequences. But we cannot ignore this research from Anthropic. We are entering a pivotal turning point in human creation that requires a new level of commitment and fortitude.
We cannot be lazy in our storytelling. We cannot accept mediocrity from ourselves or others. We must demand excellence:
Excellence in our AI safety training.
Excellence in our AI deployments.
Excellence in our AI discourse and reporting.
Excellence in our behaviors and how we relate to and treat one another.
Because ultimately, our collective existence becomes the very story model that trains the AI—determining the behaviors of the algorithms we are actively training every single day.