One thing I do not know about modern robots is how autonomous they really are. An AI with spatial awareness on a par with humans more or less needs AI capabilities. But AI is very resource intensive, usually on the level of server racks, which is not possible to carry on a robot. If this is true then all advanced robots are dependent on very stable wireless network connections. Not only are they not truly autonomous but they will also be very fragile, possible to disable with only some radio interference. Humans, on the other hand, are truly autonomous. Not to mention that there are eight billion of us. A single million humanoid robots will not last long.
Hey Anders! Glad you had something to say; this post was basically what I needed to give a sensible reply to that comment on your blog. But I have three (predictable) answers:
1. Even presuming no improvement in robots over time, robots are already able react to human attacks in real time, and recover after hockey sticks and thrown barrels knock them to the ground.
2. Even presuming the robots mostly fall apart, the AI still has countless humans willing to do things they are told. Most people will likely not be aware of what is going on as the AI takes over; how will human insurgents even coordinate with one other if the AI controls their email and cell phone communications?
3. Even presuming all of this is unlikely, it has to be nearly impossible for the issue not to matter. Risk = Probability x Severity.
I think the core of my opinion here is that humans are much stronger and more resilient than we imagine and AI/robots are much more fragile than we think. A case in point is AI's dependence on electricity. While humans are able to survive on a wide variety of fuels AI runs on computers which are wholly dependent on electricity. Cut the electricity supply and the AI is toast. Not only is it quite easy for humans to destroy the electricity supply, even a wholly natural power outage may kill off AI if there are no adroit humans there to restore power.
Come to think of it, the height or irony would be for AI to wipe out the entire human race and then a solar storm hit which wiped out the power supply and took both AIs and robots with it. It could happen. At least the second part.
Concerning that video of the soldier robot, I am pretty certain it is fake.
But what if I grant all of that? Does this force me to concede anything?
You're pretty certain the video is fake; I'm pretty certain it's fake in the way prowrestling is fake (See for instance https://www.thesportster.com/wrestling/worst-wrestling-injuries-shouldnt-happened/ .) Are you absolutely certain? Say you are; do you remain absolutely certain robots will *never* develop the resilience necessary to survive with the power cut for a few days?
Of course you do not have to concede anything. Everything is a question of risk-reward analysis. There is a risk with everything, including AI. You obviously rate the risk higher than I do. And maybe you rate the reward lower as well.
I am not absolutely certain that robots will not outperform humans. But I find it very unlikely. As I wrote in my article, evolution has worked on humans for millions of years. It is unlikely that robots will do something similar in a matter of decades or centuries.
The Youtube video is made by Corridor Digital, a company specialized in digital effects. The robot soldier video is apparently famous enough to be mentioned on its Wikipedia page: https://en.wikipedia.org/wiki/Corridor_Digital
OK; I wanted my own numbers to convey a perspective, but when you explicitly grant the probability is nonzero, I don't really have any objection.
I guess our only real difference of opinion is how effective are directed processes vs evolution? You won't be surprised that I think directed processes are far, far more effective, although I haven't said why:
Our intelligence wouldn't have evolved unless intelligence usually won. Even so, because the human brain arose through random processes, it isn't anywhere near optimal; corvid brains are close to our own in terms of intelligence, and yet are much, much smaller. We've been playing Tafl games for over 1000 years, but computers can already beat us at them. And though our robots cannot reproduce, their performance is close to the level of ants, and ants took ~3.5 billion years to evolve.
As a last point of clarification, I don't think we can put the brakes on AI research, because someone else will continue to develop it. Westerners seem most concerned about safety, so, may Westerners be the ones to let the genie out of the bottle! But deep down I wish there were no bottle. Though I do agree with you that there are objective rewards around the corner, subjectively speaking, my favorite communication technology is the pen, my favorite weapon is the spear, my favorite mode of transportation is the boat, and my favorite computer is built into the edge of a cliff and uses the flow of a waterfall rather than electricity to make calculations. Also: under the directed intelligence of a human thinker, this computer is able to solve Russel's Paradox, and it drives the bellows of a pipe organ which it can direct to play music automatically. And *I* composed the music, and if you don't like it, then you have to find our *own* waterfall.
One thing I do not know about modern robots is how autonomous they really are. An AI with spatial awareness on a par with humans more or less needs AI capabilities. But AI is very resource intensive, usually on the level of server racks, which is not possible to carry on a robot. If this is true then all advanced robots are dependent on very stable wireless network connections. Not only are they not truly autonomous but they will also be very fragile, possible to disable with only some radio interference. Humans, on the other hand, are truly autonomous. Not to mention that there are eight billion of us. A single million humanoid robots will not last long.
Hey Anders! Glad you had something to say; this post was basically what I needed to give a sensible reply to that comment on your blog. But I have three (predictable) answers:
1. Even presuming no improvement in robots over time, robots are already able react to human attacks in real time, and recover after hockey sticks and thrown barrels knock them to the ground.
2. Even presuming the robots mostly fall apart, the AI still has countless humans willing to do things they are told. Most people will likely not be aware of what is going on as the AI takes over; how will human insurgents even coordinate with one other if the AI controls their email and cell phone communications?
3. Even presuming all of this is unlikely, it has to be nearly impossible for the issue not to matter. Risk = Probability x Severity.
I think the core of my opinion here is that humans are much stronger and more resilient than we imagine and AI/robots are much more fragile than we think. A case in point is AI's dependence on electricity. While humans are able to survive on a wide variety of fuels AI runs on computers which are wholly dependent on electricity. Cut the electricity supply and the AI is toast. Not only is it quite easy for humans to destroy the electricity supply, even a wholly natural power outage may kill off AI if there are no adroit humans there to restore power.
Come to think of it, the height or irony would be for AI to wipe out the entire human race and then a solar storm hit which wiped out the power supply and took both AIs and robots with it. It could happen. At least the second part.
Concerning that video of the soldier robot, I am pretty certain it is fake.
But what if I grant all of that? Does this force me to concede anything?
You're pretty certain the video is fake; I'm pretty certain it's fake in the way prowrestling is fake (See for instance https://www.thesportster.com/wrestling/worst-wrestling-injuries-shouldnt-happened/ .) Are you absolutely certain? Say you are; do you remain absolutely certain robots will *never* develop the resilience necessary to survive with the power cut for a few days?
Of course you do not have to concede anything. Everything is a question of risk-reward analysis. There is a risk with everything, including AI. You obviously rate the risk higher than I do. And maybe you rate the reward lower as well.
I am not absolutely certain that robots will not outperform humans. But I find it very unlikely. As I wrote in my article, evolution has worked on humans for millions of years. It is unlikely that robots will do something similar in a matter of decades or centuries.
The Youtube video is made by Corridor Digital, a company specialized in digital effects. The robot soldier video is apparently famous enough to be mentioned on its Wikipedia page: https://en.wikipedia.org/wiki/Corridor_Digital
OK; I wanted my own numbers to convey a perspective, but when you explicitly grant the probability is nonzero, I don't really have any objection.
I guess our only real difference of opinion is how effective are directed processes vs evolution? You won't be surprised that I think directed processes are far, far more effective, although I haven't said why:
Our intelligence wouldn't have evolved unless intelligence usually won. Even so, because the human brain arose through random processes, it isn't anywhere near optimal; corvid brains are close to our own in terms of intelligence, and yet are much, much smaller. We've been playing Tafl games for over 1000 years, but computers can already beat us at them. And though our robots cannot reproduce, their performance is close to the level of ants, and ants took ~3.5 billion years to evolve.
As a last point of clarification, I don't think we can put the brakes on AI research, because someone else will continue to develop it. Westerners seem most concerned about safety, so, may Westerners be the ones to let the genie out of the bottle! But deep down I wish there were no bottle. Though I do agree with you that there are objective rewards around the corner, subjectively speaking, my favorite communication technology is the pen, my favorite weapon is the spear, my favorite mode of transportation is the boat, and my favorite computer is built into the edge of a cliff and uses the flow of a waterfall rather than electricity to make calculations. Also: under the directed intelligence of a human thinker, this computer is able to solve Russel's Paradox, and it drives the bellows of a pipe organ which it can direct to play music automatically. And *I* composed the music, and if you don't like it, then you have to find our *own* waterfall.