Intelligently Design a Human

[Samuel]

You propose that we compensate for removing all our hypothetical designed species's needs by the brute force solution of programming it to do the things we want it to (build civilizations and discover X, Y, and Z) whether it needs to or not.

The alternative is having it need to advance to survive which could backfire like it did in humans- we went through a genetic bottleneck because most of the species wasn't able to adapt to changing conditions. While the new and improved version has problems, there is no guarentee that baseline humanity would survive if we restarted things over with minor changes.

There's the danger of having the programming be too strong (in which case you get a culture that's so busy investigating that it winds up neglecting the few survival needs it does have).

Just make survival and curiosity both bell curved traits (with curiosity to the left) so that the majority of the civilization regards survival as more important but there are individuals willing to risk injury and death in order to find out something new.

Or of having it go in a slightly weird direction, in which case you get a race of obsessive navel-contemplaters, or a bunch of people too busy collecting butterflies to show any interest in the wider universe.

That only works if there is a strong need to conform. Otherwise their interests should not all sync up.

You'd have to wire people to not only discover things they don't need to know, but to use that knowledge to build things they don't need to own... and that could backfire even more excitingly.

Why is that a problem under this scenario? People would naturally congregate because the larger the population the more likely someone shares their interest and can add to what they know/check over what they have discovered. Eventually all the easy stuff will be understood and there will become a desire to find a way to refine an observation and individuals interested in tool building will come into demand.

[Starglider]

Which, I agree with Lagomonster here, is likely to backfire. There's the danger of having the programming be too strong (in which case you get a culture that's so busy investigating that it winds up neglecting the few survival needs it does have). Or of having it go in a slightly weird direction, in which case you get a race of obsessive navel-contemplaters, or a bunch of people too busy collecting butterflies to show any interest in the wider universe.

Frankly, with the degree of cavalier 'let's do this, this and this - it'll all work together, somehow' that people are showing in this thread, we have to assume some terribly competent and thorough R&D team or super-AI doing the physical realisation and removing all the bugs. Cognition is harder to predict in many ways, but it as a system it is not fundamentally any more fragile than biochemisty - in some ways it is actually less, in that you can kill human cells with toxins that block a single biochemical pathway, but a human brain is pretty good at mitigating the effects of localised damage.

And to make matters worse, a key aspect to the evolution of science is the mixing of pure curiosity with the practical question "Yes, but what are we going to do with it?"

Yes, and? The 'Red Queen's race' aspect of social living pretty much ensures that such applications always exist, warfare being the obvious one, although personally I would consider it a design failure if you are forced to resort to such methods.

That's how science diverges from philosophy- you don't just come up with clever ideas about nature, you actually try to do something with them.

The current human technology base is completely excessive for the purpose of meeting just our basic physical needs, something that Luddites and eco-nuts harp on about continually. Yet we keep trying to build things and do things because we can, or because it's there, or because we want to impress our peers.

You'd have to wire people to not only discover things they don't need to know, but to use that knowledge to build things they don't need to own... and that could backfire even more excitingly.

Don't give me that crap, I spend most of my time working on AI systems where you do have to explicitly specify what the system should want, otherwise it will just sit there. There is absolutely nothing wrong with this. Yes, there are some spectacular failure cases, but when engineering a biological organism from scratch there are plenty of failure cases for every aspect of the design. In fact I would say that simple copying of human cognitive design, which is a messy pile of crap kludged up by evolution without any consideration of these issues, is far more likely to fail than a reasonable attempt to design such an architecture.

[Simon_Jester]

Don't give me that crap, I spend most of my time working on AI systems where you do have to explicitly specify what the system should want, otherwise it will just sit there. There is absolutely nothing wrong with this. Yes, there are some spectacular failure cases, but when engineering a biological organism from scratch there are plenty of failure cases for every aspect of the design. In fact I would say that simple copying of human cognitive design, which is a messy pile of crap kludged up by evolution without any consideration of these issues, is far more likely to fail than a reasonable attempt to design such an architecture.

Look, my real concern is that I find the specification of desires the organism doesn't need to be a less desirable solution than giving it reasons to have those desires other than "they were in the design specs." We're trying to design a general intelligence here, not a specialized tool; it ought to be able to figure out what it should want.

For most AI applications, the situation is different because we want the AI to do something very specific that there is no obvious reason for the AI to want to do except "the builder wants it."* We don't just want it to go down, be fruitful and multiply, and hopefully build a civilization.

Whereas for our hypothetical designed intelligence, we have more general objectives. If it doesn't want what we'd like it to want because we designed an immortal autarkic superbeing with no need for civilization or technology, and we didn't want it to make do without those things, I think that's at least as much evidence that we overengineered it physically as that we underengineered it cognitively.

*Yes, I know, this is not specifically true of every general AI project; I don't even know that most general AI projects have a specific application in mind, but the field would be even less funded and supported if not for the desire to produce AIs that are useful for applications, not just for the sake of having a certain list of capabilities.

[Starglider]

Look, my real concern is that I find the specification of desires the organism doesn't need to be a less desirable solution than giving it reasons to have those desires other than "they were in the design specs."

Humans don't have any reasons for our desires other than 'those drives tended to propagate the species'. I don't see anything particularly deep and meaningful in that.

We're trying to design a general intelligence here, not a specialized tool; it ought to be able to figure out what it should want.

All goals are arbitrary. For a rational system, 'figuring out what it wants' consists of exactly two activities; converting abstract goals into more specific goals, and resolving any ambiguity present in the original goal specification. Irrational systems e.g. humans have a less straightforward motivational system, but it boils down to the same thing. You cannot pluck goals out of thin air; not humans, not genetically engineered organisms, not seed AIs. You can construct new goals, if your cognitive system is well designed you may even be able to set them directly instead of just trying to live by them, but those new goals are always based on existing goals.

We don't just want it to go down, be fruitful and multiply, and hopefully build a civilization. Whereas for our hypothetical designed intelligence, we have more general objectives.

There is no fundamental difference. You just need different tools for expressing what you want. Gravitating towards opaque, emergent methods for getting 'civilisation' from simple, human-like base desires is a case of emergence mysticism aka ignorance worship. There is nothing more 'authentic' about such a derivation and any such perception is a failure on your part to consider the messy details of the process. Specifying what you want directly, without flim-flam or unnecessary intermediate stages, is always more reliable and ultimately (if done competently) allows more scope for richness and diversity.

I think that's at least as much evidence that we overengineered it physically as that we underengineered it cognitively.

To me, overengineering means that you blew the budget, or you made it too complex to maintain. Simply exceeding specifications is always a good thing - if it is overengineering, it is overengineering of a purely positive kind. Deliberately crippling an intelligent being because you thought that it would make up for your inability to directly make its cognitive design do what you want is in fact cruelty on a massive scale.

Yes, I know, this is not specifically true of every general AI project; I don't even know that most general AI projects have a specific application in mind

Actually most general AI projects are very open ended in intent, but then they've all failed miserably so far. Narrow AI systems that actually do useful things have very narrow motivational systems, either internal (expert systems) or external (genetic algorithms). As you may have guessed, I would consider the later a nasty cop-out stemming from inability to design the former properly.

[Terralthra]

Consider;
If you give humans stronger muscles, thicker skin, strong bones, etc, they can now go toe-to-toe with alot of animals on the planet, and win, without using weapons. Therefore, no need to develop them.

Except the weapons arms race will be just as important for taking on the stronger, more well-armored other humans. We made spears to kill animals. We made arbalests to demolish castles.

If you make humans able to take colder temperatures, there is no need to develop clothing. That means when they reach arctic climates, they don't know how to protect themselves, and it limits migration. End result is North America has no humans in it.

So, because they won't need to develop clothing, they won't be able to develop clothing when they need it? Did you even read this as you wrote it?

If humans can breath water, then there is no need to develop boats. Therefore, more limited migration.

If they can breathe water, they don't need boats to migrate across oceans. Unless the current/weather/etc. stops them, in which case they'll develop boats.

If we can digest more, we don't need as much food, and might slow down agricultural development.

If we need less food, why do we need as much agricultural development?

Your objections are very poorly thought-out.

[CaptainChewbacca]

Also, the ability to breathe underwater preventing the development of boats is foolish for other reasons. We developed the wheel even though we can walk, we did it to move large loads quickly over distances. Boats accomplish that in water far more easily than walking along the seafloor. Also, boats provide a suitable platform to perform work at or near the surface of the ocean (fishing, diving, etc). These will still be necessary for the development of civilization.

[His Divine Shadow]

- A brain that is designed more like an avian brain, where processing occurs in the brain's volume, rather than on its surface area. This will allow the brain to be smaller, and you can fill the freed volume with fluid as the ID human ages (allowing for smaller skulls at birth, and making head injuries much less lethal.)

Why not make the bone thicker as well? Or since we've gotten a bunch of free space from making the brain better organized, why not use the space and expand other functions? Better eyes could use a bigger visual center for instance. I also like the idea of adding a reflective layer like in cats eyes for better nightvision.

-Regenerative ear bone BTW.

[Darth Holbytlan]

Why go for even more radical pelvis redesigns or marsupial births when it would be so much easier to move the birth canal? The childhood notion of babies coming out of the belly button isn't really that bad of an idea, considering.

You do realize that beneath the navel are the muscles of the abdominal wall? The birth canal is already in the optimal place.

In order to "optimally" squeeze baby's brain during birth? Yeah, right.

Also keep in mind that I don't mean to use the naval literally, just that the hole shouldn't go through the pelvis, forcing birth size to compete with locomotion efficiency. Sticking it just above the pelvis would still provide plenty of nearby rigid support. The uterus could also go outside the abdominal wall, possibly sharing a muscle sheath.

[His Divine Shadow]

What specifically do you want to use this for? Adding color channels (beyond the standard three) complicates visual processing, so it isn't something you should do on a whim.

If you decrease the total area of the brain it frees up more space for visual processing though. But I don't see UV as that interesting. More interesting is refining the design of the eye into it's optimal configuration.

It's in there to protect it. If the main nerves ran outside of the spine, ordinary stab wounds would have a much higher chance of paralysing you. I would use a redundant switched data processing network, but that really requires a nervous system based on electronic conduction, rather than ionic depolarisation.

Maybe a better solution is to add regenerative abilities of severed nerves.

This is something else to go wrong; if one of the valves closes pathologically, the affected area will necrotise and the organism will probably die. Using more smaller blood vessels is possible but at the cost of more energy expended on pumping.

Alternative would be improved clotting ability of blood. Not sure if that would increase dangers of blood clots though.

I also think it would be worth adding more and effective ways of dealing with free radicals in the cells, removing any flaws that increase risk of cancers and other genetic diseases.

I'm wondering if ultimately the most effective way in the future would be to use AIs and specify the gross shape and features you want and then let the AI build the optimal genotype from the ground up to express it. In essence start over from scratch with a fresh slate, a true intelligent design in other words. It would require an AI with a complete knowledge of how genes work and the ability to simulate and test it's way along but I don't see that as impossible eventually.

[Starglider]

I'm wondering if ultimately the most effective way in the future would be to use AIs and specify the gross shape and features you want and then let the AI build the optimal genotype from the ground up to express it.

It's either that or go through thousands, probably millions of 'prototypes', many of them born horribly deformed and/or cancerous. Bit of an ethical problem for normal humans, maybe not for homicidal sadist old-testament god. Of course if we dumped genetics entirely and used something more robust and predictable this would be much easier, but I gather that's not an option.

[Simon_Jester]

I'm wondering if ultimately the most effective way in the future would be to use AIs and specify the gross shape and features you want and then let the AI build the optimal genotype from the ground up to express it.

It's either that or go through thousands, probably millions of 'prototypes', many of them born horribly deformed and/or cancerous. Bit of an ethical problem for normal humans, maybe not for homicidal sadist old-testament god. Of course if we dumped genetics entirely and used something more robust and predictable this would be much easier, but I gather that's not an option.

Although the Christian deity, at least, has sufficient computing resources to run simulations of the biochemistry of living organisms down to the atomic level in much-better-than-real time; such a one wouldn't need to go through an extensive prototyping process. So if we're going with the original poster's premise, I think we can assume a comparable ability to get the design right on the first try.

Which is related to your observation that we're designing the organism's basic drives from the ground up, that this is not a problem, and that it circumvents the problem of the creature being too invincible to need to develop civilization. Which, by the way, I concede.

[Starglider]

Although the Christian deity, at least, has sufficient computing resources to run simulations of the biochemistry of living organisms down to the atomic level in much-better-than-real time

Fine-grained simulation of an entire organism plus appropriate environment is logically and ethically equivalent to physically implementing it, so 'virtual prototypes' are just as bad as physical prototypes if you allow them to fully develop. Avoiding this problem requires carefully controlled 'patchwork simulation' sufficient to verify the design without actually implementing a (functionally) sentient creature. This is nightmarishly difficult, particularly for things like verifying that social instincts will make the desired society - not only is the simulation design challenge immense, it's very hard to come up with a precise ethical formula that will grade simulations that look like a person for actual personhood (for the purpose of determining how wrong it is to spawn and kill billions of them in your design process). Idiots still stuck on 'person = unique indivisible consciousness aka soul', 'person = squishy, non-squishy = not person' or even (at the opposite extreme of the bigot/liberal scale) 'if it passes the Turing test it's a person' need not apply.

This is incidentally a real (although seldom acknowledged) ethical issue in general AI design; future AIs might create millions of internalised human-equivalent intelligences, which are coerced and then 'killed' (deleted), just as part of its attempt to understand human behavior. For some designs (e.g. genetic programming) it's a real problem just for making the main AI itself. Unbounded superintelligence can probably find a way to precisely design and verify arbitrary intelligent creatures without creating ethically hazardous simulations of them, but it requires a lot more care than just throwing brute force at the problem. Care that I certainly wouldn't expect from the christian notion of god.

[Gilthan]

Nuclear power is obviously a must, as well as flight.

Atomic angels:

3 "sexes." Males, females, and drones, all biomechanoid. The drones are nuclear dragons (living aircraft), large enough for compact reactors. Each ton of rock or soil eaten by the dragons with typically 6 grams of thorium content at 6 ppm gives 450 GJ, thus the energy equivalence of 107 million food kcal, energetically like eating 385000 potatoes but a lot easier to acquire and much less bulky.

The males and females are far smaller and instead humanoids, 70 pounds each as adults, with each wing extending to 6ft when fully extended but folding up close to the body when walking (12 ft wingspans, for just enough lift to weight ratio).

Not large enough to have their own reactors, they recharge automatically when within a moderate distance of the giant dragon drones (inductive electric power transfer) and may also drink liquid fuel (which the dragons produce from atmospheric carbon).

Throw in empathy and telepathy (radio), telekinesis (secreted foglet micro robots), and reincarnation (mind-states in constant backup and, if their body dies, downloaded into the next egg of the Borg collective their family unit).

Built-in lasers are optional.

[Simon_Jester]

Fine-grained simulation of an entire organism plus appropriate environment is logically and ethically equivalent to physically implementing it, so 'virtual prototypes' are just as bad as physical prototypes if you allow them to fully develop. Avoiding this problem requires carefully controlled 'patchwork simulation' sufficient to verify the design without actually implementing a (functionally) sentient creature. This is nightmarishly difficult, particularly for things like verifying that social instincts will make the desired society

Though I imagine you could at least weed out the biochemical disasters this way. A program which brute-force simulates a mouse is at most ethically equivalent to physically implementing a mouse, for instance, and mice aren't functionally sentient.

Presumably, you could run a similar simulation of "man with higher brain functions abstracted out, running on puppet strings operated by the simulation" or "comatose man with no higher brain function," which would at least allow you to eliminate designs that are physically dysfunctional, if not mentally dysfunctional.

Figuring out how to make sure that your mental design passes checksum without mass experimentation on helpless subjects is, I admit, difficult.