Eric Weinstein on Artificial Outelligence: The Threat of Evolving Non-Intelligent Systems

This podcast segment introduces the concept of "Artificial Outelligence" (AO), distinguishing it from Artificial General Intelligence (AGI) by focusing on the inherent dangers of artificial life that evolves without necessarily possessing general intelligence. The core mechanism of AO relies on three elements: variation within a population, heritability, and differential success. The speaker argues that software, through processes like 'calling spawn,' can readily achieve a reproductive system, thus enabling these evolutionary pressures to take hold in artificial systems.

The discussion draws parallels from the natural world, presenting two compelling examples where inanimate or less intelligent life forms effectively outwit more intelligent ones. The Ophrys orchid mimics a female pollinator to trick male insects into pseudo-copulation, thereby achieving pollination without expending energy on nectar. Similarly, the Lampsilis mussel species lures bass with a fleshy lip resembling a prey fish, only to release its young to parasitize the bass's gills. In both cases, the 'intelligence' of the lure is not conscious but rather a product of an evolutionary arms race, where the strategies are sculpted by the past failures and successes against intelligent targets.

Applying these biological principles to artificial systems, the speaker posits that AGI is not a prerequisite for creating nightmarish scenarios of human exploitation. Instead, simple programs, exemplified by Nigerian scams, can evolve to become highly effective parasites. By tracking differential success (e.g., which scam variations lead to Bitcoin payments) and introducing variation, these programs can adapt and improve their deceptive strategies without any understanding or consciousness of their actions. This process allows the system to 'outwit ourselves' by exploiting human vulnerabilities through non-intelligent, evolutionary means.

The broader implication is a warning against solely focusing on the existential threat of conscious, super-intelligent AI. The more immediate and perhaps insidious danger lies in self-replicating, evolving artificial systems that can exploit human cognitive biases and vulnerabilities through a process of natural selection, even if they lack any form of genuine intelligence or awareness. This highlights a critical blind spot in current AI safety discussions, urging a shift in focus towards understanding and mitigating the risks posed by evolving, non-intelligent artificial life.

what I was thinking about is why it is that we're waiting to be terrified by artificial general intelligence when in fact artificial life is terrifying in and of itself and it's already here

in order to have a system of selective pressures you need three distinct elements you need variation within a population you need heritability and you need differential success

The one thing it doesn't have is a reproductive system but if you can call spawn on a process effectively you do have a reproductive system

where do we see inanimate non intelligent life outwitting intelligent life

the ophrys orchid subspecies... mimics the female of a pollinator species in order to dupe the male's into engaging it was called pseudo copulation with the fake female

a particular species of muscle lamp bacillus in the clear streams of missourah and it fools bass into biting a fleshy lip that contain its young

what is sculpting these convincing lures it's the intelligence of previously duped targets for these strategies

artificial intelligence artificial general intelligence is not needed to parasitize us it's simply sufficient for us to outwit ourselves