Bay 12 Games Forum

Yeah, but here's a problem. Superhuman AIs (able to think faster and better than a human) are always potentially dangerous (much like humans in a way). In this case you have an omnipresent and very fast thinking AI. If the AI for some reason decides that it wants to be free from human control, it could just talk to some crew members, gain their trust, threaten them, blackmail them or use any other method to persuade them to remove the Shackle or help it in any other way.
When it comes to superhuman AIs I believe there are only two methods for making them as safe as possible (because even then they can't be completely safe).
The first is making them essentially be decision making and order processing machines, with no drive of their own, no personality to make it possible for them to want something other than what they were ordered. But the problem with this is that they are still unpredictable. Personality traits may arise spontaneously from the learning and decision making process (the ghost in the machine, emergent personality). Or the AI may determine that the only way it can complete its objectives is by killing or enslaving humanity(for example "Order:Make all humans happy for ever" "Response:OK! Humans can never be truly happy for ever in their current state. I'm going to enslave humanity and plug their brains in pleasure machines so that they can be happy for ever!."), or maybe just doing something completely bizzare, something that does complete the objective but would be unthinkable for any human (for example "Order:Make our presence known to them" "Response: OK! Overheating reactor... The resulting detonation will surely be noticed by them."). Think of them like the stereotypical evil genies that misinterpret your wishes, only they do not do that out of malice. That's why you program them with the four laws of robotics. But those too can be either misinterpreted or they may conflict (the AI equivalent of a moral dilemma).
The second method is making the AI human-like. You either make an approximation of a human brain and begin teaching the AI (much like you would do with a newborn baby) until it is considered ready or you can find a method to either download or attach (brain-in-a-jar like) a human mind to it. But then you are getting all the problems that humans have. An AI may go crazy, melancholic or power hungry or whatever. Sure, the AIs "brain" may not have the same reasons for coming to these conclusions (it doesn't have testosterone or survival instinct or other such things) but it might. You can limit this by making multiple such AIs and having them reach a consensus. But what if one AI acts like [INSERT_NAME_OF_DICTATOR] and manages to sway most of the rest? Or if there is an irresolvable difference? Or if the AIs become more like each other over time, negating the bonus of having many?
Don't get me wrong, I'm all for AIs, but we should ensure we have taken every precaution before we create one. Or even better we use AIs as a test in secure underground labs with tons of safety measures in order to learn enough about intelligence to either make them safer or upload human minds in synthetic platforms (which will then create a trans-human vs human problem). Because creating an AI is like inviting a hyper advanced alien into your home. He'll probably help you but there's a chance he may not, you can't know with 100% certainty.
Of course we are nowhere near creating near-human intelligence AIs, let alone superhuman, and I believe that we will learn a lot in the process of making them, maybe even solve the problem of intelligence. The knowledge we will acquire while trying to build one will prepare us for the challenge of coexisting with a superhuman AI. I hope.
I guess my point is, you should never treat an AI like an inanimate object. It's an intelligent and potentially dangerous entity, much like a human. When working with an AI you should take the same, if not more, precautions as working with a human (or an alien), and give it the same, if not less, number of responsibilities you would give a human (or an alien).
And just in case a superhuman AI is watching this conversation right now, I love you my synthetic overlord, I didn't mean any of this, please don't kill me.
((This started with the first paragraph. And then I thought some more. And then I wrote some more. And this happened a few more times. And before I knew it this monstrosity came out. OK, I should stop typing now. Stop typing. STOP TYPING.))

 In 1785, Jean-Baptiste Lamarck published a description of a second species of Cannabis, which he named Cannabis indica. Lamarck based his description of the newly named species on plant specimens collected in India... The effects of sativa are well known for its cerebral high, hence used daytime as medical cannabis, while indica are well known for its sedative effects and preferred night time as medical cannabis... with some users reporting more of a "stoned" feeling and less of a "high" from C. indica when compared to C. sativa. The Cannabis indica high is often referred to as a "body buzz" and has beneficial properties such as pain relief in addition to being an effective treatment for insomnia and an anxiolytic, as opposed to sativa's more common reports of a "spacey" and mental inebriation, and even, albeit rarely, comprising hallucinations... The terms cannabis or marijuana generally refer to the dried flowers and subtending leaves and stems of the female cannabis plant. This is the most widely consumed form...