The recent Anthropic announcement is a classic case. Take a machine metaphor for consciousness (Global Workspace Theory), then match the machine behavior to the machine model and presume that you have evidence of consciousness.
It seems that we are in an era that demands a William James or a Dewey, and we get Habermases, deeply institutional thinking that ignores difference under the guise of reasonableness.
As an aside, I am a carpenter, and the best ships and structures from the days of timber construction (or from restoration work today) use crooked timber. A shipwright will often select trees that are bent in the precise way that a frame (rib) is curved, because the lines of grain will directly support the lines of force.
Modern governance seems to be the attempt to force straightness on everything. It is no wonder so many people feel they don't fit.
+ the counterpart, the King’s Masts from early U.S. colonial times. All those massive old-growth pines are gone now. Straight Timber has a place (ha, feels like a protocol)
I was just thinking about this. Straightness as much as hardness is a necessary ingredient for good protocols. A lot of protocoling is subtle knolling.
I was thinking along similar lines, but mapping it onto lumber mills and the AI infrastructure build-out.
Today, crooked timber only survives in expensive bespoke work. The vast majority gets debarked, cut, milled, and planed to near-straightness thanks to electrical power and mechanization.
If AI infrastructure is doing the same thing to the underlying token vectors — milling crooked timber into straight stock — two questions follow. First: what structures become buildable once the raw material comes this processed? And second: what are the use cases for the “sawdust and wood chips” the milling process throws off?
I met a guy who was restoring the Mayflower II in Mystic (a replica built as a gift to the US). He had relationships with all the CT highway departments to get right of refusal on fallen trees. He just waited until an oak with the right curve happens to fall.
Even better, when Robert and James Adam built a curved gallery in a manor house, they found curved limbs on the property to create the floorboards.
Similar metaphor - AI is a field of marble boulders and each of us a Michelangelo seeking to free sculptures from the boulders? A difference from crooked timber is that, in the example of the shipbuilder below, we know what we are looking for/how we might use it. Michelangelo first saw teh sculpture (actually "angel") and only then worked to set it free.
This reminds me of how science usually works before it finds its big theory: you need a pile of weird, unexplained observations first, not just clever reasoning from a desk. If Rao's right, AI is still in that messy "collecting weird observations" stage. Which means every "AGI" framework people are pitching right now is probably too early; we don't have enough data yet to know what the real pattern is.
Interesting article but overly verbose. Took me ChatGPT to summarize it:
A much tighter version of the thesis would be:
AI reflects the complexity and conflict of the human data that produced it. Therefore, its governance should be empirical, context-specific, incremental, and continuously revised—not based on the fantasy of a universal, permanent solution.
That is probably one tenth of the essay’s length and captures most of its useful content.
Crooked timber feels like a useful corrective to the usual AI metaphors of engines, optimisation and inevitability. The systems are powerful partly because they’re irregular, improvised and difficult to reduce to one clean mental model.
I love this. But with regard to the shade thrown on Thinking Machines' latest: `It is pluralist but only in an impoverished analytic way, relying on a platonic notion of “decentralized alignment” for instance, rather than grappling with the crookedness of the timber, and the difficulties that might be involved in crafting particular pieces of furniture out of particular pieces of crooked timber.`
This feels to me like the narcissism of minor differences. There is broad alignment between the views advanced there and the views advanced here, and I see Thinking Machines as more aligned than not what with what you're arguing for here. There isn't space to delve into every detail that is relevant to wood craft in a manifesto.
The recent Anthropic announcement is a classic case. Take a machine metaphor for consciousness (Global Workspace Theory), then match the machine behavior to the machine model and presume that you have evidence of consciousness.
It seems that we are in an era that demands a William James or a Dewey, and we get Habermases, deeply institutional thinking that ignores difference under the guise of reasonableness.
As an aside, I am a carpenter, and the best ships and structures from the days of timber construction (or from restoration work today) use crooked timber. A shipwright will often select trees that are bent in the precise way that a frame (rib) is curved, because the lines of grain will directly support the lines of force.
Modern governance seems to be the attempt to force straightness on everything. It is no wonder so many people feel they don't fit.
+ the counterpart, the King’s Masts from early U.S. colonial times. All those massive old-growth pines are gone now. Straight Timber has a place (ha, feels like a protocol)
eg https://en.wikipedia.org/wiki/Pine_Tree_Riot
I was just thinking about this. Straightness as much as hardness is a necessary ingredient for good protocols. A lot of protocoling is subtle knolling.
I was thinking along similar lines, but mapping it onto lumber mills and the AI infrastructure build-out.
Today, crooked timber only survives in expensive bespoke work. The vast majority gets debarked, cut, milled, and planed to near-straightness thanks to electrical power and mechanization.
If AI infrastructure is doing the same thing to the underlying token vectors — milling crooked timber into straight stock — two questions follow. First: what structures become buildable once the raw material comes this processed? And second: what are the use cases for the “sawdust and wood chips” the milling process throws off?
That’s an amazing factoid
Thought you would like that.
I met a guy who was restoring the Mayflower II in Mystic (a replica built as a gift to the US). He had relationships with all the CT highway departments to get right of refusal on fallen trees. He just waited until an oak with the right curve happens to fall.
Even better, when Robert and James Adam built a curved gallery in a manor house, they found curved limbs on the property to create the floorboards.
https://www.instagram.com/reel/DTnl4J1jKad/?l=1
Similar metaphor - AI is a field of marble boulders and each of us a Michelangelo seeking to free sculptures from the boulders? A difference from crooked timber is that, in the example of the shipbuilder below, we know what we are looking for/how we might use it. Michelangelo first saw teh sculpture (actually "angel") and only then worked to set it free.
This reminds me of how science usually works before it finds its big theory: you need a pile of weird, unexplained observations first, not just clever reasoning from a desk. If Rao's right, AI is still in that messy "collecting weird observations" stage. Which means every "AGI" framework people are pitching right now is probably too early; we don't have enough data yet to know what the real pattern is.
Interesting article but overly verbose. Took me ChatGPT to summarize it:
A much tighter version of the thesis would be:
AI reflects the complexity and conflict of the human data that produced it. Therefore, its governance should be empirical, context-specific, incremental, and continuously revised—not based on the fantasy of a universal, permanent solution.
That is probably one tenth of the essay’s length and captures most of its useful content.
Crooked timber feels like a useful corrective to the usual AI metaphors of engines, optimisation and inevitability. The systems are powerful partly because they’re irregular, improvised and difficult to reduce to one clean mental model.
I love this. But with regard to the shade thrown on Thinking Machines' latest: `It is pluralist but only in an impoverished analytic way, relying on a platonic notion of “decentralized alignment” for instance, rather than grappling with the crookedness of the timber, and the difficulties that might be involved in crafting particular pieces of furniture out of particular pieces of crooked timber.`
This feels to me like the narcissism of minor differences. There is broad alignment between the views advanced there and the views advanced here, and I see Thinking Machines as more aligned than not what with what you're arguing for here. There isn't space to delve into every detail that is relevant to wood craft in a manifesto.