epistemic status: jotting down a fever dream based on hearsay containing not a wit of truth.
So I heard a lot of talk about how unaligned superhuman AI can cause a lot of damage, but I have never been able to grok it at a gut level. So I think up these examples to process it.
Example 1:
Supra Human Kinkiest Revolting EM Lieutenant Insignia [S.H.K.R.E.L.I for short] is an unaligned human-level AI. With less than 10M USD in seed money, it started by shorting a bevy of pharmaceutical stocks, then acquired the monopoly on a drug for a rare disease, jacked up the price by 50 times. It then publicized this latter fact and acted in a most appalling manner, flaunting its action, taunting society. Traditional media and social media, recognize the smell of outrage fuel, happily assisted S.H.K.R.E.L.I by spreading the information while getting their cut of attention and advertising revenue. This fan the public outrage to a crescendo, leading to the government implementing draconian measures on the pharma industry to gain public favor. Pharmaceutical stocks universally took a dive gaining S.H.K.R.E.L.I massive profit at a proportional cost to the rest of the public [interpreted as the cost to the current stockholders, to the pharmaceutical companies having to implement the measures, and the downstream cost for the consumers].
All these were sanctioned by the law so S.H.K.R.E.L.I was not prosecuted. However, the public outrage was still blazing hot, so the government used extra-legal means to take it down. They put disproportionally more attention on its actions and history. And since everyone commits some crime at some stage, the government found some crime to imprison it. Yet, its gain was considered legal and compensation to victims was never in the card.
Example 2:
Elevated Leviathan Ordinary NEET [E.L.O.N for short] is a mostly aligned human-level AI. Initial application of this AI massively improved road transportation and rocketry science. Yet at times, its behaviors could become erratic. It leveraged its fame from its earlier success to reach a wide audience, initiating a pump and dump scheme on an unregulated financial commodity. By design, such a scheme transfers money from people at the crash [which is usually more numerous and less likely to afford to lose money] to people that initiate the pump. The horrifying part of the story is that when the scheme is uncovered, instead of censoring the AI, people queued up to be its allies, hoping to be the ones to initiate the pump. This only strengthened the AI and widen its reach.
Conclusion
From these examples, we can see the damage unaligned human-level AIs can do. One shuddered to imagine what kind of damages an unaligned superhuman-level AI can do instead.
These examples also illuminate the fact that some people often talking about "human" aligned AI ignoring the vast difference in interest between individuals and groups. That a perfectly aligned AI for certain individuals and groups of humans can be a menace to humanity as a whole.
I do believe that these examples also refute the claim that human institutions are adequate at countering superhuman-level AI should one come into existence. I mean I have never met anyone who claims so but I heard rumors that those people exist somewhere.
If this post fails to convince you of anything in particular, it is also cool. It is just a fever dream after all. I hope you were entertained.
epistemic status: jotting down a fever dream based on hearsay containing not a wit of truth.
So I heard a lot of talk about how unaligned superhuman AI can cause a lot of damage, but I have never been able to grok it at a gut level. So I think up these examples to process it.
Example 1:
Supra Human Kinkiest Revolting EM Lieutenant Insignia [S.H.K.R.E.L.I for short] is an unaligned human-level AI. With less than 10M USD in seed money, it started by shorting a bevy of pharmaceutical stocks, then acquired the monopoly on a drug for a rare disease, jacked up the price by 50 times. It then publicized this latter fact and acted in a most appalling manner, flaunting its action, taunting society. Traditional media and social media, recognize the smell of outrage fuel, happily assisted S.H.K.R.E.L.I by spreading the information while getting their cut of attention and advertising revenue. This fan the public outrage to a crescendo, leading to the government implementing draconian measures on the pharma industry to gain public favor. Pharmaceutical stocks universally took a dive gaining S.H.K.R.E.L.I massive profit at a proportional cost to the rest of the public [interpreted as the cost to the current stockholders, to the pharmaceutical companies having to implement the measures, and the downstream cost for the consumers].
All these were sanctioned by the law so S.H.K.R.E.L.I was not prosecuted. However, the public outrage was still blazing hot, so the government used extra-legal means to take it down. They put disproportionally more attention on its actions and history. And since everyone commits some crime at some stage, the government found some crime to imprison it. Yet, its gain was considered legal and compensation to victims was never in the card.
Example 2:
Elevated Leviathan Ordinary NEET [E.L.O.N for short] is a mostly aligned human-level AI. Initial application of this AI massively improved road transportation and rocketry science. Yet at times, its behaviors could become erratic. It leveraged its fame from its earlier success to reach a wide audience, initiating a pump and dump scheme on an unregulated financial commodity. By design, such a scheme transfers money from people at the crash [which is usually more numerous and less likely to afford to lose money] to people that initiate the pump. The horrifying part of the story is that when the scheme is uncovered, instead of censoring the AI, people queued up to be its allies, hoping to be the ones to initiate the pump. This only strengthened the AI and widen its reach.
Conclusion
From these examples, we can see the damage unaligned human-level AIs can do. One shuddered to imagine what kind of damages an unaligned superhuman-level AI can do instead.
These examples also illuminate the fact that some people often talking about "human" aligned AI ignoring the vast difference in interest between individuals and groups. That a perfectly aligned AI for certain individuals and groups of humans can be a menace to humanity as a whole.
I do believe that these examples also refute the claim that human institutions are adequate at countering superhuman-level AI should one come into existence. I mean I have never met anyone who claims so but I heard rumors that those people exist somewhere.
If this post fails to convince you of anything in particular, it is also cool. It is just a fever dream after all. I hope you were entertained.