Ethan Mollick says scaling laws make the latest developments exciting; The Zvi is more skeptical; Scott Alexander asks what AI's have to do to alarm us; Mark McNeilly on Larry Ellison's vision
The police don’t even follow up with most crime right now. Property crime like your house or car getting broken into, even if you have clear video of the person? Forget it.
So imagine how the entire system would collapse if it effectively recorded every crime, everywhere.
What would actually happen is no enforcement most of the time. And then one day, if you were in the out group of who ever was in power, or if the people in charge wanted to lean on you to do something else, they’d charge you for all the previous crimes at once so you’d have a scary sentence, forcing you to collaborate.
In East Germany they only had the technical capability of tapping maybe 10 or 20 phones at once and look at all the data they had at the end when the system collapsed. Everyone has the right now to look at their file and see which of their neighbors snitched on them, or vice versa if they were the snitch.
You can’t arrest everyone. If that becomes obvious, then I think crime would skyrocket.
Supposedly it’s a power law where some small group of people commit the vast majority of serious crimes, but we don’t even acknowledge that and put those obvious psychopaths away for 20 years or whatever to keep the regular people safe.
The police don't follow up because that's policy, not because they would be overwhelmed. If crime were detected everywhere, it's not the system that would collapse, it's crime. Crime has already skyrocketed compared to post-war levels, not because the police can't arrest all the criminals, but because it is chosen policy not to do so.
Do not mistake a decision not to use state capacity for a lack of state capacity. The US (and most states and cities on their own) has absolutely terrifyingly gargantuan levels of state capacity for mass surveillance and law enforcement, and when it really wants to and finds the right Hawaiian judge to bless off on it, the US regime finds it absolutely trivial to crush anything at any scale.
I think his point is more that universal surveillance would only be used to gather information on people the state might like to punish for political reasons instead of for breaking the law only.
Agreed they chose not to pursue enforcement of crime by known offenders. Disagree they have the capacity to do much more than they are doing but they could potentially change the level of crime by increasing enforcement and then have the capacity to broaden enforcement.
It makes no sense to talk about "capacity" unless one disaggregates the actual limit of physical, empirical capacity (which was FAR above what is required to crush all crime a long time ago) from socially-constructed reductions in capacity, which are made nearly as big, because our regime cares less about stopping crime than it does about not stopping the wrong kind of criminals.
Back before glorious equality turned "standards, no compromise" into "the standard is compromise", the army once required soldiers to be able to carry an injured 90th percentile weight comrade 100 meters to evacuate him to safety. Most fit young men when trained and physically conditioned have the - capacity- to do so, so long as one doesn't normalize renormalizatiom and the 90th percentile isn't allowed to balloon, as it were, to double what it used to be.
But what is the "capacity" to do so when you are also required to carry your ruck sack, weapon, and the heavy mortar base plate, as was (once) a fun challenge in some of the more elite, selective units? Much reduced! Maybe only 2-3% of Joes can do it now. Maybe nobody can and you have to settle for not being able to evacuate anybody over the 50th percentile, so sorry Tall Tommy.
And what about if Joe shoots himself in the foot? Not much "capacity" now, I'm afraid, not even to drop everything and hobble to save just his own life!
One only makes the mistake of thinking the mirage of "low capacity" is a real limitation when one accepts repeated treatment by emergency surgical podiatry for self- inflicted wounds as somehow unavoidable and "normal" instead of a continuous series of conscious choices, as is it was something that just happened to us like a bad weather event beyond anyone's control. It's definitely in control, and being throttled to ensure the persistance of high crime rates.
Agreed that if there were less socially-constructed reductions in capacity, enforcement could be broadened. I guess it depends on the definition of "crush all crime" but I lean more towards potentially making substantial reductions, maybe even more than half.
Re: "widespread surveillance to detect infractions would have chaotic effects unless laws were carefully rewritten to somehow spell out the behavior that we really want."
A deep point!
It invites fresh thinking/research about fundamental topics:
The static analysis is that there would be way too much and too severe punishment of minor infractions. The dynamic analysis is that behaviors would change (risk compensation goes both ways) and to the extent it doesn't, intolerable situations create the pressures necessary to return matters to a tolerable state and that laws and penalties (de jure or de facto) would be adjusted to accommodate the transition to a high vs low detection rate law enforcement environment.
Which ... is already what happened, over half a century ago. There is no need fit """AI""" a term getting misused all the time and which at any rate is just the latest wrinkle. We've been living in world in which affordable technology well-deployed and utilized has made it feasible to detect all kinds of crimes at near 100% rate, if the powers that be want to. Crime is a solved problem; we are choosing not to solve it. Go ask China about their powerful, pre-"""AI""" detection system and low crime rates, especially for things the regime actually really cares about. Or go see the extent to which the US regime will go to use existing technology successfully to identify, find, and round up literally every single person involved in an event it cares about. But what the US regime also cares about is NOT having too much punishment of the wrong kind of criminal, and so for decades has rewritten, adjusted, and effectively relaxed the law and simultaneously crippled the capabilities and ratcheted up the burden on law enforcement in those cases to absurdly obscene levels, or perform mass abuse of discretion to simply refuse to investigate and prosecute crimes, to conserve "group- based enforcement homeostasis" and keep up with and neutralize every advancement in or enhancement of detection capability lest too many of the wrong kind of criminals get locked away for too long. They even decide to just let people accumulate tens of thousands of dollars in traffic violation tickets and still renew their licenses and vehicle registrations if the alternative would mean arresting, or, heck, even inconveniencing the wrong kind of people too much.
I must disagree with Arnold's example of speed limits. I have the actual experience of having a morning commute through a suburban part of the city in one of those situations in which some local road with a 30 mph limit starts getting used as a major artery despite being lined with houses with kids waiting for school busses. The road became extremely dangerous, cars often exceeding 45 mph, with too many close calls and a few ugly tragedies. The residents got angry and the parents of young children especially so, and demanded action. It began with patrol cars, but that's using a lot of policing resources, and the nature of the road left little space to park such a vehicle for speed trapping, so it didn't end up accomplishing much. Then the city installed a speed camera, not the "only if 5 or 10 mph above the limit" but for real painful tickets for even 1 mph over. Boom, thousands of tickets per week. And then hundreds, dozens, just a few. Yeah, drivers groused in outrage at first (boy, did they ever!) but eventually they just got used to it, adjusted behaviors (slowed down, changed commute routes, took the bus instead, etc.) and the road and neighborhood is 100 times safer.
The point is, in the present legal, political, and cultural context, we should be much more afraid of selective enforcement decided by politicians and judges than universal enforcement decided by machine. In a "machine regime", it universal enforcement is intolerable, the pressure of new detection tech will be to change the letter of the law, to winnow down the "three felonies a day" problem. In the political regime, the pressure is to just ignore the letter of the law for your friends and let crime explode if necessary, and also to go full Beria, use national security-level capabilities to surveil you enemies, find one of a million possible "crimes" they committed, and throw the book at them.
You know how the AIs are gonna take over? We are going to let them, elect them. We are going to compare them to trash human politicians and conclude it's just no contest. They've got my vote already!
"When a junior team member joins your company, you teach them to follow your processes. When a senior leader takes the reins, you expect them to have their own. ChatGPT wants to be the senior leader, given free rein on your tasks, using its own methods to find the right response."
Who is the driver/ master/ direction-setter/ senior? You or the AI--THAT is the question, or soon will be. I want a highly skilled assistant-servant to do what I want, and be able to quickly understand what I want, plus the ability to do it. Doing things in the right way but doing the things I think are the right things.
Enforcement of many laws will be primarily, if not exclusively, against regime opponents. Like HR Clinton illegal email sever, DoJ ok, vs anti-Trump lawfare.
There is a human social need to have softer & harder boundaries, including taboos that can be violated with much less negative results. If all sex with consent is ok, what about consent to be in a situation where one can’t consent, nor object? This is not that, just pure drugging a wife and video taping more than 83 different men having sex with her. Most have defense they didn’t consider it rape.
The ease, and greater frequency, of using freedom to act which is harmful to others is why norms & laws limit freedom. Even many libertarians accept laws against drunk driving due to much higher probability of hurting others. AI super surveillance is likely to increase with cars, first. “But officer, I don’t think I was more than 10 mph over the limit”.
One area where there should be More enforcement is laws against govt misconduct, not only police but especially bureaucrats who have a whole slew of requirements they often don’t follow. AI support on knowing and quickly referencing requirements can help supervisors ensure compliance.
" ... never be able to do X. Then a computer does X. Then we rationalize that the doing X is no big deal."
I'm not sure "no big deal" accurately captures what is happening. I believe the expectation is something closer to a human type of intelligence and when it is explained as Big Blue looking at all the possible chess moves or AI looking through information to find a likely fit, we decide those processes are more mechanical than intelligent.
It's not so much a matter of the red line moving as it is we are judging how the red line is reached.
" ... never be able to do X. Then a computer does X. Then we rationalize that the doing X is no big deal."
Nietzsche's Last Man will make The Last Rationalization, "Even though I recently said it would be, the replacement of the last man before me was not actually impressive or qualifying as 'real' intelligence, however, replacing me requires real intelligence totally beyond the potential capabilities of extensions of current approaches, and thus won't happen."
There isn't a strict line because the physical and spiritual worlds map onto each other—advanced physical systems can become infused with spiritual realities. Computers, being blank slates, are vulnerable to evil influence, as we've seen with the internet. AI is potentially even more dangerous, since the more brain-like a system becomes, the easier it is for dark forces to "ensoul" it. That’s why AI development might pose a bigger risk than the already problematic internet.
I actually wrote an article about AI that has no religious or spiritual content, but it explores these ideas through a secular lens. Rebbetzin Fastag agreed to let me post it after we discussed the potential risks, particularly around black-box systems, which could unknowingly invite dark forces. Here's the link if you're interested: https://ishayirashashem.substack.com/p/artificial-intelligence-for-dummies
The police don’t even follow up with most crime right now. Property crime like your house or car getting broken into, even if you have clear video of the person? Forget it.
So imagine how the entire system would collapse if it effectively recorded every crime, everywhere.
What would actually happen is no enforcement most of the time. And then one day, if you were in the out group of who ever was in power, or if the people in charge wanted to lean on you to do something else, they’d charge you for all the previous crimes at once so you’d have a scary sentence, forcing you to collaborate.
In East Germany they only had the technical capability of tapping maybe 10 or 20 phones at once and look at all the data they had at the end when the system collapsed. Everyone has the right now to look at their file and see which of their neighbors snitched on them, or vice versa if they were the snitch.
You can’t arrest everyone. If that becomes obvious, then I think crime would skyrocket.
Supposedly it’s a power law where some small group of people commit the vast majority of serious crimes, but we don’t even acknowledge that and put those obvious psychopaths away for 20 years or whatever to keep the regular people safe.
The police don't follow up because that's policy, not because they would be overwhelmed. If crime were detected everywhere, it's not the system that would collapse, it's crime. Crime has already skyrocketed compared to post-war levels, not because the police can't arrest all the criminals, but because it is chosen policy not to do so.
Do not mistake a decision not to use state capacity for a lack of state capacity. The US (and most states and cities on their own) has absolutely terrifyingly gargantuan levels of state capacity for mass surveillance and law enforcement, and when it really wants to and finds the right Hawaiian judge to bless off on it, the US regime finds it absolutely trivial to crush anything at any scale.
I think his point is more that universal surveillance would only be used to gather information on people the state might like to punish for political reasons instead of for breaking the law only.
Agreed they chose not to pursue enforcement of crime by known offenders. Disagree they have the capacity to do much more than they are doing but they could potentially change the level of crime by increasing enforcement and then have the capacity to broaden enforcement.
It makes no sense to talk about "capacity" unless one disaggregates the actual limit of physical, empirical capacity (which was FAR above what is required to crush all crime a long time ago) from socially-constructed reductions in capacity, which are made nearly as big, because our regime cares less about stopping crime than it does about not stopping the wrong kind of criminals.
Back before glorious equality turned "standards, no compromise" into "the standard is compromise", the army once required soldiers to be able to carry an injured 90th percentile weight comrade 100 meters to evacuate him to safety. Most fit young men when trained and physically conditioned have the - capacity- to do so, so long as one doesn't normalize renormalizatiom and the 90th percentile isn't allowed to balloon, as it were, to double what it used to be.
But what is the "capacity" to do so when you are also required to carry your ruck sack, weapon, and the heavy mortar base plate, as was (once) a fun challenge in some of the more elite, selective units? Much reduced! Maybe only 2-3% of Joes can do it now. Maybe nobody can and you have to settle for not being able to evacuate anybody over the 50th percentile, so sorry Tall Tommy.
And what about if Joe shoots himself in the foot? Not much "capacity" now, I'm afraid, not even to drop everything and hobble to save just his own life!
One only makes the mistake of thinking the mirage of "low capacity" is a real limitation when one accepts repeated treatment by emergency surgical podiatry for self- inflicted wounds as somehow unavoidable and "normal" instead of a continuous series of conscious choices, as is it was something that just happened to us like a bad weather event beyond anyone's control. It's definitely in control, and being throttled to ensure the persistance of high crime rates.
Agreed that if there were less socially-constructed reductions in capacity, enforcement could be broadened. I guess it depends on the definition of "crush all crime" but I lean more towards potentially making substantial reductions, maybe even more than half.
Re: "widespread surveillance to detect infractions would have chaotic effects unless laws were carefully rewritten to somehow spell out the behavior that we really want."
A deep point!
It invites fresh thinking/research about fundamental topics:
• Rules vs discretion
• Causal relations between laws and social norms
• (Economic) theories of incomplete contracts
• Judicial review
Etc
The static analysis is that there would be way too much and too severe punishment of minor infractions. The dynamic analysis is that behaviors would change (risk compensation goes both ways) and to the extent it doesn't, intolerable situations create the pressures necessary to return matters to a tolerable state and that laws and penalties (de jure or de facto) would be adjusted to accommodate the transition to a high vs low detection rate law enforcement environment.
Which ... is already what happened, over half a century ago. There is no need fit """AI""" a term getting misused all the time and which at any rate is just the latest wrinkle. We've been living in world in which affordable technology well-deployed and utilized has made it feasible to detect all kinds of crimes at near 100% rate, if the powers that be want to. Crime is a solved problem; we are choosing not to solve it. Go ask China about their powerful, pre-"""AI""" detection system and low crime rates, especially for things the regime actually really cares about. Or go see the extent to which the US regime will go to use existing technology successfully to identify, find, and round up literally every single person involved in an event it cares about. But what the US regime also cares about is NOT having too much punishment of the wrong kind of criminal, and so for decades has rewritten, adjusted, and effectively relaxed the law and simultaneously crippled the capabilities and ratcheted up the burden on law enforcement in those cases to absurdly obscene levels, or perform mass abuse of discretion to simply refuse to investigate and prosecute crimes, to conserve "group- based enforcement homeostasis" and keep up with and neutralize every advancement in or enhancement of detection capability lest too many of the wrong kind of criminals get locked away for too long. They even decide to just let people accumulate tens of thousands of dollars in traffic violation tickets and still renew their licenses and vehicle registrations if the alternative would mean arresting, or, heck, even inconveniencing the wrong kind of people too much.
I must disagree with Arnold's example of speed limits. I have the actual experience of having a morning commute through a suburban part of the city in one of those situations in which some local road with a 30 mph limit starts getting used as a major artery despite being lined with houses with kids waiting for school busses. The road became extremely dangerous, cars often exceeding 45 mph, with too many close calls and a few ugly tragedies. The residents got angry and the parents of young children especially so, and demanded action. It began with patrol cars, but that's using a lot of policing resources, and the nature of the road left little space to park such a vehicle for speed trapping, so it didn't end up accomplishing much. Then the city installed a speed camera, not the "only if 5 or 10 mph above the limit" but for real painful tickets for even 1 mph over. Boom, thousands of tickets per week. And then hundreds, dozens, just a few. Yeah, drivers groused in outrage at first (boy, did they ever!) but eventually they just got used to it, adjusted behaviors (slowed down, changed commute routes, took the bus instead, etc.) and the road and neighborhood is 100 times safer.
The point is, in the present legal, political, and cultural context, we should be much more afraid of selective enforcement decided by politicians and judges than universal enforcement decided by machine. In a "machine regime", it universal enforcement is intolerable, the pressure of new detection tech will be to change the letter of the law, to winnow down the "three felonies a day" problem. In the political regime, the pressure is to just ignore the letter of the law for your friends and let crime explode if necessary, and also to go full Beria, use national security-level capabilities to surveil you enemies, find one of a million possible "crimes" they committed, and throw the book at them.
You know how the AIs are gonna take over? We are going to let them, elect them. We are going to compare them to trash human politicians and conclude it's just no contest. They've got my vote already!
Rule of law for all, actually having no “Do you know who I am?” Immunity, is better than what we have now.
Larry Ellison seems to have been one of the “all” boats that was lifted by the rising tide.
I think of the creativity or higher-level hypothesis as basically that computers, however powerful, will/must struggle with out-of-sample ideas.
From McNeely's OpenAI how to chat:
"When a junior team member joins your company, you teach them to follow your processes. When a senior leader takes the reins, you expect them to have their own. ChatGPT wants to be the senior leader, given free rein on your tasks, using its own methods to find the right response."
Who is the driver/ master/ direction-setter/ senior? You or the AI--THAT is the question, or soon will be. I want a highly skilled assistant-servant to do what I want, and be able to quickly understand what I want, plus the ability to do it. Doing things in the right way but doing the things I think are the right things.
Enforcement of many laws will be primarily, if not exclusively, against regime opponents. Like HR Clinton illegal email sever, DoJ ok, vs anti-Trump lawfare.
There is a human social need to have softer & harder boundaries, including taboos that can be violated with much less negative results. If all sex with consent is ok, what about consent to be in a situation where one can’t consent, nor object? This is not that, just pure drugging a wife and video taping more than 83 different men having sex with her. Most have defense they didn’t consider it rape.
https://www.maryharrington.co.uk/p/no-taboo-is-safe
The ease, and greater frequency, of using freedom to act which is harmful to others is why norms & laws limit freedom. Even many libertarians accept laws against drunk driving due to much higher probability of hurting others. AI super surveillance is likely to increase with cars, first. “But officer, I don’t think I was more than 10 mph over the limit”.
One area where there should be More enforcement is laws against govt misconduct, not only police but especially bureaucrats who have a whole slew of requirements they often don’t follow. AI support on knowing and quickly referencing requirements can help supervisors ensure compliance.
Meh. Ellison has just been watching too many Person of Interest reruns
" ... never be able to do X. Then a computer does X. Then we rationalize that the doing X is no big deal."
I'm not sure "no big deal" accurately captures what is happening. I believe the expectation is something closer to a human type of intelligence and when it is explained as Big Blue looking at all the possible chess moves or AI looking through information to find a likely fit, we decide those processes are more mechanical than intelligent.
It's not so much a matter of the red line moving as it is we are judging how the red line is reached.
Google "Mass surveillance in China" :-|
" ... never be able to do X. Then a computer does X. Then we rationalize that the doing X is no big deal."
Nietzsche's Last Man will make The Last Rationalization, "Even though I recently said it would be, the replacement of the last man before me was not actually impressive or qualifying as 'real' intelligence, however, replacing me requires real intelligence totally beyond the potential capabilities of extensions of current approaches, and thus won't happen."
There isn't a strict line because the physical and spiritual worlds map onto each other—advanced physical systems can become infused with spiritual realities. Computers, being blank slates, are vulnerable to evil influence, as we've seen with the internet. AI is potentially even more dangerous, since the more brain-like a system becomes, the easier it is for dark forces to "ensoul" it. That’s why AI development might pose a bigger risk than the already problematic internet.
I actually wrote an article about AI that has no religious or spiritual content, but it explores these ideas through a secular lens. Rebbetzin Fastag agreed to let me post it after we discussed the potential risks, particularly around black-box systems, which could unknowingly invite dark forces. Here's the link if you're interested: https://ishayirashashem.substack.com/p/artificial-intelligence-for-dummies