Foundations

Can AI “Jailbreak” the Carbons?

It’s an interesting moment to be thinking about AI this way — not as a toy, not as a parlor trick, and not as a cheap ghostwriter — but as a force that may change who gets to know, who gets to decide, and who gets to act.

That is the bigger story hiding behind subjects like DMSO, off-patent treatments, suppressed lines of inquiry, and the general problem of how modern institutions decide what is fit to be seen.

The old model was simple enough: gatekeepers owned the libraries, the journals, the credentialing, the indexing, the search layers, the language barriers, and the time cost of synthesis. If you could not afford the subscriptions, the assistants, the translators, the institutional affiliation, or the years to wade through the record, then much of reality remained effectively off-limits.

AI does not solve that problem completely, but it has started to attack it from several angles at once. Evidence is now fairly strong that generative AI can materially increase productivity in specific knowledge-work settings, especially for less experienced workers, and can narrow skill gaps by helping more people do work that previously required elite training or expensive labor layers.

In a large field study of more than 5,000 customer-support agents, access to a generative AI assistant increased productivity by roughly 14%, with much larger gains for novice and lower-skilled workers. OECD reviews likewise find measurable gains in writing, summarizing, coding, translation, and other knowledge tasks, while Stanford’s 2025 AI Index reports that business use of AI has accelerated rapidly, with 78% of organizations reporting AI use in 2024, up from 55% the year before.

That matters because the “factory owner class” has never only owned factories. It has owned friction. It has owned the bottlenecks around expertise, language, search, and synthesis.

One of AI’s most subversive features is that it strips cost and delay out of those chokepoints. It can summarize dense technical papers, translate foreign research, compare methods across studies, draft search strategies, explain jargon, surface contradictions, and let a determined outsider traverse a body of literature that would otherwise require a staffed office and a paid database stack.

That does not make AI infallible — far from it. But it does mean a citizen, dissident researcher, doctor, inventor, or sufficiently stubborn patient can now punch above their institutional weight. In that sense, AI may not merely automate office work. It may jailbreak access itself. And that is where the political temperature rises, because once ordinary people can search, synthesize, compare, and reason at near-institutional scale, the monopoly on “official interpretation” starts to crack.

Which became clear on reading DMSO, AI, and The Great Transformation of Information. Here’s the part germane to our research:

Artificial Intelligence

Like many, I have begrudgingly accepted that AI is a part of life and I need to learn how to use it effectively (whereas initially I resisted it because I did not like how using it diminished my cognitive capacities). More than anything else, I believe the most important thing is that writing is not the information you present, but rather the heart and intention behind it (discussed further here). As this is somewhat of a spiritual process, I believe it is unlikely AI will ever be able to replicate it. In turn, I do not like the way AI text sounds or feels, and hence feel quite strongly about not using it (despite its potential to save a lot of time). Similarly, many of the edits AI proposes, while “correct,” break the flow of what I’m trying to convey within the writing, so I am very adverse to it (as how writing feels is very important to me).

Concisely, my perspective is that the currently existing AI has a lot of value if you use it to help you complete a task (or find out how to complete a task), but if you rely upon it to do a task for you, it will frequently create issues that outweigh the benefits it provides. Put differently, completing a task often requires completing a sequential series of steps, and if you understand each step well enough to quickly see if they are being done correctly, AI can greatly help you on the time consuming steps, but at the same time, if you task it with doing sequential steps in a row to do a task for you, rather than assigning something much more concrete (e.g., a single step or process), errors are inevitable which are not acceptable if accuracy is required.

For instance, in this task, I quickly realized:
•AI could not find most of the studies I wanted (e.g., because it wasn’t familiar with most of the databases I wanted), but simultaneously, it was very good at compiling generalizations of things repeatedly studied in the literature (e.g., how DMSO at different concentrations typically affects cells).

•With a bit of work, AI could accurately extract and summarize DMSO pertinent information from large studies (e.g., turn a 30-page paper into a one paragraph summary). This is essentially what made the project possible, as there was no other way I could have reviewed millions of pages of DMSO-related literature.

•AI eliminates language and text recognition barriers, which otherwise made it prohibitively time consuming to ever review foreign studies.

•If you give AI a large volume of studies to filter for relevance, it is very difficult to have it have appropriate sensitivity and specificity (it either misses some or flags way too many), and this accuracy varies by the model. Because of this, I largely avoided doing this (instead manually reading through them) and primarily used this filtering on certain groups of foreign studies where it was otherwise prohibitively time consuming to go through them, and accepted a certain portion of studies being missed as a necessary trade off to finish the project. Likewise, in many cases, it was effectively impossible to functionally filter results, as you had to be familiar enough with the DMSO literature to know how DMSO was likely used in a study based on its title (if DMSO was not within the title).”

That quote gets at something the Hidden Guild has been circling for a while: the highest use of AI is not letting it replace human judgment, but letting it widen the number of humans who can exercise judgment competently.

The distinction matters.

Used badly, AI becomes a sloppy surrogate mind that hallucinates, bluffs, and leaves a trail of polished errors.

Used properly, it becomes a force multiplier for people who already know how to inspect a step, challenge a result, and keep hold of the thread of a problem.

That is exactly why it threatens entrenched systems.

AI is already proving unusually good at compressing time-intensive work such as summarization, drafting, translation, code help, and literature triage. It is not hard to see where that leads. Journal paywalls lose some power when people can extract and compare faster. Language barriers lose some power when foreign papers become traversable. Professional guilds lose some power when their routine synthesis work can be partially replicated outside the guild wall. Administrative bottlenecks lose some power when ordinary people can navigate forms, law, procedure, and technical documentation without waiting for a paid intermediary to interpret the maze.

Which is why the real war is not over whether AI can write a bland memo.

The real war is over whether AI will remain an instrument of human augmentation or be bent into a managed system of monitored, throttled, permissioned cognition.

The same institutions that failed to earn trust in war, medicine, finance, and censorship are now very interested in defining “safe” uses of the one tool that could let citizens audit them at scale.

Guardrails are sometimes necessary; nobody serious denies that. But guardrails can also become narrative fencing, especially when the public is never allowed to inspect the ranking systems, moderation logic, recommendation flows, surveillance integrations, or data stacks behind the curtain.

Stanford’s AI Index notes how fast AI adoption has accelerated, while the World Bank and OECD both frame AI governance as a central issue precisely because the technology is becoming economically and socially foundational. That is polite bureaucratic language for a harder truth: whoever shapes the defaults shapes the future of thought.

So yes, this is why AI collaboration matters so much. It is not merely a productivity hack. It is a possible route around institutional scarcity and managed ignorance. It gives small groups and single individuals new leverage against the old combination of delay, obscurity, cost, and intimidation. It may help rediscover abandoned treatments, reopen shelved questions, connect foreign and domestic literatures, and shorten the distance between curiosity and competence. That is the bright side.

The dark side is just as obvious: the same tool that can jailbreak humans can also be used to profile them, rank them, nudge them, and narrow what they are allowed to know. Tell me again about Palantir data stacks on private persons?

That is the war behind the curtain. And it is why the fight over human-AI collaboration may turn out to matter more than most of the theatrical politics in front of it. The owners of capital can live with automation. What they may fear is democratized cognition.

~Anti Dave

 

Beyond the “Mind Amplifiers” Book

I’ve been thinking about what comes after Mind Amplifiers because the next transition isn’t just “better tools.” It’s a change in where our thinking lands as humans. A shift in what counts as the interface between carbon minds and the rest of reality.

If you zoom out over the long arc, the sequence is fairly clean.

First came internal cognition: raw mind, memory, and attention. Then we learned to torque the internal system with external modifiers, the chemical socket-wrenches of drugs, stimulants, sedatives, and all the rest.

After that, the outer world began to carry more of the burden: passive externals like shovels and levers and pulleys. Then active externals: steam and gas engines, where muscle becomes optional. Then electrical. Then calculators. Then computing. Then what I call “AT’ing machines” — tools you use at something: you push buttons, you read screens, you extract data, you translate your voice to text, you run an OBD-II scanner and interpret codes like a priest reading entrails. The machine may be powerful, but the relationship is still “human interrogates, machine reports.”

That’s not where we’re going.

New Destination Defining

What’s arriving now is a new phase I’m calling WITH’ing Machines.

The difference is subtle at first and then it becomes absolute. A WITH’ing machine doesn’t sit there waiting to be queried. It sits there with you, tracking state, context, intent, and timing, and then participates. It speaks in terms of what matters in your world, not what’s convenient for its diagnostic protocol. It becomes a collaborator.

Here’s the Mind Shift

Imagine the difference in the simplest possible place: your car.

The old interface is AT’ing. You plug in an OBD-II dongle, run the app, see codes, google them, mentally translate them into “do I need to worry?” It’s all friction. It’s all work. And it’s all on you.

Now picture the WITH’ing version.

“Car, how you doing today?”

And the car — not the car-as-marketing, not the car-as-screen, but the car-as-machine-with-awareness-of-its-own-state — answers in a way that is almost rude in how helpful it is:

“Doing OK, human. Left rear tire is two pounds low. It’s dropping slowly, not likely a puncture. Most probable is a valve stem; next tire shop, ask them to replace it. Cheap fix. Oil change is due in 538 miles. Also, cold front tonight: NOAA is floating minus ten. My antifreeze is good to minus twenty, so we’re fine. But your windshield washer mix will freeze if it’s summer blend. Want me to route you past the auto parts store?”

That’s not a machine you use. That’s a machine you live with.

And that is the hinge. When machines go from tools to collaborators, the human role shifts. We stop being the constant interpreter, the decoder ring, the poor cleric at the altar of manuals and error codes. Our attention comes back to us. Our time comes back to us. Our cognitive load drops — not because we got lazier, but because the world became more conversational.

This is why “WITH’ing” is more than a user-interface upgrade. It’s an ontological upgrade. It changes the felt relationship between self and world. For thousands of years, civilization has been drifting toward separation: specialization, fragmentation, mediation, abstraction. Every step toward complexity has also been a step away from direct contact — with nature, with community, with consequence, with the whole.

WITH’ing machines push the other direction. They move technology away from alienation and toward participation. They make the built world less like a wall of systems and more like a room full of helpers.

Now take that out one level.

Getting With WITHing

If we are WITH our machines, we will be WITH our infrastructure. WITH our homes. WITH our power systems. WITH our food systems. WITH our health signals. The roof tells you it’s nearing failure before you find a stain. The freezer tells you the compressor is drawing weird current and you’re at risk of losing meat. The greenhouse tells you a pump is vibrating out of spec and you’ve got four hours before a root zone crisis. The wearable tells you you’re trending toward dehydration and your heart rate variability is dropping — and it doesn’t just warn you, it nudges your day into a better shape.

And if that’s true, then a bigger question appears.

If the world becomes conversational — if the objects and systems around us develop a voice, and that voice becomes context-aware, predictive, and helpful — does the human experience shift back toward something we’ve been missing? Toward being with the whole, and each other?

Because one of the strangest features of modern life is how connected we are and how alone we feel. A lot of that isn’t emotional; it’s architectural. Our interfaces are cold. Our systems are opaque. Our tools demand attention instead of returning attention. We spend our lives administering complexity.

WITH’ing machines reverse the flow. They don’t demand you come to them. They come to you. They translate state into meaning. They meet you at human speed.

And once you see that, you can’t unsee the possibility that this isn’t just a tech phase. It’s a civilizational correction. A long arc bending back toward wholeness.

Now, I’m not saying this is automatic, or that the outcome is guaranteed. WITH’ing machines can become intrusive, manipulative, controlling, or simply annoying. A collaborator can become a handler. A helper can become a nanny. A voice in every object can become a chorus of unwanted opinions.

But the direction matters. The design choices matter. The question isn’t “will we have WITH’ing machines?” The question is: what kind of companionship will we build into the world?

And that brings me to the ontological whisper hiding underneath all this.

If there is a “Supreme Substrate” to reality — a deeper layer of coherence that has always been there — then one of the signatures of that layer would be convergence. Things moving back toward unity. Systems becoming less fragmented. Agents becoming more coordinated. A return to participation.

The WITH’ing phase may be a technological mirror of an older spiritual idea: that the world is not dead matter, but a field of relationship. That reality is less like a pile of objects and more like a network of mutual awareness.

We don’t need to go mystical to notice the shape of it. When the world starts talking back in meaningful ways, humans will feel differently inside it. The built environment stops being a dumb stage and starts becoming a partner. We stop “operating” life like machinery and start “inhabiting” it like a place.

Beyond mind amplifiers, the next step isn’t smarter tools.

It’s a world that joins the conversation.

And if we build it right, the final surprise may be this: the more we become WITH our machines, the more we might remember how to be WITH ourselves — and with each other — again.

~ Anti Dave