I realized what bothers me about @vitalik.eth d/acc essay. Even though it is optimistic, it shares with the pessimist side a presumption that technology, as an evolutionary process, *can* be governed according to some notion of human intent.

“Tech can/cannot be governed” is a bigger divide than optimism/pessimism

Blogger at ribbonfarm.com, running summerofprotocols.com, host of /gloom ☀️

The rare technologies we’ve been able to weakly “govern”, like nuclear have a) highly upstream proliferation choke points (fissile materials) without which knowledge is useless b) no real upside to small-scale/retail use/democratization. I want AI; I don’t really want a backyard or small-biz nuclear plant.

Even apparent counterexamples are weaker than they look:

Nuclear: Proliferation reaches North Korea

Aviation: Drones!!

Pharmaceuticals: Gray markets; India bucking western IP regimes

If you were historically first mover, you can slow things a bit, that’s it

Even apparent counterexamples are weaker than they look:

Nuclear: Proliferation reaches North Korea

Aviation: Drones!!

Pharmaceuticals: Gray markets; India bucking western IP regimes

If you were historically first mover, you can slow things a bit, that’s it 

https://warpcast.com/dwr.eth/0x6084cdcf

I’d say main e/acc axioms are:

1. Knowledge *will* leak to all who can master it, faster than you think

2. They will want to do very different things with it

3. They will not have an interest in use-consensus and 1-few defectors can blow up consensus

4. Innovation will break chokepoints

5. This is all net good

Now with AI, the obvious chokepoint is silicon. If you follow the industry (Semianalyis, China Talk newsletters), it is obvious control has *already* slipped away. EUV machines from ASML aren’t the iron grip chokehold you think they are. China will get past 3nm.

Everything above fab level is bs illusions of control.

I think this is an absolute good thing. Much as I dislike China’s authoritarian surveillance state a) the West isn’t much better b) I’d like a duopoly at least to break out of Western-centric AI monopoly dangerously open to regulatory capture by crackpot doomers

That said, I like the d-for-decentralized d/acc more than e/acc. 

The d-for-defensive d/acc I think is a dangerous impulse at political level. As a normal tech refinement vector (like nuclear safety), “defense” is important and happens as a matter of course. As a pre-emptive political concern it is a boogeyman.

What do I mean by that? 

Normal tech “defense”: FAFO, add safety protocols by trial and error, accept the risk of some loss (including loss of life) along the way.

Political “defense”: Make up boogeymen like “X-risk” that justify pre-emptive installation of massive capture surfaces for evil priesthoods

There is only one “governance” move: pump more $ into an area where you want to accelerate idea rate. Govern that area to have porous, globally accessible boundaries so people can self-select in/out from anywhere in world. Don’t try too hard to prevent discoveries leaking. 

That’s it. That’s “governance”

All efforts to *stop* technology/ideas being discovered and tried, or even crossing into your borders are necessarily local in time/space. They’ll last as long as you control the guns. Or the idea smugglers invent better ways to get around guns.

This is a good thing.

I’ll close with a discussion question: If you *could* govern technology absolutely with a magic remote control that can control global rate (including negative rates to rewind) of evolution of any technology “channel”, *should* you and *would* you? Or would you destroy that magic remote control (paradox alert!)

I think overall I fall on e/acc side because I don’t think meaningful governance is possible without large-scale authoritarianism. 

Culture is highly governable. Trump, Taylor Swift, MCU etc show how it can be done. 

Economics is weakly governable to ungovernable.

Tech is near-pure anarchy.

in my "shitposting from main" phase ||

stuff people usually put in their bios here: mosayeri.com

I think AGI is a bullshit concept to begin with. Like “God”

what do you think about the argument that AGI isn't even technology. 

i'm more on the e/acc side and like d/acc too. 
but let's contemplate this: not thinking of AGI as a 'tool meant to solve human problems' breaks the pmarca paradigm that 'this is just like all other technologies in the past'.

LLMs/AGI is gonna need huge amounts of specialized compute hardware - isnt that a pretty big upstream choke point?