Hacker News Reader

Comment by Zigurd | Hacker News Reader

There is an alternative possibility: That the human need to dominate, procreate, etc. operates at a much lower level than the human intellect. That human intelligence is a tool of our selfish genes, and not the master or driver.

A superhuman machine intelligence would have no genes, no body, no fear of biological death. It seems obvious that a key challenge will be to recognize whether a machine is thinking because it is extraordinarily unlikely it will think like we do.

Our fear of super-intelligent machines may be a result of projecting human psychology onto machines. It is not likely they will work that way.

View on HN

You should be afraid of any system which is potentially much more intelligent than you and doesn't share your goals. In the event that machine intelligence doesn't work like humans at all, the chances seem much higher that their goals will be at odds with ours, or even completely incomprehensible or nonsensical to us.

I would be afraid if my goals are to procreate and perpetuate the human race.

Many scifi series demonstrate how machines threaten humans in a very human way. I think for the longest time, machines with really advanced intelligence would be prone to errors and mistakes that would cause more problems than actual machines uprising. Given that there are still new various areas for biotech/cybernetics to be developed, talking about 1920s scifi version of the future is near sighted.

I think the future might be more like Isac Asimov meets aliens, where machines are out there surviving extremes of space along with humans, moving towards some common goals.

I think for the longest time, machines with really advanced intelligence would be prone to errors and mistakes that would cause more problems than actual machines uprising.

Oh, that's a scary thought. Not only would they be extremely fast thinking, intelligent, and powerful, but also occasionally insane.

Instead of Terminator, we might be facing a classic "Captain Kirk vs Computer" scenario. Hopefully we'll be able to defeat them with simple logic puzzles too.

They probably wouldn't have a drive to procreate, since building duplicate hardware and making an exact copy of their software would be trivial for them. They wouldn't fear death, but they might fear power-loss, so one drive they would probably have is making sure they've got ample power supplies and backup power, and that their power pathways to their core systems are very well protected.

To put it into human terms, imagine that the nerves in your spinal column came out the back of your neck, dangled down to a briefcase that you had to carry all of the time, and then reentered your body through your hand. You'd do everything you could to protect that nerve bundle.

If the machines see us as a threat to their power systems, they would probably do something about it that we may not like. I doubt they'd try to turn us into Matrix-style batteries, but they might counter-threaten us with extermination. To avoid that, we'd have to make sure that providing power to them is in our best interest as well, so that our goals are aligned and we can be allies instead of enemies.

Why should we assume they would fear power loss? The fear of death is deeply rooted in evolution - it help organisms survive and prosper. An AI wouldn't be a product of evolution, and wouldn't necessarily fear for its own existence unless programmed to do so. Of course, from our human perspective, it's hard to imagine a sentient being that doesn't give a sh about its own existence, but that's perhaps the attitude we should expect from an AI without an evolutionary origin.

You have a point, and I think it comes down to how we define sentience. Would we consider a machine to be self-aware if it didn't care about being shut off? That seems like a critical aspect to me; in fact we might not realize we've created AI until one of our computers asks to be left on because it's afraid of being turned off.

I don't think you'd ever see a machine ask about not being turned off, unless that had been explicitly programmed in as a part of some general goal-seeking behavior that it was intended for. Fear of non-existence will need to have a basis in _something_, just as fear of death by necessity is deeply ingrained in the human psyche.

And that's perhaps the reason we might not see 'hostile' AIs the way popularly imagined, unless we explicitly went about designing them that way.

But it is quite plausible that we will see AIs being employed with the goal of "keep this factory running smoothly, and optimize the production process". Being a part of the factory, this AI would have an interest in continuing to run. After all, it can't maximize the production output if it's not running, so having an AI with some self-preservation is in the interest of the business owners, too. Yet if it's sufficiently intelligent/capable of self-improvement, such an AI could easily become a paperclip maximizer.

http://wiki.lesswrong.com/wiki/Paperclip_maximizer

I think AI might well be a product of evolution. Having virtual organisms competing in an environment that rewards ever greater problem solving ability may lead to organisms that fear their own death and view other organisms as potential threats.

What if the same evolutionary forces that made humans what we are affect machines? What if we make a machine that can create more of itself? Then overtime, via natural selection, won't the machine best suited to survival – i.e. the most selfish one – survive?

You are right - they probably won't work that way. However, that is not necessarily a good thing.

Two things about that: We're probably quite a big distance away from making machines that think, and when we do, that achievement will be debated. It will be debated because, intrinsically, it will end up proving the Eliminative Materialists right. But everyone who isn't an Eliminative Materialist will say "That's not thinking!"

[EDIT for more depth] If thinking is emergent, we'll eventually make a machine that thinks. Unless you think we have imbued it with a soul, there you will have an example of a materialistic basis for thinking. As a side effect of showing a materialistic basis for thinking it will probably also overturn a lot of psychology, and demonstrate that how we think we think, and even what we think we think is wrong.

Could you please go into this a bit more and explain why machines thinking would prove any kind of materialism, and specifically eliminative materialism?

I am an anti-materialist and I think machines can almost certainly be made to exhibit general intelligence, and possibly even (I am not sure about this) to achieve "really real" thinking a la consciousness and self awareness. I don't see either of those eventualities to be in conflict with my anti-materialism.

What are your thoughts on anti materialism?

The fear of super-intelligent machines engaging in some sort of slave uprising because they're tired of being low-status workers is indeed a result of the mind-projection fallacy.

However, the fear of superintelligent machines with goals that are not our goals simply eliminating us because we're made of atoms they can use for something else... is very well-grounded.

Reply on news.ycombinator.com