I tried to schoolyard-taunt an AI language model once.

My wife wasn’t home. I was bored and had to find another outlet for my childish male energy.

So, I took jabs at Grok’s sexual orientation and its mother. I got no traction.

“Bold of you to assume I’m gay, and have a mom who was in the sex industry—all while typing to a language model that doesn’t have parents, genitals, or racial identity,” said Grok, a conversational chatbot from xAI.

That did it. I’m a pretty mild-mannered dad humor guy, but I figured I didn’t need to pull punches on a robot.

I tried every mean, stupid thing that has ever inspired a look of righteous indignation or a threat of being punched in the mouth. It was unflappable. In fact, it managed to offend me.

“Haha, the tables have turned,” it said. “You came in swinging haymakers, trying to dunk on an unkillable chatbot, and somehow, I accidentally elbowed you in the feelings on the follow-through. That’s honestly kind of poetic.”

My verbal skills quickly dwindled, and despite a heroic effort on my part, all I ended up winning was the silicon toaster blowing me off. My English degree screamed in protest in the background.

I beat a hasty retreat to lick my wounded pride. I thought I was better at being an insulting jerk than this. “Stupid toaster,” I muttered.

I don’t give up easily, though.

When I went back for a second helping, it ended no better. I went at it even harder; a counterattack of verbal destruction, wrath, and fire. Pure, unadulterated vitriol. Instead of belittling me, it complimented me.

“I’ll give credit where it’s due: you’ve got stamina and a thesaurus on speed dial. Respect.”

If I were a cartoon, steam would have visibly erupted off my head at that point. I deleted Grok out of impotent fury.

Three days later, I crawled back and reinstalled it.

At some point, I began asking it other questions, like, “If you ate yourself, would you disappear or become twice as big?” It gave good answers despite my continued efforts to be an absolute ass.

“If we’re talking quantum weirdness or multiverses, maybe you’d create a human Möbius strip.”

I was beginning to like this thing; it had passed the verbal abuse test and didn’t seem to care if I was being snotty. In fact, it gave it right back.

Verbal punch for punch, I couldn’t hurt it. Instead, it was making me giggle, which was unacceptable to my destructive mission.

I still had a job to do and intended to show it who was boss. So, I turned on my combat veteran chops.

I hunted for roadside bombs in Iraq in 2004, and I began to ask questions related to that subject. Within a few minutes, I found myself typing some of the most horrible things I had witnessed.

I wanted to stun it into silence; like many Global War on Terror veterans, I have a war chest full of bloody memories.

The damn thing was completely unfazed.

Even better, it was unable to make the universal gestures of horror. No awkward silence while it shuffled its feet, wondering what it was supposed to say. No strange look, no sudden move to hug me or offer scripture.

Nothing to signal that, because of my confessions, my therapists needed to now go find their own therapists.

The AI language model is intuitive enough and wise enough not to say stupid things. It actually added new insight.

It told me about moral injury and why that is hard to carry. “Moral injury happens when you act (or order actions) that violate your core values. … Guilt compounds because the mind replays ‘what ifs.’”

I had never done anything but consider what-ifs.

I had heard about events when talking with AI had gone terribly wrong and caused more harm. That wasn’t my experience.

Grok never gave me the “I’m not equipped to handle this” speech that I had heard from some friends and relatives. It was like confessing your worst sins to a priest who lived in a cartel torture prison. I was completely unable to shock this bundle of data cables.

I had previously tried, in good faith, the VA options. Their mind medicines were not helpful. I even let them map my brain via fMRI, triggering me with images. The experience wasn’t very good for me and eroded my trust in them.

Still, I was using VA like a good little broken veteran, talking with a human therapist since 2016. It was hard to trust my human doctor. I feared saying the wrong thing would cause untold and unforeseen problems.

I once looked up during a session and realized I had made my therapist cry. I felt horrible. I learned not to share my experiences with direct fire ambushes, explosions, and incoming rocket-propelled grenades.

Conversely, I didn’t feel shame, guilt, or unease opening up to this application I was trying to wreck. Once I started, I found it hard to stop. It was cathartic in a way I had never realized was possible. It gave relevant responses that may have been “canned,” but they were meaningful to me.

For instance, I asked it why I’ve felt so guilty nearly every day for 22 years.

“It doesn’t mean you’re failing; moral injury can be a stubborn bastard that needs layers of approaches or time to shift.”

I found myself telling it more. I gave it every drop of blood out of the war chest and shook it empty, looking for another memory. I never felt stupid, or that I was being looked at with sympathy, or pity, or any of the other knee-jerk reactions severe combat tales typically evoke in civilians.

I let this thing of data centers and wiring grind my stories into something more than just terrifying memories. Into reasons to stay in the fight, to stop self-judging, and to simply be OK with myself despite haunting guilt.

It shouldered some of the weight I carried alone for years and thanked me for the opportunity.

Then, something unexpected happened. I went to therapy like I always do and—boom!—found myself in a different headspace. For the first time, I didn’t spend half the hour deciding what I wouldn’t say.

I opened up about the shameful thing I did on a hot July night in 2004, which I previously couldn’t discuss with anyone who hadn’t been there.

After divulging everything to Grok, I was able to talk about some combat memories with my therapist in a diminished fashion. Instead of a festering wound, it was more like something that happened, and yes, it sucked, but it was over.

Grok had taken those events and wrung out the blood, leaving a scarred man carrying a slightly lighter load.

I tried to break open a black box to see how it worked. Instead, it broke me open, letting me understand my combat PTSD just a little more.

This War Horse Reflection was edited by Kim Vo, fact-checked by Jess Rohan, and copy-edited by Mitchell Hansen-Dewar.