Cephalopods are highly intelligent but also have a very short lifecycle and very large number of offspring—making them a prime target for a artificial breeding program for intelligence uplifting.
E.g. the lifecycle of many octopi is about 1-2 years. A 15 year program could potentially breed for very significant behaviourial changes.
For instance, dog and horse breeds can be made within 10 generations.
Google says the female giant pacific octopis lays between 120k and 400k eggs at the end of her life. Giant pacific octopi live about 3-5 years.
I asked Claude to make some back of the enveloppe calculations
>>>I’ll use the breeder’s equation to estimate the potential response to selection for intelligence in octopuses. The breeder’s equation states:
R = h² × S
Where:
R is the response to selection (change in trait mean per generation)
h² is the narrow-sense heritability of the trait
S is the selection differential (difference between selected parents and population mean)
Let’s make some reasonable assumptions for these parameters in octopuses:
Heritability (h²): For cognitive traits in animals, heritability typically ranges from 0.2 to 0.6. Let’s use h² = 0.4 as a moderate estimate for octopus intelligence.
Selection differential (S): With 120,000-400,000 eggs per female, we could be extremely selective. If we select the top 0.1% of individuals (still giving us 120-400 individuals), we could achieve a selection differential of about 3 standard deviations.
Let’s calculate the response to selection per generation:
R = 0.4 × 3 = 1.2 standard deviations per generation
For a breeding program over 15 years with Giant Pacific Octopuses (3-5 year lifecycle):
Using a 4-year lifecycle: 15 ÷ 4 = 3.75 generations (let’s round to 4 generations)
Total expected gain: 4 × 1.2 = 4.8 standard deviations
For octopuses with a shorter 1-2 year lifecycle:
Using a 1.5-year lifecycle: 15 ÷ 1.5 = 10 generations
Total expected gain: 10 × 1.2 = 12 standard deviations
This represents a substantial shift in the distribution of intelligence. For context, the difference in IQ between average humans and those considered profoundly gifted is about 4 standard deviations. A shift of 12 standard deviations would be extraordinary.
However, several factors would likely limit these theoretical gains:
Selection plateaus as genetic variation is depleted
Pleiotropy (genes affecting multiple traits) may create unfavorable trade-offs
The trait may reach biological/physiological limits
Inbreeding depression could become an issue with intense selection
Even with these limitations, the potential for significant intelligence enhancement in octopuses through selective breeding appears substantial, especially for species with shorter lifecycles.
Alexander reacting to Claude back of enveloppe calculation:
Narrow sense heritability is probably higher for intelligence. Indeed, at this level of selection one would have to worry about inbreeding depression, selection plateaus etc
My best guess is that 6 SD would be possible IF one was actually able to accurately select top 0.1%.
This is probably quite hard. GWAS data for humans currently does not allow for this kind of precision. Doing the selection accurately / i.e. estimating the gradient is the main rate-limiting step [as it is in deep learning!]. One would need to construct psychometrically valid tests for cephalopod, run them at scale.
How much is 6 SD? Octopi might be about as intelligent as a dog (uncertain about this). 6 SD would be quite insane, and would naively plausibly push them tonthe upper end of dolphin/chimpanzee intelligence. The main limitor in my mind is that octopi are not natively social species and are do not have a long enough lifecycle to do significant learning so this might not actually lead to intelligence uplifting. Additionlly, there is the issue that octopi dont have vocal chords so would need to communicate differently.
Ofc like the famous Soviet silver fox breeding program one could separately select for sociability [which might be more important for effective intelligence. Iirc wolfs usually outperform dogs cognitively yet some dog breeds are generally considered more intelligent in a relevant sense.]
The fact that they have a short lifecycle with only 1 lifetime breeding cycle is though. A lot of intelligent animals, like humans, chimps, elephants, dolphins, orcas, have long lives with many breeding cycles and grandparent roles. Ideally we want an animal that starts breeding in 1 year AND lives for 5+ breeding cycles to be able to learn enough to be useful over its lifetime. It takes so long for humans to learn enough to be useful!
@Ryan Greenblatt I hereby request you articulate the thing you said to me earlier about the octopi breeding program!
Cephalopods are highly intelligent but also have a very short lifecycle and very large number of offspring—making them a prime target for a artificial breeding program for intelligence uplifting.
E.g. the lifecycle of many octopi is about 1-2 years. A 15 year program could potentially breed for very significant behaviourial changes.
For instance, dog and horse breeds can be made within 10 generations.
Google says the female giant pacific octopis lays between 120k and 400k eggs at the end of her life. Giant pacific octopi live about 3-5 years.
I asked Claude to make some back of the enveloppe calculations
>>>I’ll use the breeder’s equation to estimate the potential response to selection for intelligence in octopuses. The breeder’s equation states:
R = h² × S
Where:
R is the response to selection (change in trait mean per generation)
h² is the narrow-sense heritability of the trait
S is the selection differential (difference between selected parents and population mean)
Let’s make some reasonable assumptions for these parameters in octopuses:
Heritability (h²): For cognitive traits in animals, heritability typically ranges from 0.2 to 0.6. Let’s use h² = 0.4 as a moderate estimate for octopus intelligence.
Selection differential (S): With 120,000-400,000 eggs per female, we could be extremely selective. If we select the top 0.1% of individuals (still giving us 120-400 individuals), we could achieve a selection differential of about 3 standard deviations.
Let’s calculate the response to selection per generation:
R = 0.4 × 3 = 1.2 standard deviations per generation
For a breeding program over 15 years with Giant Pacific Octopuses (3-5 year lifecycle):
Using a 4-year lifecycle: 15 ÷ 4 = 3.75 generations (let’s round to 4 generations)
Total expected gain: 4 × 1.2 = 4.8 standard deviations
For octopuses with a shorter 1-2 year lifecycle:
Using a 1.5-year lifecycle: 15 ÷ 1.5 = 10 generations
Total expected gain: 10 × 1.2 = 12 standard deviations
This represents a substantial shift in the distribution of intelligence. For context, the difference in IQ between average humans and those considered profoundly gifted is about 4 standard deviations. A shift of 12 standard deviations would be extraordinary.
However, several factors would likely limit these theoretical gains:
Selection plateaus as genetic variation is depleted
Pleiotropy (genes affecting multiple traits) may create unfavorable trade-offs
The trait may reach biological/physiological limits
Inbreeding depression could become an issue with intense selection
Even with these limitations, the potential for significant intelligence enhancement in octopuses through selective breeding appears substantial, especially for species with shorter lifecycles.
Alexander reacting to Claude back of enveloppe calculation:
Narrow sense heritability is probably higher for intelligence. Indeed, at this level of selection one would have to worry about inbreeding depression, selection plateaus etc
My best guess is that 6 SD would be possible IF one was actually able to accurately select top 0.1%.
This is probably quite hard. GWAS data for humans currently does not allow for this kind of precision. Doing the selection accurately / i.e. estimating the gradient is the main rate-limiting step [as it is in deep learning!]. One would need to construct psychometrically valid tests for cephalopod, run them at scale.
How much is 6 SD? Octopi might be about as intelligent as a dog (uncertain about this). 6 SD would be quite insane, and would naively plausibly push them tonthe upper end of dolphin/chimpanzee intelligence. The main limitor in my mind is that octopi are not natively social species and are do not have a long enough lifecycle to do significant learning so this might not actually lead to intelligence uplifting. Additionlly, there is the issue that octopi dont have vocal chords so would need to communicate differently.
Ofc like the famous Soviet silver fox breeding program one could separately select for sociability [which might be more important for effective intelligence. Iirc wolfs usually outperform dogs cognitively yet some dog breeds are generally considered more intelligent in a relevant sense.]
Amusingly, this is actually not exactly what we were discussing and is mostly off-topic for our discussion.
Also relevant: Gwern on Tryon on rats, estimate of the cost of breeding very smart parrots/Keas.
The fact that they have a short lifecycle with only 1 lifetime breeding cycle is though. A lot of intelligent animals, like humans, chimps, elephants, dolphins, orcas, have long lives with many breeding cycles and grandparent roles. Ideally we want an animal that starts breeding in 1 year AND lives for 5+ breeding cycles to be able to learn enough to be useful over its lifetime. It takes so long for humans to learn enough to be useful!
Done!