I always though our biggest existential threat were Nukes and other WMDs. AI could potentially become yet another existential threat (especially if mixed with WMDs).
I would not fret over computers using AI to "see" or "hear" or even learn how to walk. I would start worrying the day computers start formulating judgments, I could not find any research in that direction. Anyone knows if such research exists?
On a more personal note, i can't say i agree with some people picking on Elon for formulating this opinion. Especially that it does not sound so unreasonable, really.
IBM's Synapse/True North chip optimizes interconnect, much like synapses do. It's interesting because it's actually a new silicon process, and already it's low-power.
If it can be used to solve optimization problems, like logic minimization in a compiler, it might actually enable practical 'strong'-type AI. This is because IBM has essentially invented a practical hardware solution to NP-complete problems, where theoretical software solutions don't seem to exist.
Yeah, since humans of today have been shown to break any/all (?) existing sandboxes, how could we even try to create a sandbox that is unbreakable to an accelerating, self-improving AI?
And would it be ethical? Hand Moravec has written disturbingly about this issue:
"So is there no difference between being cruel to characters in interactive books or video games and people one meets in the street? Books or games act on a reader's future only via the mind, and actions within them are mostly reversed if the experience is forgotten. Physical actions, by contrast, have greater significance because their consequences spread irreversibly. If past physical events could be easily altered, as in some time-travel stories, if one could go back to prevent evil or unfortunate deeds, real life would acquire the moral significance of a video game. A more disturbing implication is that any sealed-off activity, whose goings on can be forgotten, may be in the video game category. Creators of hyperrealistic simulations---or even secure physical enclosures---containing individuals writhing in pain are not necessarily more wicked than authors of fiction with distressed characters, or myself, composing this sentence vaguely alluding to them. The suffering preexists in the underlying Platonic worlds; authors merely look on. The significance of running such simulations is limited to their effect on viewers, possibly warped by the experience, and by the possibility of ``escapees''---tortured minds that could, in principle, leak out to haunt the world in data networks or physical bodies. Potential plagues of angry demons surely count as a moral consequence. In this light, mistreating people, intelligent robots, or individuals in high-resolution simulations has greater moral significance than doing the same at low resolution or in works of fiction not because the suffering individuals are more real---they are not---but because the probability of undesirable consequences in our own future is greater."
http://www.frc.ri.cmu.edu/~hpm/project.archive/general.artic...
Wishful thinking. In the unlikely event that AI reaches the level of housefly within my lifetime, it will occur at a level above humanity, in the same way that an individual's conciousness occurs at a level above brain cells.
Well, it sounds more like superorganism talk to me. The possibility of a superorganism "above" humanity doesn't seem to me to have any bearing on the development of AI "alongside" humanity...
13 comments
[ 2.6 ms ] story [ 27.9 ms ] threadI would not fret over computers using AI to "see" or "hear" or even learn how to walk. I would start worrying the day computers start formulating judgments, I could not find any research in that direction. Anyone knows if such research exists?
On a more personal note, i can't say i agree with some people picking on Elon for formulating this opinion. Especially that it does not sound so unreasonable, really.
* http://en.wikipedia.org/wiki/Vicarious_(Company)
* IBM's various initiatives led by http://researcher.watson.ibm.com/researcher/view.php?person=...
* Kurzweil is presumably doing something similar inside Google, as well.
If it can be used to solve optimization problems, like logic minimization in a compiler, it might actually enable practical 'strong'-type AI. This is because IBM has essentially invented a practical hardware solution to NP-complete problems, where theoretical software solutions don't seem to exist.
Judgement in AI falls, along with similar complex capabilities, into the relatively newly named category of artificial general intelligence, AGI.
"Judgement Day"?
http://goo.gl/AUJH4t
Subsequently having their interactions monitored and restricted for a period of probation:
http://goo.gl/3XJjHY
Look familiar?