Thank you for writing this great essay! IMHO the 'singleton' vs. 'ecology' debate is a critical question . This article correctly identifies that real capability and intelligence lives in the knots of relationships between discrete entities, not just the solitary agent. If we treat AI as an incremental addition to familiar workflows, we are merely building faster horses. The real change might be when we redesign coordination itself. Eager to see more from Cosmos Institute on the system dynamics lens
Great read. Couldn't agree more with the concept that "People each hold fragments of truth, most of it tacit and hard to articulate, which is why spontaneous order tends to get the better of best laid plans."
Outstanding framing of the core reasons why group intelligence is what leads to human flourishing and will lead to AGI that works.
The horse metaphor also extends to the difference between "centaurs" (human heads controlling horse bodies) and - what we may be headed toward without more care - "reverse-centaurs" (AI heads controlling human bodies).
It also parallels the reasons why monolithic control of social media feed attention agents fails to serve humans, and why open interoperability (“middleware”) must work not only for individuals, but for the “social mediation ecosystem” of communities that always guided human discourse, but is now being disrupted by the dominant platforms – a cautionary tale for AI (https://www.techpolicy.press/three-pillars-of-human-discourse-and-how-social-media-middleware-can-support-all-three/).
I think as a consequence, it is quite difficult to get good market feedback on projects in this directions as the you're sampling from people who are imagining "more horses" and so you're left in this semi no-mans land without much proper feedback for the phase transition of the car transition hasn't happened yet (or if it has people still haven't noticed it).
*Checks clock*, When can I actually build something?
*Checks clock*, common people, it can't be that hard to see?
... , Whatever let's just build what is feasible right now...
With regards to the section on Aristotle and the many being better than the individual, it is easy enough to note that an individual human is rarely as intelligent as the accumulated cultural intelligence of humanity. Humanity is where the intelligence of the human resides, not in its individual members.
This perspective is exactly what brought me towards smaller models. I might go slightly technical on this, but if you have a base model that is great at reasoning and planning, you can add “domain knowledge” on top of it through fine tuned adapters. This, complemented with various tools, gives rise to an amazing and efficient system that is much more capable than a single larger model.
And why is this? Because while a single large model may consume too many resources, the smaller models require very less of those.
great read. I see past the horse i see the car. I see even further that that! I can see no more starvation, No more greed and hate. I see a world where we stop wasting everything do to greed and use it to make the world great. Not part of it ALL OF IT!
Thank you for writing this great essay! IMHO the 'singleton' vs. 'ecology' debate is a critical question . This article correctly identifies that real capability and intelligence lives in the knots of relationships between discrete entities, not just the solitary agent. If we treat AI as an incremental addition to familiar workflows, we are merely building faster horses. The real change might be when we redesign coordination itself. Eager to see more from Cosmos Institute on the system dynamics lens
Great read. Couldn't agree more with the concept that "People each hold fragments of truth, most of it tacit and hard to articulate, which is why spontaneous order tends to get the better of best laid plans."
Outstanding framing of the core reasons why group intelligence is what leads to human flourishing and will lead to AGI that works.
The horse metaphor also extends to the difference between "centaurs" (human heads controlling horse bodies) and - what we may be headed toward without more care - "reverse-centaurs" (AI heads controlling human bodies).
The same thinking drives some related ideas on how personal AIs can serve individuals (and communities of individuals) and negotiate how they interact with business and institutional Ais (https://www.techpolicy.press/new-perspectives-on-ai-agentiality-and-democracy-whom-does-it-serve/).
It also parallels the reasons why monolithic control of social media feed attention agents fails to serve humans, and why open interoperability (“middleware”) must work not only for individuals, but for the “social mediation ecosystem” of communities that always guided human discourse, but is now being disrupted by the dominant platforms – a cautionary tale for AI (https://www.techpolicy.press/three-pillars-of-human-discourse-and-how-social-media-middleware-can-support-all-three/).
I think as a consequence, it is quite difficult to get good market feedback on projects in this directions as the you're sampling from people who are imagining "more horses" and so you're left in this semi no-mans land without much proper feedback for the phase transition of the car transition hasn't happened yet (or if it has people still haven't noticed it).
*Checks clock*, When can I actually build something?
*Checks clock*, common people, it can't be that hard to see?
... , Whatever let's just build what is feasible right now...
i found this dry
Kudos for mentioning Polanyi. His perspective on "The Tacit Dimension" is sorely missing from today's AI theories.
With regards to the section on Aristotle and the many being better than the individual, it is easy enough to note that an individual human is rarely as intelligent as the accumulated cultural intelligence of humanity. Humanity is where the intelligence of the human resides, not in its individual members.
This perspective is exactly what brought me towards smaller models. I might go slightly technical on this, but if you have a base model that is great at reasoning and planning, you can add “domain knowledge” on top of it through fine tuned adapters. This, complemented with various tools, gives rise to an amazing and efficient system that is much more capable than a single larger model.
And why is this? Because while a single large model may consume too many resources, the smaller models require very less of those.
great read. I see past the horse i see the car. I see even further that that! I can see no more starvation, No more greed and hate. I see a world where we stop wasting everything do to greed and use it to make the world great. Not part of it ALL OF IT!