Playing Possum with AI

AI, fear, and choosing happiness anyway

This post isn’t about the evolution of AI. It is about my own struggle to productively respond to some of the fastest technological changes in human history.

A little background context

Five years ago, AI could barely write a coherent paragraph. Today, it passes the bar exam, generates video indistinguishable from Hollywood cinematography, discovers new drugs, predicts the function of human DNA, and writes most of the code used to build itself.

Ten days ago, a Citrini Research thought experiment imagining America’s 2028 AI meltdown helped trigger IBM’s worst single-day stock drop in 25 years — and on Friday, the Pentagon gave Anthropic a deadline: remove your safeguards against mass surveillance and autonomous weapons, or lose your contract. Anthropic’s CEO, Dario Amodei, refused.

Last month, Amodei wrote a public essay on the existential risks posed by the technology he has helped create. I respect the folks who question whether anything a CEO says publicly is bullshit, but CEOs of companies in the middle of historic wealth creation don’t typically write to the public to warn that their technology could destroy the world. It seems reasonable to me to pay attention when one does.

Crushed like ants

I still remember the way my throat tightened the first time I felt real fear about AI. It happened three years ago. I was seven years into scaling the company I had started in college. We had wrapped up a grueling quarterly planning retreat, and I could feel the dark circles under my eyes. I wanted to laugh and shoot the shit. I felt some relief as we all sat down to dinner.

That is when my Co-Founder, Eric, looked up at the leadership team gathered around the table and said something like:

“We could be crushed like ants by machines that don’t give a damn about us.”

I felt anger creep through my chest. Eric sounded crazy. I could see how his curiosity with early LLMs was pulling him in a new direction. I knew it could hurt what we were working together to build. But underneath the anger, I felt a constriction in my throat that didn’t have anything to do with how he was talking to our leadership team.

“What if he is right?” I thought.

I spent a year trying not to think about it.

Two wolves

There is a popular parable commonly attributed to a Cherokee Chief:

A grandfather says to his grandson, “In life, there are two wolves inside of us, and they are always at battle. One is a good wolf representing hope. The other is a bad wolf representing fear.”

The grandson stops and thinks about it for a second. He then asks: “Grandfather, which one wins?”

The grandfather replies:

“The one you feed.”

Bad wolf

I hate how we have algorithmized the feeding of our fear wolf. To me, the most corrosive force for American democracy in the last hundred years has been the way social media companies have weaponized fear to maximize profits.

Facebook’s internal reports show that they understand that using algorithms to inspire fear and anger is easier and leads to more durable monetization than pulling users towards other emotions. Despite the howl of whistleblower complaints, Meta and many other social media companies continue to pump out the kind of content that traps people in scarcity and fear.

And I see this same fear-stoking playing out with AI. At first, I noticed it in my friends. And then, more embarrassingly, I noticed it in myself.

I realized I was getting this weird glee from warning others of their coming despair. Something about telling my friends about AI doom activated a part of my brain that liked to feel like a Roman Sentinel heroically alerting my people of the coming threat. Yikes.

I’ve learned that any time I feel like I can puff up my chest because I forwarded an email, I probably need to get off my computer and go for a walk.

I understand what Oliver Burkeman meant, writing The Imperfectionist, when he chose not to share the anxiety-inducing AI articles he has been reading “for the same reason he doesn’t go around sneezing on people.”

So why am I still sneezing on you?

Possoming

As much as I don’t like giving more oxygen to fear, I also believe my fear had something to teach me. It just needed to be examined and understood before I could place it back in its proper place.

Most folks in tech are now folding some combination of GPT, Claude, Claude Code, and other agents into everything they do, but for the 70% of Americans who aren’t using AI regularly, I sense that most folks are playing possum.

When fight and flight both seem futile — when the nervous system concludes that the threat is overwhelming and inescapable — the dorsal branch of the vagus nerve takes over. We play dead. And hope the threat goes away.

Many of us are already at our limit. We are doing our best to meet our responsibilities and provide for our families amid political dysfunction and fear.

If installing AI on your computer feels overwhelming, how the hell are you supposed to save humanity from it?

Play with fire

Amodei lucidly unpacks his concerns about existential risk here, and I am not going to rehash them. But I will say I find his logic, on autonomy risk, misuse for destruction, and how AI could help bring about a totalitarian nightmare, thoughtful and compelling. Amodei’s warning reminds me of what Robert Oppenheimer would have tried to do if he had access to WordPress.

But the truth is, if I’d read Amodei’s essay during my possoming period, I might have tuned it out. In the same way that you don’t understand the power and danger of fire by reading books on fire, you can’t understand the power of AI unless you are playing around with it. This is obvious to people who like tinkering with a car or codebase, and much less obvious to people who like to spend all day in the New York Times or the Wall Street Journal.

My period of possoming didn’t end with some terrifying realization or worldview-shattering moment. It ended with wondering how best to turn the bok choy in my fridge into dinner. From noticing I now had the power to fix my grill despite having the mechanical ability of a goldfish. And then, eventually realizing I could use AI to automate all sorts of things that I used to pay people to do for me.

In the same way, you notice how fire accelerates in the wind, I started to notice the almost unbelievable jump in sophistication I was experiencing as models evolved in real time. It was only from that base that I came to believe many of the most important questions for democracy, the economy, and our world will have to do with our response to AI.

Taking action

Eric left the company we had founded together to start Goodfire, an interpretability startup focused on making blackbox AI models safer. He gave up his life in New York and started over in San Francisco. I was hurt and angry with him at first for leaving RippleMatch, but I also deeply respect his conviction and desire to make the world safer.

If we see a fire spreading that could envelop our community, we need to do what we can to ensure it is controlled.

Most of us won’t start AI safety companies, but that doesn’t mean we should keep possoming.

You don’t have to agree with Amodei’s level of concern on existential risk to appreciate how AI is going to transform the economy and automate a lot of what college-educated workers have traditionally done. Just look at what played out at Block last week. Jack Dorsey fired 40% of his workforce AFTER beating earnings and raising guidance for the year.

Dorsey remarked:

“Intelligence tools have changed what it means to build and run a company. A significantly smaller team using the tools we’re building can do more and do it better.”

Block stock surged 24% in after-hours trading. This should terrify you, excite you, or both, but you should feel something.

There are such good questions.

How should wealth be distributed in a world of abundance with less need for labor?

What would a world look like in which our job was less central to our identity?

Who should control such a powerful technology, and with what safeguards?

Do we want the government’s ability to surveil to become absolute?

Good wolf

I am done possoming. I want to play with AI, talk about AI, and vote for leaders who are thinking thoughtfully about AI. And I want the rest of the world to do the same.

But I don’t want to give in to the doom. My fear wolf is fat enough - fed by algorithms, headlines, and the part of the brain that likes to play Roman Sentinel. I’m choosing to put him on a diet.

A few years ago, my uncle Sam showed me Mary Oliver’s Poppies, and it has been one of my favorite poems ever since. In it, she talks about how “the black curved blade” of death is inevitable — but that happiness, done right, is a kind of rebellion “palpable and redemptive.”

AI or no AI, we don’t make it out of life alive, but we get to choose the way we live. My marriage, my friendships, and my best work have all come from feeding the right wolf.

Once I finish writing this, I’m going to go play in the snowy woods with my dog. I might even eat chocolate and play Fortnite tonight.