The Singularity

[warty goblin]

It might not necessarily be a case of large-scale active testing, but statistical analysis of gene-frequencies within existing populations and using- I dunno, hidden markov models, or something- to tease out the causal factors involved. Then taking that knowledge and applying it clinically to individuals- desperate parents with a history of genetic illness in the family, for example- to confirm the hypotheses.

In theory that'd be nice, but that's not really something that I think can be statistically teased out to a degree of specificity I'd feel comfortable messing around with people's babies.

Another data point: over 25% of pregnancies miscarry within six weeks as it is. Monkeying with the process is hardly likely to decrease this, and there's obvious reasons to object to highly expensive ways to get more miscarriages. While it's not impossible that eventually such genetic therapies could prevent miscarriage by selecting against embryos with a high likelihood of not carrying to term, we certainly aren't there yet.

I suspect it'll be limited to treating simple, and severe, mendelian disorders for the first decade or so. But it will move up from there, and by the end of the century, I suspect most aspects of gene function will be thoroughly mapped.

In theory, yes. What I'm hoping for is that advances in detailed understanding of genetic cause-and-effect will be able to keep reasonable pace with black market offerings, given the latter's drawbacks in terms of price and safety. Consumers are willing to tolerate restrictions on legal services provided they (A) give you most of what you might want and (B) have some acceptable moral mandate behind the limitations.

I don't disagree with any of your critiques of pure free-market solutions or with the range of negative outcomes you project. And in the short term- say, maybe the next 10-30 years- blanket bans on everything except severe and specific developmental abnormalities might be viable. But beyond that, as treatments become cheaper, the techniques more refined, and more genetic traits are identified, the pressure to turn to the black market will become more and more intense if legal outlets for genetic augmentation are absent.

Now, if you want to argue against, say, gender descrimination, or against genes for hyper-competitive sociopathy (assuming those exist,) or against tailoring for prevailing beauty-standards (as distinct from metabolic health,) by all means do. I think there are reasonable grounds for banning forms of gene tailoring which confer individual advantage without any tangible social benefits, and I think most people will be willing to accept the moral reasoning behind such legislation. But a blanket ban on gene-selection/tailoring for anything but curing the nastiest diseases, however laudable in intent, strikes me as difficult to permanently enforce within our current socioeconomic framework.

Most of the reason I'm highly skeptical of gene tailoring is that the costs of being wrong are high, and the probabilities of being wrong are high.

As I pointed out, we get things wrong with clinical trials fairly frequently. And that's when there's at least some ability to control confounding variables, outside variation, and using a well understood probability structure.

You don't get that with genetic tinkering in humans, at least until you start doing it.

Statistics is a marvelous tool for modeling, predicting and describing things. It's absolutely terrible at making a cause and effect argument - to the point where most of my statistics professors basically just say 'don't even try.' If you have a very well designed experiment for controlling outside variation, hidden and confounding variables, replication, and a matching theory to support your conclusions you can maybe, maybe make a causal argument. Industrial statisticians doing reliability studies can do this, because they have very good controls, and physics is a well developed theory.

Genetics is a well developed theory, but we can't control people's genetics, or their lifestyles. There's all kinds of confounding and variation there. I'm not into gene mapping or bioinformatics, so I won't go so far as to say its impossible, but it strikes me as a very problematic application of the methodology.

(From a slightly different angle, statistics is good at prediction and modeling because it doesn't care if you've looked at the truth or not, only something that tracks the truth reasonably well. Predicting things from genetic material is a reasonable application - I've talked with people who do that - but actually being able to concretely say 'gene X causes Y' with enough certainty to bet somebody's life on plugging gene X into them is not something you can do every day of the week. It's not impossible, and I've seen cases where it's happened. I'd be surprised however if its occurrence is common enough that we're anywhere close to being able to choose a child's intelligence. Not least because measuring intelligence in the first place is hard*.)

(*Quite a few things that seek to measure intelligence and ability turn out to be essentially useless. The GRE for instance is a very good measure of socio-economic background, and a terrible measure of actual ability to succeed as a graduate student. By some methodologies it's actually negatively correlated with success for women. In the science of education, I've heard it said that an R square of .3 or so is a really good model. Yes this is a footnote to a parenthetical. You may now shoot me)

Oh, open-source doesn't necessarily mean 'not paid for it'. (It does mean that any key ideas are easier to copy/steal, but that's not necessarily a problem if different economic niches are addressed by different products.)

I know, I'm just skeptical of open source as a movement. Mostly when I'm trying to get R to do something SAS does automatically.

Anyways. My more general, long-term concern here would be that a sufficiently devious AI, even if limited to low-frequency trades, might be capable of large-scale analysis of market forces and long-term price manipulation in favour of it's parent corporation. That's a ways off yet, but I wouldn't put it outside the realm of possibility within a century.

That strikes me as unlikely. The stock market is notoriously hard to predict, stock prices actually make reasonably good random numbers.

True. But nukes were developed- and even used on human targets- before they got banned. Rollbacks in these areas are possible, though I hope we don't have to repeat the precedent.

I think the key difference is that nukes - once more than one person have them - come with a built in lose-lose guarantee, one that getting better nukes doesn't take away. Robots, at least sub-nuclear robots, don't come with that. The presence of nuclear weapons on both sides raises the cost of hostilities to the unbearably high for the 'victor,' before anybody fires a shot. Combat robots are, unless somebody's idiotic enough to give them nukes, entirely favorable to the party who spends the most and builds the best.

In other words the fact that a potential enemy has Mk.III Skullreaper Killdroids doesn't bother me if I have Mk.V Skullreaper Killdroids, since I'm secure in the knowledge that my army will provide a more effective skull-reaping solution for the competitive modern battlespace. The fact that my enemy has a 5 megaton nuclear weapon that can vaporize me and my little dog does bother me, even if I have a 10 megaton weapon pointed right back at them. I don't get bonus points for killing the enemy more dead because dead is binary. It doesn't come in fifty shades of grey.