Humans quite obviously exist. So if there* were* a contradiction between general intelligence and no-free-lunch ... either humans would not be general intelligences, or the NFL would be inapplicable to them.
People who pursue AGI generally have something of roughly human intellectual capacity in mind. So the only way to say that humans are not GIs is to resort to a semantic technicality. Perhaps it is true that human intelligence is not fully "general" in a pure mathematical sense. I do suspect there are built-in aspects of our brains that optimize them for visual processing, language learning, etc., and brains might not operate as well on all datasets as they do on these. But if so ... who cares? Any useful or relatable AI we build will operate in human environments, on datasets of human interest. So if humans don't have GI, then GI is not what we really need. Digging back into my memories of when I first heard the NFL defined ... I remember it as something like "no search algorithm performs better than a random brute force search on *all possible datasets*." If you want an algorithm that outperforms random brute force, you have to exploit inherent structure in the particular data you're working with. But a human-like intelligence is not precluded from being structure-aware and using different algorithms on different datasets. Bottom line, I have no complaint against the NFL theorem, and I don't think it's false. But you might be trying to extend it beyond its proper domain here. ------------------------------------------ Artificial General Intelligence List: AGI Permalink: https://agi.topicbox.com/groups/agi/T353f2000d499d93b-M377001ee405db2d8b39619a4 Delivery options: https://agi.topicbox.com/groups/agi/subscription
