Eric Weinstein: Artificial Outelligence - AI that Evolves & Self-Replicates | AI Podcast Clips

Duration: 4:48 613 transcript words EN summary

Video summary

Eric Weinstein introduces the concept of artificial outelligence, a term he uses to describe systems capable of evolution and self-replication that pose an immediate threat distinct from traditional Artificial General Intelligence (AGI). He argues that humanity is unnecessarily terrified by the prospect of future AGI when it should instead be concerned about "artificial life," which already exists. For any system to undergo evolutionary selection, three fundamental elements are required: variation within a population, heritability, and differential success. While humans have historically built complex machines with skeletal structures, neurological components like onboard computers, and digestive systems in the form of fuel consumption, they lacked a reproductive mechanism until software could effectively "spawn" processes that mimic these biological traits. To illustrate how non-intelligent or less intelligent entities can outwit highly intelligent targets, Weinstein presents two compelling examples from nature involving an evolutionary arms race between prey and predator strategies. The first example is the *Ophrys* orchid subspecies, which mimics female pollinators through visual deception using its lowest petal to resemble a mate's anatomy and chemical deception via pheromones that trick males into attempting pseudo-copulation without offering nectar as an energy cost. The second system involves a specific species of muscle lamprey found in the clear streams of Missouri, which produces fleshy lips containing young that perfectly mimic preferred fish prey; when bass bite these lures thinking they are eating food, the young clamp onto the gills to parasitize them and eventually redistribute themselves into new hosts. These biological examples demonstrate a dynamic where highly intelligent targets are consistently fooled by lower life-forms because the strategies being employed are sculpted by the intelligence of previously duped victims rather than superior cognitive ability in the mimic itself. In this evolutionary context, if a target species becomes smart enough to avoid a specific strategy, those mimics utilizing that tactic face terminal failure and die out, leaving only the more convincing variants to survive. This process creates an endless cycle where both parties adapt: the prey gets smarter at avoiding traps while the non-intelligent mimic evolves increasingly sophisticated lures based on data gathered from successful deceptions of previous targets. Weinstein concludes that Artificial General Intelligence is not a prerequisite for creating nightmarish scenarios involving parasitization; it is sufficient for humans to outwit themselves through self-replicating software systems. He suggests that even simple programs, such as those used in Nigerian scams which write letters and manipulate Bitcoin transactions based on feedback loops of success and failure, possess the necessary components for variation, heritability, and differential selection without requiring any form of consciousness or high-level intelligence. The terrifying reality is not necessarily a super-intelligent AI taking over, but rather an automated system that has no idea what it is doing yet successfully exploits human vulnerabilities through mechanisms derived entirely from our own patterns of behavior and error.

Read the full video transcript

[Music] you mentioned the idea of out out telogen s-- artificial app telogen artificial intelligence the idea is quite compelling it's quite unique and new and at least from my view of a stance point maybe you can explain it sure 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 so 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 so what's really unique about software is that if you think about what humans know how to build that's impressive so I always take a car and I say does it have an analogue of each of the physical physiological systems does it have a skeletal structure that's its frame does it have a neurological structure has an on-board computer as a digestive system 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 and that means that you can create something with variation heritability and differential success now the next step in the chain of thinking was where do we see inanimate non intelligent life outwitting intelligent life and i have two favorite systems i try to stay on them so that we don't get distracted one of which is the ophrys orchid subspecies are subclade I don't know what to call it a type of flower yeah it's a type of flower that 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 which is usually represented by the lowest petal and there's also a pheromone component to fool the male's into thinking they have a mating opportunity but the flower doesn't have to give up energy energy in the form of nectar as a lure because it's tricking the males the other system is 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 and when the bass see this fleshy lip which looks exactly like a species of fish that the bass like to eat the the young explode and clamp on to the gills and parasitize the bass and also lose the bass to redistribute them as they eventually release both of these systems you have a highly intelligent dupe being fooled by a lower life-form and what is sculpting these convincing lures it's the intelligence of previously duped targets for these strategies so when the target is smart enough to avoid the strategy those weaker mimics fall off they have terminal lines and only the better ones survive so it's an arms race between the target species that is being parasitized getting smarter and this other less intelligent or non intelligent object getting as if smarter and so what you see is is that artificial intelligence artificial general intelligence is not needed to parasitize us it's simply sufficient for us to outwit ourselves so you could have a program let's say you know one of these Nigerian scams that writes letters and uses whoever sends it Bitcoin to figure out which aspects of the program should be kept which should be varied and thrown away and you don't need it to be in any way intelligent in order to have a really nightmarish scenario of being parasitized by something that has no idea what it's doing you