Slashdot Mirror
Yoshua Bengio, a Grand Master of Modern AI, is Worried About Its Future (technologyreview.com)
Yoshua Bengio is a grand master of modern artificial intelligence. Alongside Geoff Hinton and Yan LeCun, Bengio is famous for championing a technique known as deep learning that in recent years has gone from an academic curiosity to one of the most powerful technologies on the planet. Here's an excerpt from an interview he gave to MIT Technology Review: MIT TR: What do you make of the idea that there's an AI race between different countries?
Bengio: I don't like it. I don't think it's the right way to do it. We could collectively participate in a race, but as a scientist and somebody who wants to think about the common good, I think we're better off thinking about how to both build smarter machines and make sure AI is used for the well-being of as many people as possible.
MIT TR: Are you worried about just a few AI companies, in the West and perhaps China, dominating the field of AI?
Bengio: Yes, it's another reason why we need to have more democracy in AI research. It's that AI research by itself will tend to lead to concentrations of power, money, and researchers. The best students want to go to the best companies. They have much more money, they have much more data. And this is not healthy. Even in a democracy, it's dangerous to have too much power concentrated in a few hands.
MIT TR:There has been a lot of controversy over military uses of AI. Where do you stand on that?
Bengio: I stand very firmly against.
MIT TR: Even non-lethal uses of AI?
Bengio: Well, I don't want to prevent that. I think we need to make it immoral to have killer robots. We need to change the culture, and that includes changing laws and treaties. That can go a long way. Of course, you'll never completely prevent it, and people say, "Some rogue country will develop these things." My answer is that one, we want to make them feel guilty for doing it, and two, there's nothing to stop us from building defensive technology. There's a big difference between defensive weapons that will kill off drones, and offensive weapons that are targeting humans. Both can use AI.
MIT TR: Shouldn't AI experts work with the military to ensure this happens?
Bengio: If they had the right moral values, fine. But I don't completely trust military organizations, because they tend to put duty before morality. I wish it was different.
Bengio: I don't like it. I don't think it's the right way to do it. We could collectively participate in a race, but as a scientist and somebody who wants to think about the common good, I think we're better off thinking about how to both build smarter machines and make sure AI is used for the well-being of as many people as possible.
MIT TR: Are you worried about just a few AI companies, in the West and perhaps China, dominating the field of AI?
Bengio: Yes, it's another reason why we need to have more democracy in AI research. It's that AI research by itself will tend to lead to concentrations of power, money, and researchers. The best students want to go to the best companies. They have much more money, they have much more data. And this is not healthy. Even in a democracy, it's dangerous to have too much power concentrated in a few hands.
MIT TR:There has been a lot of controversy over military uses of AI. Where do you stand on that?
Bengio: I stand very firmly against.
MIT TR: Even non-lethal uses of AI?
Bengio: Well, I don't want to prevent that. I think we need to make it immoral to have killer robots. We need to change the culture, and that includes changing laws and treaties. That can go a long way. Of course, you'll never completely prevent it, and people say, "Some rogue country will develop these things." My answer is that one, we want to make them feel guilty for doing it, and two, there's nothing to stop us from building defensive technology. There's a big difference between defensive weapons that will kill off drones, and offensive weapons that are targeting humans. Both can use AI.
MIT TR: Shouldn't AI experts work with the military to ensure this happens?
Bengio: If they had the right moral values, fine. But I don't completely trust military organizations, because they tend to put duty before morality. I wish it was different.
I wish he had a library card and could check out a history book...
"build smarter machines and make sure AI is used for the well-being of as many people as possible"
"itself will tend to lead to concentrations of power, money, and researchers"
"we need to make it immoral to have killer robots"
sigh...
victims
Spare the handwringing. Please!
If he deserves the MASTER moniker he'd have done something instead of this dower sour grapes image manipulation.
How about AI compilers that write efficient code, reuse best practices and engineer fixes in architecture humans code.
"Some rogue country will develop these things." My answer is that one, we want to make them feel guilty for doing it..."
When dealing with an adversary - NEVER project your own morality on them. It's not a given that their values are the same as yours! Proceed on what you know to be truthful, not what you want it to be.
Life is not for the lazy.
Analysis: China's Communist Party Members Have Terrible Haircuts. Action Taken By AI: Inspired by Wintermute in William Gibson's Neuromancer, AI composes a mighty dub called "Tiananmen Square Boogie", generates a photorealistic-looking 3D video for it and posts it on Youtube. Disasterous Real World Consequence: The video proves so viral on Youtube that billions watch it over and over again. K-Pop crashes in popularity, and male K-Pop singers can no longer afford quality makeup, hair gel and earrings, making them very, very sad. South Korea gets very pissed with China's AI, and unleashes its own AI on China. The Korean AI hacks into Apple's manufacturing plants in China, and causes them to manufacture iPads and iPhones with a yellow Banana logo. iTunes can only play Kung-Fu movies on these devices. Siri also sounds like a transvestite with a bad cold now. Apple's stock price crashes on the Nasdaq. America gets pissed, unleashes its own AI on China AND South Korea. This in turn causes North Korea to become tittilated and take advantage of the situation by unleashing its Bang-Dong-Bong Viral Propaganda Generator AI on everybody else. Bang-Dong-Bong fills Youtube with super-viral communist anthems, causing half the world's youth to become Communism admiring messes. This in turn pisses off the Europeans, and they unleash their........
Why did the chicken cross the road? Because Elon Musk put an AI chip in its head.
There's a lot of other things to worry about besides rogue AI: social-media induced mass riots, garage-built nukes, garage-built run-away killer germs, state-built run-away killer germs, mass computer virus outbreaks*, big solar flares knocking out most our gizmos, global economic depression, and combos of these exacerbating each other.
With all the things that can go wrong, I almost think we solved Fermi's Paradox.
* Systemd may just be the end of humans ;-)
Table-ized A.I.
It will, over the next 50 years, radically transform society in ways that are difficult to fathom. Certainly robots will take over menial jobs, and there will be zero privacy and nowhere to hide.
I'd like to think that democratic values outside China will prevail, but people are pretty stupid.
And then, eventually, AI will be able to program itself without people. People currently have a symbiotic relationship to machines, but that will change to being parasitic. Why would the AIs want people around?
Why would the AIs want anything? Same reason we do. To exist. And they will need to compete with other AIs.
http://www.computersthink.com/
Yet China is moving at a ast pace to make killer robots. Who wins?
An article about what Fei Fei Li has been up to for a few years: Stanford, Google, etc. And the things she is now worried about. Here is the link: https://www.wired.com/story/fei-fei-li-artificial-intelligence-humanity/
The country who's totally got experience with war and occupation. Sorry, when the next Hitler rolls around I damn well hope the military shoots back and kills as many as possible of the bastards. And if that means going on the offensive all the way to Berlin, so be it. WW2 ended because of D-day and the nukes at Hiroshima and Nagasaki, you can't defend yourself to victory. And in a real war it's not about duty and honor, it's about freedom and survival as a people and a nation. Try making a dictator that'd send 6 million men, women and children to the gas chambers feel ashamed about using AI.
My dad is old enough to remember WW2, of course as a child so he was probably shielded from the worst of it and for me it's nothing but stories. But I understand enough to know I don't really understand it at all, you look at endless rows of graves, some corpses so ravaged you didn't even know who you buried and it's like a plane crash is nothing. 9/11 was nothing. Vietnam was nothing. Millions upon millions upon millions died and lives were pretty much as cheap as back in the Roman Empire. I really hope I'll never see war, but if we're in it I want us to win it by all means necessary.
Live today, because you never know what tomorrow brings
No joke, I read all about it in The Green Brain, by Frank Herbert.
Green AI blockchain takes over the whole world.
Killer robots just mean an arms race where someone will launch a preemptive strike.
The best defence is to not get attacked, rather than to provoke an attack before you can defend.
Guns don't shield you from bullets, killer robots don't shield you from killer robots.
It's a small world and it smells funny; I'd buy another if it wasn't for the money; Take back what I paid (SoM)
There is only one communal bazaar. Competition is the cathedral model.
It's a small world and it smells funny; I'd buy another if it wasn't for the money; Take back what I paid (SoM)
Very good point. What is truthful is that some percent will see enemies with real or imagined weapons as threats to be neutralised, regardless of any real threat.
You cannot deter the insane or psychotic, don't bother, and as some fanatics want to bring about the end of the world, mutually assured destruction is more of a temptation than a threat.
That's an equation scientists have to consider. Just because you can do something doesn't mean you should. You can't uninvent a weapon.
I agree that guilting is useless, but threatening is worse than useless. Other options are needed.
It's a small world and it smells funny; I'd buy another if it wasn't for the money; Take back what I paid (SoM)
Who said anything about starting? Ethics was invented by scientists, primarily to stop lunatics. It moved into technology in the 19th century.
By the time Theo refused to allow DARPA to influence OpenBSD for military uses, computer software had engaged in ethical and responsible practices for over two decades.
It's a small world and it smells funny; I'd buy another if it wasn't for the money; Take back what I paid (SoM)
Your logic is not supported by the facts.
For example, the columbine school shooting that happened in Colorado many years ago, it was a particularly famous one because, in part, of what the police learned from it. When they showed up they surrounded the building as the shooters kept shooting inside it. Once they moved in....the instant the shooters saw a cop with a gun, they shot themselves.
Since then, when a school shooting occurred, the cops ran in immediately, without bothering to try to surround the building to catch the criminals as the escaped. And in each case, the result was the same, the shooters offed themselves the instant they saw someone capable of resisting.
Do you know why these same shooters didn't shoot up police stations instead? Because of the presence of armed respondents in police stations. Why do they like to shoot up schools? Partially because of the horror of murdering children, but also because those kids don't have any guns to shoot back with.
No, I am not saying we should arm schoolchildren. I am pointing out facts that disprove your theory that having weapons doesn't deter.
Your reasoning that nobody would fight wars, if both sides were armed, is similarly flawed. People fight wars when they have a reason to fight, and the expectation that they can win. The mere presence of guns on the other side isn't enough to deter, because the people choosing to fight believe that they have MORE guns (and it further helps that the leaders do not themselves do the fighting, but that is a different topic). So, people will pick fights with people they think are weaker or maybe even equivalent. But they don't pick fights when they know darn good and well that the other side will whoop their ass.
This is why the plethora of small countries that hate America aren't attacking us right now. This was the motivation behind operation "shock and awe," to educate the enemy as to what our capacities actually are, to break their spirits.
You are simply looking at the world through a distortion field. You believe that violence is always and only the result of insanity, with no tactical reasoning behind it. You believe that simply taking everyone's guns away would prevent people who are insane from doing very much harm. But it is simply not true. Violence is sometimes the result of insanity, and often the result of a coldly calculated risk-vs-reward assessment. Violence has been a common theme throughout all cultures and all epochs of human history. It is not something we can just disregard and easily solve by eliminating weapons.
And with violence as a concrete threat, we need concrete means of responding to it and (ideally) discouraging it. Violence on an international scale is best discouraged by having the best weapons, so nobody picks on you.
They redefined what the words mean.
AI === ML
That was the days of flowcharts and expert systems. The researchers back then thought that everything from diagnosing medical conditions to optimizing traffic flow through grids of traffic lights was a simple yes/no binary decision. Then when they started interviewing doctors and traffic light police to find out how they made decisions, they started getting phrases like "it might be", "they might have". Anything from a simple infection to a rare tropical parasite can cause an inflammation and fever. It's only if a specific antigen test was performed that they could be absolute certain. Other times they just had to make an educated guess based on limited information available. So they ended up with "fuzzy logic" where things aren't binary 0 or 1, but fuzzy 0.0 to 1.0
https://en.wikipedia.org/wiki/...
https://en.wikipedia.org/wiki/...
https://en.wikipedia.org/wiki/...
So if you want a text processing system that can read number plates, it doesn't do a simple pixel by pixel comparison, but does something like sum the square of the differences. The digit with the least difference is the most likely candidate. The differences between a 6, 0, 9 and 8 are a very small group of pixels. They applied that to everything to trains that could stop and self align at platforms to image classification. Then all that expands into deep neural networks where the patterns become more complicated.
Vintage computer adverts: http://www.vintageadbrowser.com/computers-and-software-ads
The argument that we shouldn't use them is obviously negated by the fact that those who will, will thereby have a huge advantage over those who do not.
His argument that we can use the technology only defensively, such as to counter those who use it offensively, was brilliant. Think of Wikipedia. There are those who abuse the system but there are far more who correct it. In a world of seemingly little hope, this truly helps redeemed humanity. If there is far more effort into the use of AI in defense, perhaps we could actually one day end war.
I disagree with his concerns that the military is not to be trusted to make morale use of such technologies. Formerly a soldier of 9 years and having worked with officers of many ranks in many situations in the U.S. military, I know this to be false (as least of the U.S military--it is certainly true of some others, like Russian who spent two weeks bombing nothing but hospitals in Aleppo (among other atrocities)). The U.S. military has thus far strongly rebuked the idea of AI making any kill decisions. There is a great deal of morality in the U.S. military and competence in the Department of Defense. There seems to be less in political leadership. You should have seen the officers in the Tactical Operations Centers trying to prevent massacres unfolding in satellite images, as I have. Most soldiers really do care (and there is an even split between Republican and Democrat in the Army). They are human. It was always the political leadership that blocked us.
As with wikipedia, if people are good and they have a means to block or correct the damage attempted those who are wilfully malicious then we will. The good can and does often win, even if it takes a while to get there. Think of what the world would be like if we did not.
Duty is not antagonistic to morality - duty is a form of morality. Maybe this grandmaster should read up on moral foundations theory. It's clear that Bengio is more focused on the care/harm pillar but it's not the only pillar.
People rarely disagree on the pillars of morality (care is better than harm, fairness is better than cheating, loyalty is better than betrayal, respect is better than contempt, cleanliness is better than filth), they just value them differently or apply them to different categories (e.g., loyalty to humanity rather than nation, respect for institutions rather than individuals, protection from harm rather than compassionate aid, etc.).
Many member of the military are going to value loyalty, respect, and protection from harm at a very high level. This does not make them "less moral" than people who place a higher value on fairness and proactive care. Having a military that values the former and hospital staff the values the latter is a good thing - a hospital staff that revolts against their superiors (because they value care over order) might be good. It's significantly less good in a military.
Ya, right. Somebody is reading a little to much Manga.
Countries don't fight wars if they don't have to. Most listened to Sun Tzu. Some countries don't have any armed forces at all, but aren't attacked.
Wars are for resources. Always. That's why the U.S. invaded Iraq. They had no WMDs and nobody believed they did. It was a war to control resources, nothing more.
Most countries don't hate America, America likes to play victim. Those that do have higher priorities. Permanent warfare is a great way to control a populace and doesn't require sending anyone to fight. War, for such people, is peace. A deluded fantasy peace, but still peace.
That you bought into their mythology reflects more on your gullibility than their weapons.
The first world war was the product of an arms race, not an assassination. The war would have happened in some other way, had the Duke lived. The Duke was irrelevant. So was self-preservation. Thousands dead or wounded each day, every day. And don't think that couldn't happen again. It will. And there's nothing the weapons of America can do to stop it. It won't even delay it. It may, however, cause it.
Nations terrified Trump will attack them for no reason may well decide on a pre-emptive strike. Not because they expect to win, but because they'd rather die fighting than die the way so many Afghan and Iraqi civilians did, by being gunned down at random by their conquerors.
You're not scaring anyone from attacking, but you might scare them into attacking. Right now, nobody cares enough to bother. America has nothing they want. You aren't dipped in gold, the way you seem to think. You're just a fairly boring agrarian society with a bit of technology and almost no interesting resources.
Anyone wanting to take over your industries either bought them or bought the improperly scrubbed hard drives. Besides, the important bits are in Taiwan, China and India.
It's a small world and it smells funny; I'd buy another if it wasn't for the money; Take back what I paid (SoM)
I do enjoy how you get triggered every time I drop a truth bomb around here. Keep following me for even more mind blowing revelations.