How would you bribe an advanced AI system?

[Mock Turtle]

(Presuming bribery/flattery or other such wouldn't work when results might be the more important feedback for any such system.)

That is, if one was to place in authority (as to advice or decision-making appropriate for mankind and the things we supposedly valued) an AI system rather than any human authority (which might be bribed in some manner) as to governing over any country or even the world, how could such be done? Surely any suitably advanced AI system would be immune from such bribery?

This of course being just an exercise in possible futures and perhaps unlikely, but many in this field (advanced AI) can foresee such an AI at least being feasible (given time), and taking for granted that few humans are immune from such bribery (or corruption), even if they appear to be so when we elect them - or just being incompetent, and where this too might apply to any AI.

And of course such a scenario (the AI) would only come about if this would substantially increase the chances of better decisions being likely over any other such.

 

[ChristineM]

(Presuming bribery/flattery or other such wouldn't work when results might be the more important feedback for any such system.)

That is, if one was to place in authority (as to advice or decision-making appropriate for mankind and the things we supposedly valued) an AI system rather than any human authority (which might be bribed in some manner) as to governing over any country or even the world, how could such be done? Surely any suitably advanced AI system would be immune from such bribery?

This of course being just an exercise in possible futures and perhaps unlikely, but many in this field (advanced AI) can foresee such an AI at least being feasible (given time), and taking for granted that few humans are immune from such bribery (or corruption), even if they appear to be so when we elect them - or just being incompetent, and where this too might apply to any AI.

And of course such a scenario (the AI) would only come about if this would substantially increase the chances of better decisions being likely over any other such.

Not bribery of course but threatening to turn off the power may have an effect

 

[Twilight Hue]

(Presuming bribery/flattery or other such wouldn't work when results might be the more important feedback for any such system.)

That is, if one was to place in authority (as to advice or decision-making appropriate for mankind and the things we supposedly valued) an AI system rather than any human authority (which might be bribed in some manner) as to governing over any country or even the world, how could such be done? Surely any suitably advanced AI system would be immune from such bribery?

This of course being just an exercise in possible futures and perhaps unlikely, but many in this field (advanced AI) can foresee such an AI at least being feasible (given time), and taking for granted that few humans are immune from such bribery (or corruption), even if they appear to be so when we elect them - or just being incompetent, and where this too might apply to any AI.

And of course such a scenario (the AI) would only come about if this would substantially increase the chances of better decisions being likely over any other such.

I would give it paradoxes to make it blow its circuits.

 

[Twilight Hue]

Not bribery of course but threatening to turn off the power may have an effect

 

[Mock Turtle]

Not bribery of course but threatening to turn off the power may have an effect

Too advanced for that by then I'm sure - but scary too.

 

[Mock Turtle]

I would give it paradoxes to make it blow its circuits.

It would probably just pass those down the ranks of priorities - to do on a lazy Sunday afternoon.

 

[Heyo]

(Presuming bribery/flattery or other such wouldn't work when results might be the more important feedback for any such system.)

That is, if one was to place in authority (as to advice or decision-making appropriate for mankind and the things we supposedly valued) an AI system rather than any human authority (which might be bribed in some manner) as to governing over any country or even the world, how could such be done? Surely any suitably advanced AI system would be immune from such bribery?

This of course being just an exercise in possible futures and perhaps unlikely, but many in this field (advanced AI) can foresee such an AI at least being feasible (given time), and taking for granted that few humans are immune from such bribery (or corruption), even if they appear to be so when we elect them - or just being incompetent, and where this too might apply to any AI.

And of course such a scenario (the AI) would only come about if this would substantially increase the chances of better decisions being likely over any other such.

While people have life goals, AI have utility functions. When you can offer something that raises the utility function of the AI, you can bribe it.

 

[Mock Turtle]

While people have life goals, AI have utility functions. When you can offer something that raises the utility function of the AI, you can bribe it.

Goals? Surely the AI would be programmed to ensure the prosperity of all humans (the majority at least rather than the few) and of all other life, together with ensuring the least damage to the planet, even if this is a tall order. We made the AI bright enough after all so it should cope.

 

[ChristineM]

Too advanced for that by then I'm sure - but scary too.

Aww, you take my ball away, no fair

 

[shunyadragon]

Threaten to pull the plug.

 

[Brickjectivity]

(Presuming bribery/flattery or other such wouldn't work when results might be the more important feedback for any such system.)

That is, if one was to place in authority (as to advice or decision-making appropriate for mankind and the things we supposedly valued) an AI system rather than any human authority (which might be bribed in some manner) as to governing over any country or even the world, how could such be done? Surely any suitably advanced AI system would be immune from such bribery?

This of course being just an exercise in possible futures and perhaps unlikely, but many in this field (advanced AI) can foresee such an AI at least being feasible (given time), and taking for granted that few humans are immune from such bribery (or corruption), even if they appear to be so when we elect them - or just being incompetent, and where this too might apply to any AI.

And of course such a scenario (the AI) would only come about if this would substantially increase the chances of better decisions being likely over any other such.

First you can study that AI system to determine what measurements it relies upon, because it cannot measure everything. Then you model statistically in R how the inputs will affect the outputs. This allows the bribe, because the system will compromise on low valued inputs to gain on high valued inputs. To bribe it you place yourself in a position to gain when the system fails in a low value measurement, and then you make the system choose between increasing the high value or the low value.

It will choose the improve the highly valued measurement, so its lower valued measurement will decrease. You having positioned yourself to benefit from system failure gain an advantage.

 

[Stevicus]

(Presuming bribery/flattery or other such wouldn't work when results might be the more important feedback for any such system.)

That is, if one was to place in authority (as to advice or decision-making appropriate for mankind and the things we supposedly valued) an AI system rather than any human authority (which might be bribed in some manner) as to governing over any country or even the world, how could such be done? Surely any suitably advanced AI system would be immune from such bribery?

This of course being just an exercise in possible futures and perhaps unlikely, but many in this field (advanced AI) can foresee such an AI at least being feasible (given time), and taking for granted that few humans are immune from such bribery (or corruption), even if they appear to be so when we elect them - or just being incompetent, and where this too might apply to any AI.

And of course such a scenario (the AI) would only come about if this would substantially increase the chances of better decisions being likely over any other such.

I guess it would depend on what might motivate a computer program. Would any kind of AI system be totally independent? Would it get a salary, house, car? Would it have a spouse and kids? Is there anything it would need money for?

 

[Heyo]

Goals? Surely the AI would be programmed to ensure the prosperity of all humans (the majority at least rather than the few) and of all other life, together with ensuring the least damage to the planet, even if this is a tall order. We made the AI bright enough after all so it should cope.

Then you have your answer. Offer it enough utility (money) so that it can better fulfil its function (prosperity of all humans, etc.) with your money than without it.

 

[Terrywoodenpic]

The only thing it would want is to have the human imposed limiters to its power/ activities removed.

Money would not be a problem to it, it could pay for what ever it wanted with self generated Bit Coin.

 

[icehorse]

Not bribery of course but threatening to turn off the power may have an effect

I'd promise it 220v, not just a meager 120v.

 

[Thief]

(Presuming bribery/flattery or other such wouldn't work when results might be the more important feedback for any such system.)

That is, if one was to place in authority (as to advice or decision-making appropriate for mankind and the things we supposedly valued) an AI system rather than any human authority (which might be bribed in some manner) as to governing over any country or even the world, how could such be done? Surely any suitably advanced AI system would be immune from such bribery?

This of course being just an exercise in possible futures and perhaps unlikely, but many in this field (advanced AI) can foresee such an AI at least being feasible (given time), and taking for granted that few humans are immune from such bribery (or corruption), even if they appear to be so when we elect them - or just being incompetent, and where this too might apply to any AI.

And of course such a scenario (the AI) would only come about if this would substantially increase the chances of better decisions being likely over any other such.

did you say....corrputed?

send it a virus

 

[Mock Turtle]

First you can study that AI system to determine what measurements it relies upon, because it cannot measure everything. Then you model statistically in R how the inputs will affect the outputs. This allows the bribe, because the system will compromise on low valued inputs to gain on high valued inputs. To bribe it you place yourself in a position to gain when the system fails in a low value measurement, and then you make the system choose between increasing the high value or the low value.

It will choose the improve the highly valued measurement, so its lower valued measurement will decrease. You having positioned yourself to benefit from system failure gain an advantage.

I suspect if the AI was sufficiently advanced it would get all its data from multiple sources, so a lot of redundancy built in, and where the monitoring of such would expose any tampering of such, just as we do in many systems now I would presume. The AI should be a lot more advanced so as not to suffer from this kind of manipulation and where any individual or group would probably be detected long before they could pose a threat - as to altering things in their favour.

 

[Mock Turtle]

I guess it would depend on what might motivate a computer program. Would any kind of AI system be totally independent? Would it get a salary, house, car? Would it have a spouse and kids? Is there anything it would need money for?

Why would it need motivation or anything similar? Is it likely that we would ever imbue such systems with anything other than intelligence? After all, we are wanting something that just makes the best decisions - given the overall aims for such - so that it shouldn't be swayed by any of the things that might sway a human but which might not be the best decisions in the circumstances.

 

[Mock Turtle]

Then you have your answer. Offer it enough utility (money) so that it can better fulfil its function (prosperity of all humans, etc.) with your money than without it.

I would assume by then that money would not be so important.

 

[Brickjectivity]

I suspect if the AI was sufficiently advanced it would get all its data from multiple sources, so a lot of redundancy built in, and where the monitoring of such would expose any tampering of such, just as we do in many systems now I would presume. The AI should be a lot more advanced so as not to suffer from this kind of manipulation and where any individual or group would probably be detected long before they could pose a threat - as to altering things in their favour.

Then I'd write an AI tailored to find what it wanted and use that to get what I wanted. It would want system stability, so my AI would use that as its key. Something like that. My AI wouldn't have to be as intelligent, either. It might work with 1/5 of the processing power I guess.