There is a character limit in the titles, my question is can you use selective breeding to push a character trait 20-30 standard deviations w/o creating an entirely new species, or w/o causing so many other problems that the life form becomes useless?
I was reading an article about breeding humans via a mix of selective breeding and possibly CRISPR, they mention how in agriculture they have already pushed animal traits 20-30 SD in a direction that is more productive.
However by the time you get to 8 SD, you are already in the 1 in a trillion category in a normal distribution. Plus with return to the median and all that, I don’t see how something so extreme as a 20-30 SD movement of a certain trait is possible over such a short time frame. Plus, who knows what other issues that would create by doing that.
Over time, things like weight can be pushed pretty far from where they started. Dinosaurs evolved into birds which are many thousands of times lighter, but they became a new species in the process that looked and acted totally differently.
I’m not a geneticist but isn’t the limiting factor that the gene must exist? You’re not going to breed people who are ten feet tall unless somewhere in the world there are people who have the genes for being ten feet tall. If you just breed the tallest people you can find together, you’ll end up with a bunch of people who are between seven and eight feet tall because that’s the “tallest” genes that exist. You can isolate genes so they express themselves but you can’t create genes or cause genes to change in a planned direction. That would be Lysenkoism.
Also not a geneticist, but the since Wolfhounds and Chihuahuas are the same species and can apparently interbreed, it seems the answer to your question must be “yes”. This is of course over millennia, not a particularly short timeframe.
Even that is not even remotely close to 20 or 30 standard deviations. Wesley Clark isn’t asking if you can breed for large variations within the same species - you can in everything from corn to dogs. The question is about a statistical reference to an unexpected number of standard deviations. Standard deviations quickly become increasingly more rare as they get bigger to the point they get almost unusable for any standard deviation over 6. Standard deviations aren’t even measured or referenced in the 20+ range because that is an absurd claim for almost any measure that we can conceive.
I think someone writing about it just didn’t understand statistical measures very well. They probably just heard that a SD of 4 was pretty big so it is better to tack on some generous multiples to that to make it even more dramatic. Unfortunately, standard deviations don’t work that way and the claim sounds stupid to anyone that understands them at all.
Not necessarily. First, “tallness” or whatever other characteristics you’re looking for is probably not a single gene but a combination of multiple genes and other factors such as nutrition. Even if the genes themselves exist, they may not have existed in a particular combination, or in conjunction with the outside variables, until you started your breeding program.
Second, Lysenkoism concerns the heritability of acquired characteristics (if you cut the tails off mice, soon mice will be born without tails–that doesn’t work). What you’re talking about, though, is not acquired characteristics but reinforcement and extension of existing characteristics. Look at miniature horses (or pigs, or dogs, or any of the various other species that exist in smaller-than-normal forms, or conversely in larger-than-normal forms such as Flores giant rats). In species that have a substantial degree of variability anyway, breeding the two largest specimens together will typically produce offspring that cluster around the parents, but can be slightly larger or smaller, so yes you can obtain a horse that is fully mature and two percent smaller than either parent, or a dog that is several percent larger than either parent. Continue breeding the outliers together, and the offspring will tend to cluster around a new normal. This is essentially what happens with insular dwarfism, e.g.
Is genetic information (in a species with a high variability anyway, such as dogs) necessarily in a normal distribution? If your distribution is out of whack, you might have a greater number of SD without getting to the one-in-a-trillion level (I think).
20 - 30 standard deviations isn’t useful no matter what the curve looks like. It exceeds the number of atoms in the universe many times over and there are no SD tables that go up nearly that high. It wouldn’t matter if they did because nothing you can measure can ever meet it.
The fact that you have to include the examples that deviate from the population mean preclude you from ever reaching 20 - 30 standard deviations. To give an extreme example, let’s say you develop a breed of horse that is 1 inch tall. That is a huge difference from a normal horse but you will still end up with a few of those examples mixed in with the Thoroughbreds and Clydesdale’s when you take your sample.
There aren’t enough horses in the world to even bring the 1 inch tall horses out to the 7 Standard Deviation level let alone 20 - 30. That math just doesn’t work that way and it is effectively impossible.
I am going to assume that a science journalist did a really terrible job (as usual) rather than assume he is an idiot. It may be both though. There is no such thing as an IQ of 1000 because IQ is a man-made construct that can’t be tested for currently at anything over about 160 and even that is iffy and controversial at that high a level.
You also can’t have a standard deviation of 100 because the the person being tested is part of the sample and there aren’t that many people in the world (or grains of sand in the universe for that matter).
The whole conjecture is complete crap. He may be a genius at physics but he seems to know almost nothing about IQ tests and basic statistics.
Actually, there is a bigger problem with that. The notion of the standard deviation is tied to the normal or Gaussian distribution. There are certainly a lot of things that can be approximated by a Gaussian distribution or as a function to a Gaussian distribution (e.g. lognormal) in most of the population as long as their parameters are truly randomly distributed about a central mean, but almost nothing except in quantum mechanics is literally Gaussian. This means that the further you get away from the mean and into the tails, the less precise the approximation is going to be. Once you start estimating parameters outside of five or six standard deviations the estimates are essentially meaningless; it takes so little difference to change the “thickness” of the tails that they don’t provide any credible estimate of probability. As a practical example, buying two PowerBall tickets numerically doubles your probability of winning, but the odds are still so overwhelmingly against either ticket being the winning one that as a practical matter the likelihood that you will win has not substantially changed, as the likelihood estimate is lost in significant figures.
Speaking of a trait being “pushed 20-30 standard deviations” is not really a meaningful question to begin with. There is a level of natural variability in many traits, such as height, skin and hair color, et cetera, but these are not typically normally distributed and are often governed by more than one gene or expressed due to actions of other genes or epigenetic factors, so there is a complex interplay that defines how traits are distributed within a population. In many cases, there are physiological limitations as to how far a trait could be altered. The giraffe, for instance, is just about as tall as it could possibly be without having to rework the anatomy significantly; the Trumpeter Swan is about as large as waterfowl can get; a humanoid can’t grow to much more than 9 or 10 feet in height before bipedal plantigrade anatomy is not longer workable; no creature will every be able to see radiation above the ultraviolet levels because the radiation is inherently harmful to organic materials; et cetera, ad nauseam. Selective breeding by itself can only emphasize a few characteristics in their extreme, and if you notice, most highly breed domestic creatures such as dogs, cats, horses, et cetera, tend toward neotenic forms; in part because it makes them more emotionally appealing to us, but also because it is fairly easy to breed for characteristics that already appear during development rather than to advance those which appear only in adulthood.
This argument is missing the point. It’s true that the average size of chickens has significantly increased in the last sixty years via selective breeding.
But there were no mutations involved. The genes for weighing over 4000 grams existed back in 1957 but they were rare so the average chicken weighed less than 1000 grams. Selective breeding isolated the larger weight genes and made them common so now chickens that weigh over 4000 grams are common.
But the genes never changed. There are no genes in the chicken gene pool that didn’t exist back in 1957. It’s just that some genes that were rare have now become common. You can’t breed a species beyond what its genetic potential is.
And I think Hsu might agree with me. From reading his article, it appears he feels that current human genes have the potential to produce people with 1000 IQ’s. We just need to line existing genes up right. And on that issue, I have to question his premise.
Hsu’s argument is that there are an estimated ten thousand alleles which affect intelligence and we have found that as few as a hundred alleles can raise intelligence by a factor of 10-15 IQ points. So he multiplies the two and figures ten thousand alleles could raise intelligence by a factor of 1000-1500 IQ points.
That seems highly unlikely. Hsu’s assuming that each group of alleles will have a separate cumulative effect. Considering the amount of redundancy known to exist in genetics, it’s more likely these ten thousand alleles have effects that generally overlap rather than stack.
Probably true, but it doesn’t necessarily change the argument. Standard deviations are about frequencies in a population. A 1000 IQ is by definition 60 standard deviations above the mean, but it doesn’t say anything about how smart that hypothetical person is.
The upper probability for 60 standard deviations is about 1.2e-784. That is equivalent about 2600 bits worth of “information”. So, if you can find 2600 alleles with about a 50/50 distribution in the population, and are uncorrelated with each other, and can create a person with those 2600 alleles all optimized to a specific value, then you have a human that’s 60 standard deviations out from the baseline.
The trouble really is in the step going from those alleles to IQ in general. You can’t know if those are the only ones associated with intelligence; and if there’s more than that, it fattens the tail on your distribution. Still, it remains true that within the group of alleles you’ve identified, any one particular configuration can be said to be X standard deviations out, even if the numbers are way higher than the actual human population.
Using the most favorable numbers: I’m finding that the weight of wolves average at 98 pounds, with a standard deviation of 6.5 pounds. The smallest chihuahuas are about 3.3 pounds. That’s about 14 standard deviations away from the wolf average.
It’s conceivable that there’s some other domesticated animal with an even greater change, though I’m not sure offhand what it would be. Clydesdales to wild ponies, maybe?
Since intelligence is the one and only competitive advantage H. sapiens has ever had, I rather suspect natural evolution has pushed the intelligence of the species as far as can be done. If the structure of our brains permitted any significantly (like 2x) greater intelligence – 4 million years of brutal natural selection would have caused it to happen. It didn’t. A reasonable first hypothesis is that it’s not possible. Which means natural or artificial selection could probably not produce even so much as a 50% increase in mean human intelligence.
Which would mean to get much further you’d have to design a much greater intelligence de novo, by designing the “hardware” (biological substrate), the basic “operating system” level functioning – how information flows around, gets stored, categorized, recalled – and then the higher level functions that give creativity, insight, superb inductive skill, whatever you want to call “intelligence.” We currently have no idea how to design a machine with our own level of intelligence, or even something as smart as your common dog. How would we ever design something 100x smarter? That’d be like a horse designing and programming a computer that ended up with the intelligence of a man.
Interesting argument but I’d question the premise. I think our main advantage is our social grouping. Humans live and work together. I feel this is more important because a group of humans working together don’t have to be individually all that intelligent; they have the collective intelligence of the smartest members in the group. The average human only needs to be intelligent enough to communicate with the other members of his or her group.
Memory is imperfect. It would help if you could link to this source that you were reading. And note that CRISPR is not selective breeding-- it’s genetic engineering.
Are you assuming all distributions are Gaussian? Standard deviation is a well-defined statistic which does not require Gaussianness. The world’s tallest man really is over 8 feet tall, which really does put him more than 10 standard deviations from the mean. That’s much more than would be plausible if tallness truly fit a Gaussian model.
Using variance as a dispersion measure and assuming distributions are Gaussian are both artifacts of “the light is better here” phenomenon. Stats can get weird. For example, do recall that the Cauchy distribution, which looks a bit like a “bell-shaped curve,” has infinite variance.
Humans could certainly be more intelligent. We know for an absolute fact that it’s possible for a human to be as smart as Einstein, since, after all, Einstein was human. Now, only a very small proportion of humans are as smart as Einstein, but there’s no reason that proportion couldn’t improve. Evolution or genetic engineering might increase human intelligence by raising the floor, bringing people up closer to Einstein.
Could it result in someone even smarter than Einstein (or whomever you consider to be the most intelligent human)? That, we don’t know yet. But intelligence on that level is so rare, that we can’t really even guess with confidence yes or no.