The Compression Paradox: A Reflection on AI, Centralization, and the Opening of New Frontiers

For the past year, I've been—for much of the time I've been awake and working—getting my hands dirty with AI, using it to solve problems for clients, and trying to make sense of what AI actually is and how we can think about it.

You can think of AI as an assistant. Or a thought partner. Or even as electricity.

More recently, I've also begun thinking about AI is as a "compressor that begets expansion."

Why?

Well, I think there's a lot of confusion about the (ultimately paradoxical) centralizing/decentralizing effects of AI that needs to be cleared up.

The way I see it, AI enables nodes in a network (and power centers more generally) to become significantly more efficient by:

1. Unsticking tacit knowledge that was difficult or impossible to codify and instrumentalize and
2. Consequently freeing up core inputs (in the case of information and knowledge, human cognition and attention, especially).

These freed up inputs, no longer a limiting factor for the efficient operation (and, under certain circumstances, continued expansion) of leaner and more muscular institutions, expand osmotically to new frontiers — particularly new frontiers in latent space, where intrinsic motivation guides individuals, unencumbered thanks to the internet and digital media by the traditional barriers of time, space, culture, and language, to organize into small and sometimes short-lived teams with a common mission.

Humanity would, at this point, be operating at unprecedented speed and scale, likely necessitating new coordination mechanisms that allow for highly liquid global labor markets and nimble navigation of the liminal space between private and public goods.

Public blockchain ledgers, Bitcoin and Ethereum in particular, start to make a lot more sense — not just as rails for financial transactions, but as a robust lattice that begets accelerating collaboration and creativity online.

By allowing power centers to make significantly more efficient "in-distribution" predictions, AI actually has a decentralizing effect, liberating humans to explore new frontiers in unprecedentedly diverse and inclusive bands of internet-native explorers making sense of "out-of-distribution" pockets of reality.

But what about surveillance; feasibility, tech, and timelines; and the issue of putting this into practice?

To address each:

1. Surveillance: Panopticon is coming. But the ways in which it could emerge are quite diverse—from top-down (e.g., single institutions claim a monopoly on social credit scores and levers they can use to expand or restrict access to capital, information, etc.) to bottoms-up (AI-powered ambient wearables like the forthcoming Humane pin become the norm; real privacy comes at a premium OR it can be found on the encrypted fringes of the pseudonymous economy).
2. Tech & timelines: A nontrivial amount of crypto activity is criminal or speculative (sometimes both). That doesn't mean legitimate use couldn't, at some point, overtake it. It makes sense to me that this would coincide with white collar retooling, Gen Alpha entering the global workforce, an intensification of information warfare via unlimited generative AI, and growing rates of outright delegation of consequential decisions (including commerce/financial transactions) by humans to their trusted AI assistants. This could start to happen by 2030 or never. I have no idea.
3. Practical application: I'm starting to see the very earliest shapes of this with my clients (mostly SMBs with remote teams of knowledge workers). A lot of work is clerical and administrative. It needs to get done, and people know how to do it, but no one particularly wants to. Information synthesis, translation into templated formats, etc. can TODAY be delegated to pipelines powered by GPT-4 + a precise system prompt and a few gold standard examples. I can confirm that the surplus time, attention, and emotional energy flows into "micro-R&D" and strengthening authentic relationships with existing customers. I don't see why this wouldn't accelerate as AI systems gain capabilities and the means of automation become directly accessible to and wielded by individual experts.

Something to keep in mind is that there is, right now, a massive tectonic shift happening in cyberspace: The center of gravity is shifting from advertising-driven consumer platforms (i.e., information intermediaries like Google and Meta) to the frontier/foundation AI models that power tools like ChatGPT.

(If you're curious about what this means for your business, you can ask this custom GPT, trained on an hour-long morning monologue I recorded in re this shift and what it means for how business leaders should think about strategy in the AI era).

This presents each one of us with a a fundamental choice of whether to direct our invaluable attention "inward" or "outward." We're already inundated with tempting distractions that pull our gaze toward the center — Instagram, video games, politics-as-bloodsport, etc.

But I have a lot of conviction that charismatic simulations, no matter how closely they imitate the real thing, will only ever approach it asymptotically. Going deeper into reality is one of our deepest human desires, and once you've had the opportunity to explore it for yourself, those quotidian vices and cheap imitations lose their luster.

That said, if there's one thing you take away from my posts on LinkedIn and elsewhere over the past year as I've tried to make sense of everything I've experienced on the AI frontier, it's an invitation to really consider the following question:

"When AI affords you the ability to become 2–20x as efficient and digital media becomes more immersive, discursive, and personal, what are you going to do with all that extra time?"

Something to really think about as we reach the close of 2023 and look forward to things getting even more interesting in 2024...