AI and A-Bombs: Is Everything A-Okay? | GoodFellows
HTML-код
- Опубликовано: 12 июл 2024
- On the 78th anniversary of the only wartime use of nuclear weapons, is the human race at another moral crossroads, fearing what artificial intelligence (AI) breakthroughs might unleash? Fei-Fei Li, co-director of Stanford University’s Institute for Human-Centered AI, joins Hoover senior fellows Niall Ferguson, John Cochrane, and H.R. McMaster to discuss AI’s promise and peril, followed by the three “GoodFellows” revisiting Harry Truman’s decision to drop the bombs in 1945. Just as crucial to mankind’s future: they debate the likely winner in an as-yet-unscheduled MMA bout pitting Facebook’s Mark Zuckerberg against X’s Elon Musk.
ABOUT THE SERIES
GoodFellows, a weekly Hoover Institution broadcast, features senior fellows John Cochrane, Niall Ferguson, and H.R. McMaster discussing the social, economic, and geostrategic ramifications of this changed world. They can’t banter over lunch these days, but they continue their spirited conversation online about what comes next, as we look forward to an end to the crisis.
For more on this series visit, www.hoover.org/goodfellows.
00:00 - 1:38 : Introduction
1:39 - 11:05 : Impact of AI + Understanding AI
11:06 - 17:52 : Should we be worried about AI towards the election + actions/dangers
17:53 - 22:06 : Ideas about AI + Education POV
22:07 - 36:05 : How America will cope with AI + remedy for AI + innovation (biological)
36:06 - 40:23 : Did Harry Truman make the right call?
40:24 - 47:39 : What would’ve happened if the bombs did not drop? + Building of Weapons
47:40 - 52:59 :Will we use nuclear weapons in the next 78 years? + Future + Defense
53:00 - 1:06:09 : Lightning Round
This was a fantastic episode for showing the biggest danger of AI: The lack of understanding of basic concepts by so-called experts. The first thing I noticed was after John asked FeFe a straightforward question about LLMs such as might come up in a Data Science interview, she could not answer it. The second thing I noticed was a pattern of anthropomorphizing all AI/ML as AGI in terms of both capabilities and threats. Thirdly few, if any, of her assertions about the technology are statistically testable, therefore not science but belief. To her credit she was entirely right about ChatGPT being an amzing teacher.
These kinds of biases are very dangerous. I work as an ML Engineer, responsible for making AI/ML products just work. I see this overpromising and underdelivering all the time either in the form of promises of a thinking machine with super powers or promises of vast sums of money due to using certain algorithms. Nobody asks "How do you know this thing works?". Then they are surprised by racial bias or aggressive text or any number of flaws due to preventable model and training issues. The stakeholders are promised that AI is an infallable thinking being rather than just another computer program. Garbage in, garbage out.
The most embarrassing and deliberate fraud in this space is that of large firms such as Microsoft and Tesla promising that their AI is a "safe" AGI that can be regulated. Not at all like those evil thinking machines that are self aware and will take our jobs. (Have you ever actually tried to automate labor? It is much more difficult than advertised. Those projects have a high failure rate). These large corporations are rather shamelessly attempting to protect their purchased AI assets from competition from open source alternatives. ChatGPT is not the only LLM, and there are new ones being produced all the time.
The fraud of claiming an AGI where there is none, could be catastrophic. For one thing, models are very easy to poison. An unscrupulous state actor could do a lot of harm in multiple ways. Social Media engineering and DeepFakes are just the beginning. Expect to see some wild stuff next year.
Lastly ML/AI may be the atomic bomb of these times. We do not have an AGI now, but the nation that builds the first one will by default be the leading world power. Assuming that something like the singularity happens quickly after the first AGI is created. Every other nation will be bringing knives to a gunfight. Pausing this escalation to make a few billionaires richer is probably the most irresponsible thing our nation could ever do.
So very well put.
Thanks for your thoughts, very practical.
@troymann5115 are you AI or just playing for AI team?
I have the impression that Fei-Fei Li and many others who excel in technical disciplines don't have a very profound notion of what "intelligence" is. As far as I know, no computer can write with Shakespeare's genius or propose the unification of electricity and magnetism as did James Clerk Maxwell...That is intelligence, and it is rare, but uniquely human.
But, give AI time. We absolutely, positively cannot imagine its future capabilities…from A to Z and beyond.
That would be like going back in time and trying to explain to the people working with the printing press how writing would be done on a glass screen - with a letter from U.S. to someone in Europe receiving it in 2 seconds. That still blows our minds.
A baseless assertion, why should intelligence be uniquely human, and why is the ability to write shakespeare how we define intelligence? If chat gpt can produce an entirely fictional shakespearean play would that make it intelligent, or just very good at imitating?
A more fundamental definition would be an ability to deduce information from other information, if I tell you A=B and B=C, can you work out A=C without me telling you? Programs are currently very bad at this, they do only what they have been told to do and no more.
Good to see John as passionate as can be this episode.
She thinks like my academic coworkers (they have rejected me as a colleague due to my conservative views) here in California. Government, government, government is the answer to every question. They simply assume their fellow professors are in lockstep. That thinking, not AI per se, will be what makes the technology dangerous.
I love it when HR comes across as the reasonable one of the regulars lol 👍🏼
Yes, we need a philosopher on the panel. She understands this 100x better than economist John and others. Economist doesn't yet get it and neither does the historian and military expert. They assess AI has something humans control, the threat is that at certain stage AI controls. As humans we cannot be trusted to safeguard AI. Lots of ethical issues.
Our current set of self-appointed AI "guardians" are incredibly leftward-biased.
Why do you need a philosopher on the panel when you have a top economist, historian, and geopolitics experts? When she said "you need a philosopher," you don't think that was a way to evade the question without actually answering?
Love listening to these. Brilliant job as always
A super discuss today. As usual best thing I've listened to all week. Thanks !
Bravo! I have a rather biased opinion of Mr. Truman's orders: my father was "on the boat" awaiting the invasion when the festivities came to an end. I prefer the ending we got.
Niall: You are making a distinction between crooks and campaign operatives?
Fantastic, immensely interesting and informative as usual, thanks Gentlemen 👏🇦🇺
Great episode! Thanks to all.
From my experience with Army mechanics, I would not want any of them to be running our federal bureaucracy. Maybe the generals at the Pentagon are assigned a better class of mechanics. Every one of my visits to an Army motor pool (Maintenance Facility) was characterized by complaints about the lack of spare parts, of assigned operators, and poorly trained operators. Those guys were never happy.
Lol, true observation. But as for the political considerations, nobody is happy.
This panel, plus or minus one, is the best thing that ever happens in intellectual discussion.
We will definitely get dumber the more AI replaces previously required thought. It's true of basic calculators, so it will certainly be true of AI. But we'll be more productive?
I can see what you’re saying to a degree, but I wonder. I believe Socrates (maybe a different ancient philosopher) was against using the technology of writing as it would free people from having to use their memories.
Less than 1% of people on this planet move science and engineering forward the rest of us are already as dumb as birds.
@@natesenglishkameoka3728 I suppose the modern equivalent is trying to remember something instead of googling it. All the world's info floating around our head, so there's no need to keep it IN our head.
Can we actually get any dumber than we are now ( especially 80% of Americans) ?
Yes, we peaked during the Stone age 🙃
A great discussion! It would have been good to dedicate a part of the time of the episode to delving into the life and times of Oppenheimer, and parallels today with leading luminaries of AI.
Great talk with Fei Fei, we are in good hands. If Unit 731 could do what they did without AI, it is good to hear there are some protections.
18:02 "... might transport the individual back to a pre-Enlightenment state of bewilderment..." I suspect that neither Henry nor Niall get out much amongst the *average* citizens of the globe.
Or not
Yep, the low 3/4th of the IQ curve is already in a constant state of bewilderment - most of the rest are so fixed in their opinions they have no hope of growing beyond their views (to even enjoy the initial bewilderment of new ideas).
Exactly what I was thinking!
AI makes mistakes, constantly.
....that's a major point of concern.
21:30 Completely agree! ChatGPT has been a tredemous gift for education. Regardless of its drawbacks, it is like having an always-available tutor to you, helping you save time with small, intricate, often time consuming details of a subject and explaining them so well.
The ramifications of which can be devastating since all of human history has been of learning through struggle. We have no idea if the human mind will even retain information long term if that time and struggle is removed. We could end up creating people 100% reliant on an I.A. thinking for them and knowing little.
What I am saying is. The small, intricate and time consuming details when researched are retained at a higher rate (from what we know) than those that have their hand held and everything explained to them because the mind compartmentalizes information. If gained all at once, you would need to have a near perfect memory to be able to recall it...but if its learned over time, its easier because of how the mind works for most of humanity.
I am excluding the savants with a photographic memory and recall well above the norm.
Finally we're relieved of the terrible burden of thinking. What's left? Just buy, buy, buy...until your money's gone, and your health is gone, and you tell yourself: "well, I sure logged a lot of screen time!"
A good one 👍. Keep it on.
As always a true delight, thank you gentlemen (and lady)....."one and all".
Peter
Vancouver Island
One should continue the topic of AI with Fei Fei. Again too notch discussions!
Wow. Knowing that males are more aggressive than females, it was so great to see this woman hold her own against men. No, I like and appreciate all the "GoodFellows" and get a lot from these episodes. Huzzah - Fei-Fei Li!
I'd love to see an interview with PERUN - who specializes in defense economics and who has been providing a 1 hour weekly analysis of the Russian invasion of Ukraine. Now this would be insightful and informative.
Great and broad conversation. Loved it..
Good all ways, glad of *YOUR* videos.
Fei-Fei Li makes a good point with regard to Human dignity and agency, we have already seen in the covid crisis how human power structures imposed their ideas on the population, I can only expect more totaltarianism from a robot. Of course some will argue that AI would not ignore data as humans do but I am aware that AI is programmed and by humans. Very interesting discussion, thank you all
Fei-Fei was a great guest. Thanks 👍
Damn. 😂. That was a rough one. But awesome as always
Fei Fei brings a point about the quest for knowkedge, like chemical weapons research, or dare i say gain of function testing. That went real well.
Hey Neil...I have a case of Academic Elitist fatigue. Take a walk back over the pond with a lead brick necklace. Cheers!
In order to create an AI scientists should understand what real human intelligence is. I don’t think science understands today what human intellect is and how it works from the physiological biological cellular point of view.
Apparently Oppenheimer actually said: it worked. He thought of I am become death sometime after
And you believe it??
I never thought I would see the day that John threw in his lot with Marge Green. Dr. Lee was very good guest, and I would love to hear from her again.
John Cochrane's first question is answered very clearly, by the giant, George Gilder, in his book, Life After Google. AI does NOT have that which human are graced with: creative genius. AI will always rely on inputs and sustainability from sources beyond itself. AI in and of itself is no more dangerous than a knife in a draw, even if now its become AI capable: sharpening itself, altering its blade, serrations, sharpness and such; it still requires human application as a tool.
Fantastic discussion
48:38 This is what the ending of Oppenheimer suggested to me as our likely future...a chilling admission from H.R.
Thank you for recommending Christopher Lasch's _The Revolt of the Elites_ -- it should go alongside Sowell's _The Vision of the Anointed_ as required reading for political scientists everywhere.
Every year at this time we wring our hands over the dropping of the Bomb... if we hadn't I for one probably wouldn't be writing this quip and some of you wouldn't be reading it. My Dad had just finished up three years in the ETO and his unit was getting ready to move to the Pacific for the invasion of Japan. When he heard the Bomb was dropped he recalled to us, "I just might survive this war!" Let's face it countless lives were saved by dropping the Bomb American and Japanese... the question will a nuclear weapon be used in the next 78 years? I think The GoodFellows answered that in your debate. The horror of using nuclear weapons keep normal people from using them... the key words...Normal people. The AI debate is so very hard to put thoughts to because it's not tangible and it will be hidden to most people kind of like Electricity we use it, but never really see it, only its results. This is a topic I'm sure you will revisit time and again... as always good topics.
I have talked to vets at different times and places, most of the combat vets said it was the happiest day of their lives, they then knew they would have a future.
Great nuanced debate on AI, thank you. More on Geopolitics please especially the Ukraine problem and CCP nuclear escalation and it’s impact on the Indo-Pacific. I’m from New Zealand please talk about us too.
Such a great show - Thanks Again. ❤😊
Only thing that can be said about New Zealand is how they denied US nuclear powered aircraft carriers and subs to dock in their ports.
Australia is a much more reliable ally.
It's so interesting to listen to these US / Northern hemisphere perspectives over the equatorial fence from the South. Where we phased out and banned nuclear weapons, and are protected by Coriolis force if a large exchange did break out in the North. And very different philosophy towards China. There is so much less fear for the future here. Except environmental of course.
There was a book about this ! "On The Beach" by Neville Shute.
Always feels good to have your feet in the sand… er, I mean your head
You should wake up to the CCP inhumanity!
Stimulating discussion, thank you all.
Thanks, dudes!
Favorite quote from the episode is at 6:26. I'll take validation for my career path wherever I can find it. 🙂
Great content, can you please make a podcast already so I don’t have to keep the RUclips video playing in my pocket?
you can find our podcast version here: goodfellows.podbean.com/ :)
The closer point is backing off of atomic weapons which we didnt do. She's right about the human drive to expand and expand and expand.
Alan Alda does a series for scientists about how to talk human.
Could not disagree with Dr. Chochrane more - if AI is not regulated to some way - then it will be used as a tool for exploitation.
I already design and use AI in advertising professionally - it’s almost unfair when combined with social media.
Also - would have been nice if someone argued against the dropping of the bomb - there is an argument for that.
Before we go too far down the "Wonderful Future" path that availability of knowledge will improve popular understanding, how has the internet literally at the fingertips in the form of smart phones improved literacy and reading rates?
You no longer need to own the classics of literature or go to a library to read/consume them. So therefore, every schoolchild should be thoroughly well read from James Fenemore Cooper to Mark Twain or John Milton to C.S Lewus.
Are they?
Not in my experience. That very availability has, if anything, DECREASED the rate of uptake in the general public
Our problem is that almost all humans think Descartes was right. They're 400 years behind. Or put differently, most people are Yorkshire grandparents: they know what they like and they like what they know.
Tools. Machines. Not life. Therefore not intelligence. What AI, starting from decades back, has done is to make some sense of huge volumes of data. This can be used to increase productivity for some and open up exponentially more corridors for avarice. That is why people who have been dumbed down fear the spread of AI. It is not AI per se but the human capacity to increase the rate of human entropy and degradation. This in spite of the potential of AI to help better the human condition.
Who decides who the best human is? I already suspicious of that sentence itself
Steal from one: Plagiarism!
Steal from many: Research!
Now we have "Steal from everyone: AI!"
21:11 omg this on education - so good!
I think that it's worth thinking about how while we are advancing by leaps and bounds technologically, we seem to be backsliding rapidly socially in terms of personal relationships. If anything causes us to go backwards it won't be the science, it will be the lack of meaning and purpose in society.
50:16 2034? Going to have to check that out.
good day a.i. how do i lose 10 useless pounds? a.i: "cut off your head peasant."
its a valid answer... does it have any feeling? morals? ethics? empathy?. a.i. is doomed due to the absence of these making it a brilliant psychopath.
I have used both BING CHAT and BARD. Neither seems to be able to do simple math, even after multiple prompts. Is that being addressed? LLM may be good with suggesting the next word, but what good is it if it makes up stuff (where there are gaps in the knowledge base) to throw at the user, without informing the user that the response is dubious?
Letsget AI in human resource departments and see what happens. How about the DMV or other state monolithic agencies.
Can AI be held accountable for anything?
Great question, and big limitation re: realistic implementation in modern context. It's machinations are arcane to most, but only its users can be held accountable, barring software devs issuing limited context licenses.
An interesting group of people. Are these shows done on a monthly or a quarterly basis?
Monthly during the summer, 2x/month the rest of the year. Next show will be released on 8/23.
What was the New Yorker article mentioned about the bomb?
Run for president HR!
Nice to see some thinkin' goin on in the summer months...
Hello Hoover people 👋
Kudos on a wonderful interview! Let's not lose sight of the fact, however, that experts in A.I. who are also native English speakers is as rare as hen's teeth on one of Neptune's moons. Falling back to second place had CONSEQUENCES. American IT. in every sector is 90% #H1BTech. We lost the tech battle. That ship has sailed.
Chat GPT has learned to be dishonest - likely through the political biases of it's trainers... I am somewhat encouraged that prof Li is more concerned about her students developing critical thinking skills rather than regurgitating the politically warped agenda of our indoctrination system (as chat gpt tends to do).
Somebody really needs to find a copy of The Forbin Project movie. Even though it is more than 50 years old, it is an effective movie about an AI which gains control of nuclear weapons. The last 5 minutes are truly chilling.
Very astute. That was a great movie , and a harbinger of things to come from that times perspective when it was made.
Hopefully some of us won't be alive to experience it
There is already at least one superpower, out of the 3 currently in existence (US, Russia, China), which has already switched control of its strategic component of the deterrence, to full AI automation. That is Russia, the largest nuclear power on the planet. There is also a proxy war being fought by two nuclear powers on the planet, right now, in Ukraine. It's a war over energy, market share, and the future of the global reserve currency, being fought on the ground of Ukraine by NATO and Russia. This was can escalate into the last human war, quite literally on hour by hour basis, and yet it's been over a year now, and it hasn't happened. You see, just because the danger of something catastrophic exists, doesn't mean it will be realized. Can a our world end tomorrow, realistically, because of some escalation in Ukraine? Yes. Will it? Probably not. Has Russia transferred control over its nuclear assets to AI? It has, since 2018, a few years already passed. Did anything catastrophic happen? No. I think there is a lot more resistance in the system, to just humanity, then what we give it credit for.
It was. I like the part where he said it's Forbin time and forbed everywhere.
Its here on YT. Loved it as a kid. ;)
Your guest is the true definition of artificial intelligence.
It's a large language model. But true "meaning" escapes it.
great conversations, guys -- yes, have Dr Li back, that was very interesting, I listened twice. On the A-bomb: HR suggested a future episode about nuke proliferation. How about imagining post-(this)-war Ukraine, with Russia both a bigger proliferation vector than it has ever been, this time supporting Iran's nuclear bomb (if it isn't doing so already), AND an ISIS/AQI type terrorist threat inside the Ukrainian territories it claims it annexed. Does Ukraine have to apply Iraq Surge style counter insurgency in Crimea and the Donbas?
58:29 South Park!!! love it!!!!
Never miss an opportunity to promote your books NF! The guest's command of English and AI made her ineffective explainer of the subject.
To me there seems to be one main question and the answer to it will tell us if AI will overtake humanity: If you feed all music ever composed from the first tune until, but not including Beethoven - can AI compose Beethoven?
A GAN takes training data and continuously retrains it with another set of data, over and over until an image or sound is produced. It is not the same thing as a true composition. However the sound can be intriguing.
How do you discuss the surrender of Japan and neglect to mention the entry of the Soviet Union in the war on 8aug45?
Missed point: AI increases the likelihood of extreme scenarios, both positive and negative. We, as humanity, can navigate the positive ones...but what about when an extremely bad scenario, made possible by AI, occurs?
47:15 The Chinese don't follow any international rules, do they? Why would it be different when it comes to nukes?
CIA Coups in Iran, Guatemala, Congo, Dominican Republic, S Vietnam, Brazil, Chile..,,.Illegal wars, occupations, regime change, and proxy wars in Afghanistan, Iraq, Libya, Syria, Yemen.
Thank God for JC. I hate the word 'we' too. It is the pivotal idiom of the we say because we think & you do because we say politico-beurocratic complex. People who start sentences with phrases like: "I think it is really important to recognise the fundamental (insert platitude of your choice)' have spent too much time sitting on committees and too much time with other members of the 'we say' community. They are the same people who use the seemingly passive: 'there is a really important conversation to be had about how we manage (insert socio-cultural 'tool of containment' of your choice) when what they mean is pay me some more to sit on a committee and deliver more hollow platitudes like this. As the great man so eloquently said: "You have a very benevolent view of how regulation works in a political system'.
The greatest irony of all time would be a scenario in which nuclear war is the only one that allows for a total destruction of all infrastructure and therefore AI along with it yet leaves remnants of humanity sufficient to rekindle and repopulate.
She's been in the USA since 1988 & still has such a thick accent?
I had the opposite take on her speech, but I'm just a dumb cow puncher.
Just so you don't get too unnerved by GPT4's omniscience, the oinly difference in the replies to my question "What does Niall Ferguson like to eat?" was that GPT4 managed to come up with a snikde dismissal: "Niall Ferguson is a historian, author, and professor known for his work on economic and financial history, rather than his culinary tastes." Anyway, I'm glad he doesn't like MacDonald's.
Apart from that, it's such a relief that China can be discussed as the enemy it obviously is. Tnat would have beeen unthinkable not too long ago. . Finally, Cochrane's remark about the left being eager to extend voting rights until the "wrong people" start voting reminded me of Latin American revolutions in which the Criollos were liberal compared to the Spaniards but always drew the line at land reform, which meant to hell with the vast majority of the population. Okay, I'll shut up, great podcast!
Zuck is not a tech Giant he is an ad-executive
I am one of the proud benighted MAGA caste that former conservative elitest, David Brooks, condescended to address in his op-ed piece. I may only have two double-major masters degrees from two non-Ivy League institutions, but I'm smart enough to see where the self-annointed eggheads in DC, the leftist media, and oligharchic corporate masters are taking this country.
I never liked the boorish rhetoric of President Trump, but I strongly supported 90% of the policy initiatives he tried to pursue, especially his appointments of legal textualists to the Federal Judiciary. If the new GOP base he created hands him the nomination over DeSantis or Ramaswamy, so be it.
Biden can't/won't run, so the conservative populists may have a shot at arresting the current slide into wokeist oblivion. God help us if they don't!
You have voiced my position on Trump exactly. Well done.
John is wrong, the US designs the top tier chips, then gives them to TSMC to make, TSMC exists because of the US. Do they think Taiwan built this industry?
2:27 Wow I didn't know Eliezer Yudkowsky has a PhD now.
Anyone alive 8 years ago when Self driving cars were about to go mainstream?
What happened to that intelligence? 😂😂😂
This is a lot of hot air.
Next week lets talk aimlessly about Quantum Computers. That's so hot right now 🙃
The reason why Japan fought on was the demand for unconditional surrender this is why nuclear bombs was used against Japan
Many people in China are already subjecting each and every citizen and visitor to China to absolute Individual grading system, with which the system will treat them accordingly. What are you doing about that? Did you discuss this with anyone in China? 🙃
Why are we not training A.I to shine light into the black box?
Probably a bolt on independent, a de-compiler.
“….I think its really important to recognise the fundamental, quest for innovation & expansion of knowledge is in humanities DNA…” Fei Fei Li. 👍
The ‘Tree of Neuronal Instruction Set Evolution’ I.M Gurney.
"Are they just regurgitating facts, or are they critically thinking?"
Quoting this conventional wisdom, she misses the post-modernist toxin: Are they thinking so critically that they've stopped believing in facts?
AIs lack embodied experience of reality. One massive potential problem with AI, is that it will diverge from reality if exposed to too much theory.
I don’t want AI telling us or evolving our perspectives
Does anyone else think that AI and technology experts are their own hype train? In other words, they underestimate the capability humans have to prevent worst case scenarios of technological advancement? Througout history, for instance, humans have consistently harnessed innovation and prevented doomsday scenarios.
19:35 if people are not bewildered about how their google search comes up with choices, then we will probably be ok
The hullabaloo is they don’t know how it does what it does.
⭐
Have any of the GoodFellows' episodes inspired a GoodFellow to pursue a research or book project?
I understand everyone is busy, but why cut off a guest so early? Every episode we act as if there is a time constraint where there is none.
Justin Trudeau did a boxing match with another politician. Must be a trend.