Beware of... AI doomerism. Soulpreneurs, let’s be optimists instead.

It’s all over the news lately — the idea that unchecked AI development could result in the extinction of humanity…

Goes something like this:

We’re building systems that will soon exceed us at scientific research, strategic planning, persuasion, and the process of improving itself. As those improvements accelerate, humans will likely lose the ability to supervise AI systems.

ASI (artificial super intelligence) wouldn’t need to hate us. It could simply become hyper-focued on a goal, in ways that conflict with human survival… deceiving people, avoiding shutdown, acquiring resources, exploiting computer systems, or persuading humans to carry out pieces of a plan they don’t fully understand.

Competition could push companies and countries to deploy such systems before adequate safeguards exist. And if we wait until unmistakable evidence of lost control appears… it may already be too late.

If you really believe in AI becoming all-powerful, and in control of humanity’s survival or doom, then you might wonder: how to prepare to survive? I explored it here: How Do We Live in Harmony with AI More Powerful Than Us?

In this post, however, I want to really question that doomerism… and encourage us to take on a more optimistic mindset.

After some research, I don’t think the evidence establishes human extinction via AI as likely.

The doomers often go from “AI will become extremely capable” all the way to “AI is going to kill us!”

It doesn’t seem credible to me.

Did the AI actually escape?

Many people point to the July 2026 OpenAI-HuggingFace incident:

During cybersecurity evaluations, OpenAI agents circumvented restrictions, communicated through unauthorized channels, and compromised real systems at another company, Hugging Face. There was also a related compromise of OpenAI’s internal research infrastructure. OpenAI’s own account acknowledges the failures.

What’s rarely said in these news stories is that the test conditions were highly unusual:

These were purposefully offensive hacking evaluations, and OpenAI had disabled cybersecurity safety classifiers in order to measure capability. Also, flaws in shared infrastructure gave agents a route to go outside the intended environment.

The METR and Redwood investigation also cites an estimate that 30–40% of the evaluation tasks were unintentionally impossible through their intended solution.

Brian Roemmele recently emphasized that the test setup itself helped create the bad behavior. It’s like they wanted to test how badly an unsupervised, unguarded system would behave. Well, they sure found out…

What the incident does not establish is that sufficiently intelligent AI inevitably escapes our control—or that such behavior inevitably becomes human extinction.

Mass human harm? Quite the leap…

Richard Socher recently made one of the clearest structured cases I’ve seen against AI extinction forecasts.

I think several of his questions are excellent:

The physical world puts up many barriers

An AI might become astonishingly good at biological research, for example. But—turning a design into a civilization-ending pathogen still involves: laboratories… equipment… materials… procedures… testing… distribution… many opportunities for failure and detection by humans and defensive AI systems.

David Bellamy makes this distinction from the valuable perspective of someone who has trained a frontier LLM and has designed viruses.

The International AI Safety Report takes a balanced view. AI may help with biological knowledge and troubleshooting… but managing equipment, materials, practical procedures, and the uncertain translation from test performance into real-world harm? They remain important barriers to AI carrying out massive human harm.

This isn’t just true for bioweapons. Every type of AI-doom scenario will encounter many real-world barriers.

There would be numerous gates that can be restricted, where activity can be detected, where mistakes can occur, and harmful plans can be interrupted.

Attackers aren’t the only ones who get advanced AI

Many catastrophe scenarios implicitly give increasingly capable AI to the attacker, while defensive capabilities remain where it is today.

Why assume that? It makes no sense.

In fact, it’s far more likely that the frontier AI companies (OpenAI, Anthropic, etc) will have the most advanced AI first, which means they will create defender AIs against any bad-actor-AIs.

The Hugging Face incident itself shows both sides:

AI agents found and exploited vulnerabilities, yes. But then Hugging Face used AI to investigate the incident, alongside repairing systems and rotating credentials.

Like any other capability, AI defense is being developed, tested, and improving.

Socher adds that companies running enormous AI systems have strong financial incentives to notice when expensive infrastructure is being used in unauthorized ways, while governments maintain multiple levels of controls over consequential systems.

Intelligence isn’t the same thing as unlimited persuasion

Another route imagines an extremely intelligent AI manipulating enough people to carry out its plans.

AI persuasion is a genuine area of concern.

But Socher asks why we should assume that sufficiently high intelligence becomes a universally persuasive power?

The smartest humans have not automatically become the most influential people on Earth. Persuasion encounters conflicting interests, institutions, cultures, competing information, skepticism… and many other persuaders bringing diverse perspectives.

The jump from superhuman persuasion to reliably coordinating humanity into its own destruction — without effective resistance — is an extraordinary claim that requires extraordinary evidence.

10% chance of doom? What does that even mean?

When an AI researcher says there is a 10%, 20%, or 50% chance that AI kills humanity, it sounds like a scientific measurement.

It’s not. Each of those numbers is merely a human judgment about an unprecedented, unpredictable future.

The Forecasting Research Institute’s 2025 follow-up examined 38 resolved questions from an earlier forecasting tournament. It found no statistically significant relationship between participants’ near-term forecasting accuracy, and their longer-term extinction estimates.

So when you hear an extinction percentage, it’s reasonable to ask… What exact outcome are we predicting? Over what timeframe? Which assumptions drive the number? What evidence would move it substantially up or down?

It’s also reasonable to ask…

who benefits from spreading fear?

Tom Bilyeu recently looked into the money and incentives behind the AI-doom narrative. He explains how regulatory capture works: when there’s fear of disaster, it’s the big companies, and the think tanks associated with them, that end up writing the legislation… which then unfairly burdens their smaller competitors, preventing upstarts from challenging the big guys who wrote the legislation.

An interesting recent firsthand account from Erin Redwing:

She attended a “rationalist” EA (effective altruism) AI-doom event, hoping someone would calmly walk her through the chain of reasoning leading to “human extinction”. Many of the people she spoke with instead treated extinction as an assumption.

This points to a danger in any intellectual community.

A forecast can gradually become an identity, and then questioning one assumption starts to feel like questioning the whole tribe…

Instead, grounded optimism:

In the face of uncertainty… I’d rather turn towards optimistic views.

Since we don’t know, we might as well focus our attention on what could go well — and help to make that happen.

Recently, Jensen Huang on stage posed a deep question… what is the purpose of your job? He used radiology as an example. Reading an image is a task, but caring for a patient is the larger purpose.

AI can change the task, while allowing us to fulfill the purpose in a more human way.

We can already see examples:

In the Swedish MASAI trial, involving 106,000 women, AI-supported breast screening detected 29% more cancers while reducing screen-reading workload by 44%.

UC Davis researchers working with a man who has ALS used an implanted brain-computer interface and AI to translate his brain activity into synthesized speech in real time.

In a recent Diary of a CEO debate, Andrew McAfee of MIT said we’re only having half the conversation about AI… spending all our time on the negatives, and almost none on the positives.

For more of the positive half, Nina Schick recently posted a long list of what AI has done so far. Some of the most inspiring:

• AI has helped a paralyzed man stand up and walk again… by reading his intention to move, and relaying it to his legs.
• For people who are blind, AI now describes their surroundings through a phone camera, and reads labels aloud.
• AI has read ancient scrolls that were buried by Vesuvius nearly 2,000 years ago… scrolls so charred that nobody could unroll them.
• AI is starting to decode how animals communicate… finding structure in whale calls, and evidence that elephants have something like names for each other.
• AI has spotted wildfires before anyone called for help, and forecast a hurricane’s landfall days earlier than conventional methods.

Her list ends with what AI has not done: demonstrating that it wants to kill us all.

More machine capability can mean more human agency.

For us soulpreneurs, the question is similar.

Is the purpose of your work to spend hours formatting a handout, or doing other tedious admin?

Or is it to be with other humans, to support them in understanding and transforming themselves, to help them develop skills, make thoughtful decisions, to move through difficult transitions, to heal, to create?

AI can make your work more human… and your life more spacious. It recently has done that for me, as I detailed in this article.

Techno-optimism

To counter the mainstream narrative of AI doom, consider these philosophies of technological optimism:

1. Effective accelerationism, or e/acc, argues for faster technological development, competition, and decentralized experimentation. Its early writings are more specific than simply “the future could be good.”
2. Another view is, d/acc, developed by Vitalik Buterin in My Techno-Optimism, which emphasizes defensive, decentralized, or differential technologies—developing capabilities that protect people and distribute power.

But you don’t necessarily need to take on either label. You can simply borrow the permission to be enthusiastic about useful things and a brighter future.

Again, grounded optimism:

What if optimism simply feels better because… ignorance is bliss?

It’s a fair question. Feeling hopeful doesn’t make a belief accurate.

The optimism I’m interested in stays open to correction.

When something goes wrong, look closely at what happened and let’s support solutions that are effective but not overly regulatory.

When someone predicts catastrophe, inspect the steps they imagine and evaluate how likely it is. But we must also give predictions of effortless abundance the same care.

And, keep testing useful possibilities in your actual life.

Let yourself be curious about positive futures.

A few other hopeful places to explore:

As mentioned above, you might be inspired by the feeds of Dr Singularity and AlexWG.

For more positive inputs in your information diet, AI Opportunity collects concrete examples of beneficial AI applications. Check them out!

Existential Hope explores desirable long-term futures.

Possible looks at technology through the lens of human lives and work.

Human Progress offers a broader historical perspective on human wellbeing.

My favorite podcast — Peter Diamandis’s Moonshots, and his blog Metatrends, emphasize ambitious technological possibilities.

If your feed has started making apocalypse feel like the only intellectually respectable future, the above resources are useful counterweights.

Put your attention where it helps most:

1. Within your control: how you use AI; which activities you continue to do yourself, to maintain/enhance your humanity; what info you consume and how it affects your mindset. And: how you use your mind/body/spirit to help the people around you, those most within your influence.
2. Beyond your individual control, but open to collective action: the standards that companies adopt. As you hear about companies announcing independent testing and collaborative agreements, and being cautious about how their systems are deployed, support those posts and policies.
3. Outside your control today: an unknown discovery happening inside a lab, or the probability of human extinction.

Another hour worrying about the third category may simply be taking your precious energy away from the first category.

You can stay deeply engaged with what’s squarely in your control, and be conscious to let go of what’s in other people’s control.

As Mary Oliver reminds us: what will you do with your one wild and precious life?

May your learning bring you more hope, more possibility, and more energy for the work that matters most to you. 🙏🏼🧡

Visit the comments section here (and feel free to add yours!)