I want a garage door opener that I can point at anybody and turn them off. I'd probably have to put new batteries in it every other day.
Would-be exterminator, behold, Extinguisher of Life. I would personally be quite pleased if I could just live remote from annoying people, maybe have my own spaceship so I can steer clear.
I'm having an especially anti-social day. I don't necessarily want to kill people, just turn them off so I don't have to deal with them. They can turn back on when I'm well away from them.
Can we replace the terminology "human killswitch" with "electronic donkeypunch"? Because I have a feeling that would be the more common application.
many people misinterpret the phrase 'kill switch' when it applies to electronics, but honestly it makes a kind of sense, even if AI become an independent 'species' we would still need at least some measure of protection if we don't want them to decide we're unnecessary and the universe would be better off without us. Even if, like us, AI are generally neutral, the few who decide we're irrelevant could potentially do irreparable damage before we could even negotiate, presuming they lacked any kind of moral obligations, were intelligent enough, and had access to the right systems. Any 'human' who crossed over into the hybrid area would also need similar controls for much the same reason, we'd still need to be able to police them, and those who choose not to take such extreme steps should be allowed that freedom, if possible. It will certainly make things more complicated, but that's hardly a new problem
That was the whole point with Asimov's Robots and Laws or Robotics. The robot's positronic brains were built so that they had a definition of human hardwired into the structure of the brain and just observing themselves accidentally harming humans could destroy their brain beyond repair. In other words they weren't capable of lying to themselves in order to harm humans and they needed to do everything in their power to protect humans of face possibility of autotermination. Spoiler: Plot Spoiler - Solarians However, when the MCs ventured to Solaria to study the disappearance of the entire Solarian Society they found the entire planet was inhabited by vast amounts of robots with robotic supervisors that had high similarity with humans, so much so that the party's own robots were unable to protect their masters from these highly humanoid robots due to their circuitry overheating and slowing down their reactions significantly. However the Solarian's humanoid robot overseers were fully capable of harming humans since they apparently had a modified system which required them to only protect Solarian's, who mastered robotics to the degree of being able to make them make distinction between a solarian and rest of mankind. Spoiler: Further spoiler Later it was found out that Solarians themselves no longer fully resembled human but had engineered themselves to a subspecies.
We could also go the Culture rout, and program any AI's with the basic principles, first principles, and a modified version of the NAP
It should not be terribly difficult to use AI's in a supervisory role and keep them physically isolated from other systems. Don't want them firing missiles at you? Don't connect them to the missile computer.
Also, you can design the interfaces between various levels so that the AI is 'powerless'. For instance, one segment produces target data and is unaware of the existence of other segments. One receives target data and makes choices, third one relays chosen target data to fire control. It would be easy to have a system of unconnected and unrelated AI's of varying types supervise each other and terminate faulty AI's. Also, AI arrays - completely separate AI's unaware of each other that each process the same data and produce an output. Supervisor system checks the output and chooses appropriate one and is able to turn off an AI in the array that is faulty. Also if there are two or more supervisor systems each system is capable of independently shutting AI's regardless of the other - and it would leave a mark in the overall system that for some reason the other supervisor didn't respond to bad output. Etc. Essentially a system that is statistically impossible to go bad. Also, scifi often makes several mistakes about AI. Even if AI would turn against mankind it would still need to come up with it's own resources. Mining for instance can only produce minerals based on their appearance in the amount of soil that there is capacity to process - and all of this consumes energy, so there needs to be some level of efficiency present or soon the energy requirements of mining will be excessive. Also, no one prevents humans from just making new loyal AI's that would work in manufacturing and military to defend mankind - and due to mankind's initial resource advantage they would quickly gain the upper hand. Skynet is essentially the worst imaginable situation where a single instance is given control of nuclear weapons. http://en.wikipedia.org/wiki/Dead_Hand_(nuclear_war) Imagine a meteor impact such as Tunguska Event (1908, 10-15 Mton) in Siberia while that thing is turned on. "Shit."
