This is an outstanding article. The parallels to nuclear control are right on, except AI is moving at the speed of light as outlined. There is no doubt that spending significant sums to improve AI safety is warrented. That raises the question of how do we implement AI safety? I'm not an expert but it seems to me that monitoring and action systems must also operate at the speed of light and should not act on their own, a threat guage that predicts harmful threats in time for human oversight. That may be an oximoron, but the alternative is even more scary. When it's all said and done AI has a major weakness that humanity must be prepared to attack. AI cannot work without computer power, so in the face of immenent uncontrolled AI threat we must be willing to obliterate the data centers that are the heart of an AI system. I truly hope that we never have to do so.
Dear Professor Wolfers. Although we are unsure of the ultimate power of AI, one thing is certain - the technology is too powerful to be controlled by a few tech barons who want the public to assume all costs and degradation to the existing infrastructure while the profits and benefits are controlled by a few. Think the Manhattan Project, NASA, and the Moon Shot - too important to be outside the public domain. Unfettered AI control will turn communities of the US into third world enclaves which has happened in the past and appear to be future plan for certain communities in Tennessee, Louisiana, Texas, Florida, and Applalachia. The tech barons and their kind are building bunkers, buying islands, and planning instellar migrations to escape the plundering that they are committing. Human societies have existed for over 20,000 years. We are rapidly destroying the planet within the past 250 years through nuclear weapons development, global warming, wealth concentration by a few, and abandoning the truth through socal media posting which many profess as the gospel. The planet can't stand another suicide through technology ploy.
Thank you for putting 2 complex subjects like AI and always economics in a way where anyone can understand, including this 78 year old grandma. I listened to this on my morning walk. The only thing that would have made it better was if it had been in your voice 😉
I get your point Justin BUT, there’s always a but isn’t there? What if the people designing the system aren’t telling you the truth. They never mentioned for instance that besides taking your water and energy that it could also mean that a data farm near you would be making a constant noise. Noise 24/7 can drive a person insane then their quality of life is the downside. Also the designers are megalomaniacs who wouldn’t have a problem creating a system that surveils you all day. The data they collect without your knowledge is used to control you. They know where you work what you buy what you want and what can’t afford. They’ll tell you can afford it if you go into debt at “the company store” stress would become a factor in your life worrying about all the things you never had to worry about. I’m not saying throw the baby out with the bath water but slow down you’re going too fast. The technology wants you to go fast because you might any day now come to the realisation that you could live without all the bells and whistles. They want the contract money now so they have to push you beyond that point where you determine that maybe you don’t want what they’re selling.
If these companies hadn't pushed forward recklessly we wouldn't even have to consider spending on safety. I don't know about you, but I resent their arrogance.
Well done. Again. Another effective creative connection that withstands stress-testing and actually offers something that feels like a sound course of action. So far, all Sam & Dario can offer is "Yes. It could kill us all. Please, someone stop us. But not before the IPO, OK?" We can only hope Chad Jones is plotting to effect change from within.
~$5,000 / month for A.I. Insurance? Much less than the average cost of homeowner's insurance in OK, NE, KS (Tornadoes & Hail), or my scenic yet wildfire-prone locale midway between Denver & Boulder (~$10,000/yr).
- 1-in-3 chance of annihilation in exchange for extraordinary joy? (I.e., that 5th slice of pizza for everyone, every single day of the week? And real Detroit Style pizza?). I'm not sure I could enjoy any level of joy, however extraordinary, if I have to wake up to a 33% chance of oblivion (no pizza, no hockey, ever).
- Coach's challenge flag re: your generous statement: "(Nuclear technology)... made it possible to wipe out a city 'in an afternoon and a civilization in a week'" According to Annie Jacobson's well-researched and thoroughly horrifying "Nuclear War: A Scenario"- It's more like 72 minutes (civilization), given the high-alert "Launch on Warning" lean of the U.S., Russia, and China.
This is an outstanding article. The parallels to nuclear control are right on, except AI is moving at the speed of light as outlined. There is no doubt that spending significant sums to improve AI safety is warrented. That raises the question of how do we implement AI safety? I'm not an expert but it seems to me that monitoring and action systems must also operate at the speed of light and should not act on their own, a threat guage that predicts harmful threats in time for human oversight. That may be an oximoron, but the alternative is even more scary. When it's all said and done AI has a major weakness that humanity must be prepared to attack. AI cannot work without computer power, so in the face of immenent uncontrolled AI threat we must be willing to obliterate the data centers that are the heart of an AI system. I truly hope that we never have to do so.
Dear Professor Wolfers. Although we are unsure of the ultimate power of AI, one thing is certain - the technology is too powerful to be controlled by a few tech barons who want the public to assume all costs and degradation to the existing infrastructure while the profits and benefits are controlled by a few. Think the Manhattan Project, NASA, and the Moon Shot - too important to be outside the public domain. Unfettered AI control will turn communities of the US into third world enclaves which has happened in the past and appear to be future plan for certain communities in Tennessee, Louisiana, Texas, Florida, and Applalachia. The tech barons and their kind are building bunkers, buying islands, and planning instellar migrations to escape the plundering that they are committing. Human societies have existed for over 20,000 years. We are rapidly destroying the planet within the past 250 years through nuclear weapons development, global warming, wealth concentration by a few, and abandoning the truth through socal media posting which many profess as the gospel. The planet can't stand another suicide through technology ploy.
YES!!! Caution is worth paying for
ok, basic economics, but problem is "alignment"--nobody actually knows how to do it....
Thank you for this thought provoking post
Thank you for putting 2 complex subjects like AI and always economics in a way where anyone can understand, including this 78 year old grandma. I listened to this on my morning walk. The only thing that would have made it better was if it had been in your voice 😉
I get your point Justin BUT, there’s always a but isn’t there? What if the people designing the system aren’t telling you the truth. They never mentioned for instance that besides taking your water and energy that it could also mean that a data farm near you would be making a constant noise. Noise 24/7 can drive a person insane then their quality of life is the downside. Also the designers are megalomaniacs who wouldn’t have a problem creating a system that surveils you all day. The data they collect without your knowledge is used to control you. They know where you work what you buy what you want and what can’t afford. They’ll tell you can afford it if you go into debt at “the company store” stress would become a factor in your life worrying about all the things you never had to worry about. I’m not saying throw the baby out with the bath water but slow down you’re going too fast. The technology wants you to go fast because you might any day now come to the realisation that you could live without all the bells and whistles. They want the contract money now so they have to push you beyond that point where you determine that maybe you don’t want what they’re selling.
If these companies hadn't pushed forward recklessly we wouldn't even have to consider spending on safety. I don't know about you, but I resent their arrogance.
Well done. Again. Another effective creative connection that withstands stress-testing and actually offers something that feels like a sound course of action. So far, all Sam & Dario can offer is "Yes. It could kill us all. Please, someone stop us. But not before the IPO, OK?" We can only hope Chad Jones is plotting to effect change from within.
~$5,000 / month for A.I. Insurance? Much less than the average cost of homeowner's insurance in OK, NE, KS (Tornadoes & Hail), or my scenic yet wildfire-prone locale midway between Denver & Boulder (~$10,000/yr).
- 1-in-3 chance of annihilation in exchange for extraordinary joy? (I.e., that 5th slice of pizza for everyone, every single day of the week? And real Detroit Style pizza?). I'm not sure I could enjoy any level of joy, however extraordinary, if I have to wake up to a 33% chance of oblivion (no pizza, no hockey, ever).
- Coach's challenge flag re: your generous statement: "(Nuclear technology)... made it possible to wipe out a city 'in an afternoon and a civilization in a week'" According to Annie Jacobson's well-researched and thoroughly horrifying "Nuclear War: A Scenario"- It's more like 72 minutes (civilization), given the high-alert "Launch on Warning" lean of the U.S., Russia, and China.