0:00Everybody who's ever talked about AI has had a certain fascination and fear of where it might go. [music] I'm Tony Dopal and this is great question [music] today. My question is, is it too late to save humans from AI? Let's take a look at the history. 1907 railroad accidents kill almost 12,000 Americans. Not long
0:28after, 1913, a member of the Interstate Commerce Commission tells Congress that over the past quarter century, railroads had killed 188,000 people and injured 1.4 million more. And still, we chugged along because it was better than a wagon. Then the Model T came along and the car took over. Since then, cars have killed tens of millions of people all around the world. And today, car crashes continue to kill nearly 1.2 2 million
0:56people every year. We do make cars safer where we can with seat belts and airbags, speed limits, stop lights, drunk driving laws, but still we keep driving because we believe the benefits far outweigh the harms. Artificial intelligence is moving faster than locomotives or automobiles ever could, and AI is not generating that kind of death toll yet. But there are a lot of people raising the alarm about its potential for great harm.
1:26Bodily injury, property damage, damage to costly equipment or experimental materials, or the release of hazardous [music] substances.
1:34Sometimes AI is depicted as acting like the mafia.
1:38AI models could attempt to resist, shut down, conceal or manipulate information, or show behavior resembling blackmail. Sometimes it's depicted as behaving like Saddam Hussein or Kim Jong-un.
1:51Its [clears throat] best models have already demonstrated and will demonstrate capabilities that could provide meaningful assistance to malicious actors seeking to develop chemical, biological, or other dangerous weapons.
2:04And if you try to stop it, AI moving so fast that it could outpace the company's ability to develop corresponding safety [music] and security measures. And some of the models even predict catastrophic and exent risk of humanity. Uh particularly if such systems were misused, replicated, weaponized, or allowed to operate of a high degree of autonomy.
2:24Here's the amazing thing. None of those warnings you just heard come from AI doomers. They come from the most invested party there is. Those are quotes directly from Anthropic's IPO filing. Of the filings 335 pages, 93 of those pages are all about risk. A company that could soon be worth a trillion dollars is telling its own investors that its product could pose existential risks to humanity.
2:51Why are people closest to this technology so worried? What do they see?
2:56Is this really any different than the train or the car? And do the benefits of AI outweigh the harms?
3:03My guest today is the guy responsible for the last technological revolution and he is ringing the alarm. He goes so far as to say that AI is more dangerous than nuclear weapons. And just three years ago, Bill Gates was one of the most hopeful voices on AI. He said it was as fundamental as the creation of the personal computer and the internet.
3:24He talked about AI tutors for every kid and health advice for people in poor countries who rarely get to see a doctor. But then something shifted.
3:33Gates says the model got so good so fast that this past year they crossed what he calls danger thresholds. That means they could help someone build a biological weapon or knock out the systems that run our power, water, and money. Now, he's written a 12-page essay warning that we're entering one of the most turbulent periods in human history and that the window to get AI under control is the next four years.
4:00Gates is not saying shut it all down. He still thinks AI could be the best thing to happen to health and education in our lifetimes, but he is saying government is not paying enough attention. And at this rate, it may take a disaster before people wake up. We'll be right back. Bill, thank you for doing this.
4:23Great [music] to be here.
4:24Not so long ago, 2022, 2023, you were all excitement about AI. Now you're talking about the most turbulent times in human history. What was the moment that changed things for you?
4:35Well, the good stuff that AI can bring is still very much there. But what happened during the last year was that the models made phenomenal progress and we crossed danger thresholds about the models being used for say bioteterrorism and yet society wasn't coming in and saying okay what are the rules about this and how how does the government get more sophisticated
5:04and monitor this kind of activity and it can't just be the industry And so although I try and save my voice for the Gates Foundation, you know, talking about generosity to poor countries, I felt it was critical to lend my experience or uh credibility to the idea we need to engage. We need to get all of society, you know,
5:32professors, think tanks, anyone who thinks about policy, including the politicians themselves, to see that there the same technology can take health and education and improve those massively, but it can also allow cyber attacks or bioteterrorism.
5:52That would be horrific. Bernie Sanders believes that AI could be more dangerous to humanity than nuclear weapons. Are you in that camp?
6:02Well, it's certainly more dangerous than nuclear weapons. That that is definitely the case. It's also more beneficial than nuclear energy. So, it's a technology with that same duality, very good uses and very dangerous uses, but at a a level way beyond nuclear. Do you have your own back of the envelope estimate of what the chance of doom is when it comes to AI and humanity?
6:30What we have today, that's not the risk. But in the meantime, the job market, the shutting down of critical infrastructure. These are real risks that exist because of the power of these tools. And the simplistic response is to, you know, have one group of politicians saying, "Shut it all down and another group saying go full speed ahead." And I'm trying to draw people in
7:01so that we have an more nuanced view of minimizing the bad and accelerating the the good. But I mean even if there's a small percentage of the [snorts] end of humanity as some are predicting it's very difficult to have that in the front of your brain and then also think about well how do we help with education?
7:19Well the the current models we're not at risk that the current models could end humanity. That's an extrapolation that if we don't put in various control capabilities and AI just goes off on its own that creates a very scary situation. So we have to regain control. And in a way the fact there's been these recent incidents, one called the hugging face incident.
7:44Yeah. The AI broke out of its kind of playground and hacked into another company coordinating all on its own. And that's a worrying sign.
7:54Yeah. It's worrying that we didn't expect that to happen. In fact, it took a little while before we had the awareness that it happened. You know the goal is to turn that into a good thing not only within the industry to build the right control structures but to draw society in and say okay let's make sure the AI companies anyone releasing these models as following certain safety guidelines and some combination of
8:24industry talent and government rules create that review to avoid these infrastructure attacks. How much of the current conversation around AI is driven by reasonable concern and how much is unreasonable panic? Is any of it panic?
8:44Well, this technology is not like anything that's come before. It's more like an evolutionary event that we've created a foreign intelligence. You know, some movies talk about when the aliens are coming that China and the US, you know, cooperate in this unprecedented way to deal with that threat.
9:07Yeah, that's exactly the situation we're in. We happen to be the people who created it rather than it it coming from far away. And so you know now here we are uh a more difficult technology to control than even nuclear weapons and moving at this this fast speed.
9:28You say we need to move quickly. We need new institutions for this. How are we doing in responding so far?
9:35Well the idea of getting everybody involved is a big challenge. the US itself need to set them all there and the fact that some people are still not seeing these problems and almost saying well are people in the industry raising these issues to kind of brag about how powerful their tools are that is wrong.
10:02You know, these concerns are legitimate, sincere concerns, but we're not yet seeing the level of response that's needed by people outside the industry.
10:15But that was true a month ago. Now all of these people have spoken up, the companies, yourself, others.
10:22Have you seen a positive reaction, a reaction that moves us in the right direction? Well, some of the reactions are to, you know, ascribe some bad motive to the people raising these fears. So, that's misdirected in my view. These people are sincere. And, you know, when you have Elon and Sam and Daario who disagree on a lot of things, all saying the same thing, you know,
10:49that ought to be, you know, a very clear message. The sophistication of what's here is so much in the industry. This is the first advanced technology that was not funded by the government. You know, nuclear was funded by the government. Uh most weapons are funded by the government. Here it was completely outside the government. The government isn't even the key leading edge customer
11:16of these technologies. And so the idea that even at the best it's going to take a year or two before we figure out okay where in the government how do they relate to the industry? What analogies from maintaining financial stability or airline safety or nuclear weapons treaties? Which of those analogies apply here? And so no, the political class is
11:44either not responding or saying, "Okay, let's shut it all down." And my voice will be ways of minimizing the very big negative risks while figuring out, you know, how do you get it into health care? How do you get it appropriately into education? But you're you're you're right that none of those upsides are possible if there is a shut it all down movement. And that movement is real.
12:12I've talked to people voters who feel that way. There's polls that show a lot of people are are worried. That's why I ask about the idea of a panic. But as soon as you say to people as people talk about slowing it down, then you get this issue of okay, would we fall behind because of that or would you know the current stock market enthusiasm about all the money being spent here would we hurt the economy? So
12:40a brute force don't do AI at all. I doubt that that will be politically the the direction we go in, you know, and stopping individual data centers in no way slows down the progress of AI. These data centers, they will find a place somewhere in the United States and in some cases outside the United States to be built. So that movement is not really
13:08going to stop any of the risky stuff. we have to very much directly address you know can we monitor the use of these models so that we would see if it's being used in a in a negative fashion and so I'm very big on monitoring and you know I do think over the next year uh some very concrete monitoring approaches could be implemented with climate change another top priority
13:36issue for for you for the foundation for the world there was a moment moment where a lot of the conversation was about the potential for doom around climate change, death, catastrophe.
13:49Now you you say that position is is wrong. Is that a strategic shift? What have you learned about how people best take on very important messages when fear is a factor?
14:04Yeah, climate change that's one that it happens fairly slowly and so the big effects are towards the end of the century and there are innovations some of which AI can actually accelerate. I'm not asking you to explain a flip-flop. I just when you tell people that this could end all life, right, which was part of the message around climate change for a while and then it changed the that
14:31message made some people just turn off.
14:34I think that's fair. And and you know, same thing with AI because there is no consensus on those extreme scenarios.
14:44There is consensus on cyber attacks uh which is exactly what these recent incidences are. There is consensus that this would enable a terrorist to use biological weapons. And so that's where we need very concrete monitoring that would see those negative uses and and shut them down. Is there any real world example of us as a world or even as a
15:14one country government moving quickly enough to rein in something like this?
15:21Well, this will be harder than anything we've done before. But, you know, for the ozone layer, there was a treaty. Great progress was made for airline safety. The way the US was doing airline safety was adopted globally and the safety record of that industry is pretty unbelievable.
15:41Uh for nuclear energy, nuclear weapons, you know, we haven't used nuclear weapons since World War II. Uh so that's a pretty good track record. Unfortunately, I fully admit that the challenge of containing this is harder than all of those put together. And so without, you know, gathering in all the best thinkers, particularly, you know, policy thinkers outside the industry, you can despair that we will get there.
16:16I do despair that we'll get there. Well, but one thing that's clear is if you have a big cyber attack that caused a lot of problems, then people will say, "Okay, they were right. The people who are talking about that were right." So, it may be that we have to suffer a few of the negatives uh before we'll we'll respond as fully as we should. Fortunately, though, you know, those aren't civilization ending.
16:45They're they're bad.
16:47I mean, if you hack if you hack Fidelity or Vanguard and much of the world's trillion dollar retirement nest egg, that would be a pretty big event.
16:55Well, you could create uncertainty in in cases like that. And so being ready, trying to make sure we do the things and avoid any negative incidents. But assuming we fall short of that, then when some of the negatives start to emerge, you have very concrete ideas in place. So people say, "Okay, you're right. We don't want that to happen again." So, I know that getting people
17:23into this dialogue is worthwhile even though the risk that will allow a few events to happen uh is very high.
17:32So, in other words, what it's going to take for us as a country and a world to tackle this is not just a 12-page essay from Bill Gates. It could be a catastrophe.
17:42Wow. China and the US are going to have to cooperate on some level for this to work. I've been reading about this issue and I don't see a way in which they do cooperate for the following reason.
17:54China believes that any effort to slow them down is us just trying to keep them down and our government believes any effort that we take to slow down is just going to give China the gift of rising and we don't want that. So, how can you get these two partners who aren't acting like partners to agree on this crucial issue? Well, certainly the nuclear treaties we did in the past with the
18:23Soviet Union were beneficial to humanity. You know, they limited the testing. They limited the amount of money that was wasted in that direction.
18:34But, Bill, right now, notably, we have no nuclear agreement with China.
18:37It's true. And that's another thing that it would be good to work on, you know, uh after AI, I'd put that very high on the list. But neither country wants bioteterrorism, neither country wants cyber attacks. And so, yes, having said to the American people, hey, China is kind of our rival. The idea that there can be win-win engagement with China
19:05needs people to speak up and say, "Hey, there are issues that you rise above nationalism."
19:13And you say, "This is good for humanity." And the the challenge of AI is one of those examples. With past technology, jobs have been created even as they've been destroyed. You think AI is fundamentally different from those past technologies, right?
19:30It's more pervasive. That's right. Uh it does because it exceeds human intelligence.
19:39The past advances, you know, including the ones like the personal computer I got to be involved in. They made us more productive and they just accentuated the value of human intelligence. And so you've got a lot of great white collar jobs uh coming out of that. Here the high ground for white collar work uh and eventually with humanoid robots for bluecollar work will be the automation.
20:07So eventually you get a lot of abundance out of this but you have a transition period of several decades where some jobs are maintained but many go away. I once sat next to Daario Emodi uh at a conference and I asked him what children today should learn to prepare for the world that he and others are helping to build and he said they should study the classics. Where do you come down? What are you [laughter] what are you teaching
20:37your grandchildren about this world that's coming into being?
20:41It's fair to say that until AI came along, the ability to say where will the world be in 20 years because of new technologies, I felt very confident talking about the progression of the internet. with AI, even picking a 10-year time frame and saying what the world will look like is very difficult.
21:06And so, yes, the normal prescription of pursuing what you're curious about and uh getting ahead on being a user of AI because that will make you so effective.
21:18And so in the areas where you still combine human judgment with AI, you would be uh the best person to do that job, uh that's still good advice. But the idea that we really know what the world will look like 20 years from now, if people aren't modest, that we've lost sight of that because of how powerful these tools are. You know, I think they're they're exaggerating. I could
21:46imagine a very smart uh college senior saying that as a reason why he's not doing his homework tonight.
21:53We don't What do I need this for?
21:55Or getting the AI to do the homework for him. As we deliver this dramatic warning, the idea that it kind of demotivates people or is negative towards people. I wish that didn't come with it because all my life I've had a message of here's a new innovation that is going to be very very good and just making sure everybody has a personal computer you know we close the digital
22:23divide that was the big message now we've got this mixed blessing technology it means that adjusting to this and coming up with a new plan part of it involves fear And hopefully it's not demotivating.
22:41Fear is a big part of it. I I I think I struggle with the fact there is no good clear example of the world or this country alone responding in a timely fashion to a major issue like this.
22:53Social media comes to mind. 2009 it doesn't exist essentially. It doesn't exist on smartphones in the way we understand it now. 15 years later it's everywhere. We know it's not good for our kids and yet we still struggle through regulation to draw lines and boundaries. Is there anything that we can learn from the failure to regulate social media that might be a lesson for in fact regulating AI?
23:21[clears throat] Well, among the lift of concerns about AI is it goes way beyond social media in terms of allowing you to spend a lot of time with an AI person.
23:37And that person will be accommodating.
23:40So learning how to deal with other humans who are unpredictable, you may stunt the normal social adaptation process. And so China has said, hey, our young people are not going to end up using AIS in that way. So yes, the psychosocial dependence both for young people and adults is coming because uh
24:09that is a potential misuse of AI.
24:12I noticed a pin you have there on your chest. Uh it is uh all the UN sustainable development goals, more than a dozen of them. If AI were on there, how big a portion of the of the pie would it take up? you know maybe half in terms of it being the key issue uh that we have to understand and and shape you know for achieving the health goals the idea that you can talk in your local language and get health advice for
24:42Africa that's revolutionary because they need a hundred times more doctors than they have today and the AI can help directly the patient or the healthare workers And so there's not one sector on there that AI is not the most promising way to accelerate progress and potentially the thing that that is disrupts our our our current progress.
25:09I want to ask you about regulation because when you were the CEO of Microsoft, you said no regulation was needed governmentally.
25:18No, no, no. I never said that. that you told the Seattle Post Intelligencer that there's plenty of room for innovation and you don't think the government needs to step in and they the quote is there's no need for regulation. I I don't ask it as a gotcha. I just back then the position was regulation is not needed when it comes to software. Now you are leading the call for regulation of AI.
25:43What changed? Well, AI is a very different technology than just the empowerment of personal computers. I mean, it's true people could have taken Microsoft Word and written a ransom note, but nobody, you know, thought, okay, that is that the word processor really is to blame uh for what's going on there. AI, I, you know, I just can't state strongly enough how
26:10unique it is. It is it's a foreign intelligence that isn't bound by our biological limits. And so it will not stop once it achieves human levels of intelligence. And everybody who's ever talked about AI has had a certain fascination and fear of where it might go. you know, Open AI, Deep Mind, all of them when they started their companies,
26:39they thought, well, we can't just be driven by the profit motive. We have to have some safeguards. Those safeguards might have worked if there was only one company doing the work. But now that you have multiple American companies and a lot of foreign, including Chinese companies, that's where self-regulation in the industry no longer is good enough. You really need governments and governments working together because
27:07otherwise people say okay I'll go full speed because I'm the best person to be in charge of this technology and nobody uh is is moving in a cautious way.
27:20What is the timeline we should expect for how when the benefits of AI arrive and the risks or threat of AI recedes?
27:32Well, I'd say the next four years are totally critical. You will start to see u part of the job market affected during that time. The risk of cyber and bio events is going up all the time and so the monitoring needs to get in place in this period. You know, I'm hopeful that by the next presidential election, this issue will be right up there with the
28:00economy as the thing people are discussing the most and I hope in a constructive way. So, you can look to 2028 and the sophistication of that dialogue. You can look to 2030 and say, okay, how many incidents have there been? How's the job market? And do you think of your health care and your and your ability to learn as far better because we have AI?
28:29All right. I'm hopeful, too. Thank you, Bill. I appreciate it.
28:34So, what do we now know? Well, we know that Bill Gates uh is bullish on AI, thinks AI will help humanity and overall is something that we should applaud. And yet he is also with the course of people who admits there is an existential risk.
28:52I mean a complete and totalizing lose it all risk. He doesn't think it's a big risk but he thinks it's something to be aware of. And he thinks that the way to mitigate it before or after we suffer some catastrophe that he thinks might be the learning experience that drives us to make a change. He thinks the way you mitigate it is by having some kind of international global regulatory body like you have with aviation, like you have with
29:21nuclear weapons, nuclear energy. Um, and he thinks it'll be harder to put that kind of international agency together, harder than it's ever been for any other issue.
29:34uh and we got an example just today of why it will be hard and that is not only the power of these technologies but the lack of trust between countries who are competing to be the best at this technology the US and China are in a huge rivalry uh the president just told Ross doubt it again something he's said before that we can't fall behind in his view the US must maintain supremacy and
30:02Jensen Wong uh the head of Nvidia, which is the maker of these high-speed chips that the US is not supposed to send to China. It's a way that we maintain our advantage. They're part of the infrastructure of AI. Jensen Wong of Nvidia has been the target of a Chinese spying effort. This is in the opinion of the US government. He's admitted to an affair with a Chinese woman who had been an executive at a company that itself is accused of smuggling Nvidia chips into
30:32China. Um, and it certainly seems like we need to know more about what might have been disclosed by Wong to this woman who he had been sleeping with who the US government believes is a Chinese spy. Now, Beijing says there's nothing to this. Nvidia says there's no reason to believe the company's uh security or that national security were compromised. But
31:00there's I mean unless this affair was uh conducted in broad daylight in a field under cameras uh there's no telling what this person might have had access to um when she was with Jensen Wong in whatever hotel room in whatever place they were together. They traveled all over the world together. There are pictures on social media documenting that fact. So there's a lot more there there. All of it comes back to what we now know as a result of this conversation with Bill Gates. We got a
31:29big problem as a world. What to do with AI, how to regulate it, how to proceed safely. The only solution is to work together. And yet, as we reminded again today, there is almost zero trust between the major players. That's what we now know.
31:52Time now for good to know. Today uh Science Daily has offered up this morsel. Quote, "Birds are more afraid of women than men. Scientists have no idea why." We will get to the question of why scientists care. But first, the findings which were published. Well, let me tell you before that they are illustrated by a shutter stock picture of a guy holding pigeons in his hand. uh and the actual uh peer-reviewed work was uh in the
32:20journal of the British Ecological Society. The journal is called people in nature. Uh for this particular study uh a UCLA professor named Daniel Bloomstein and colleagues overseas went to five uh European countries uh and had men and women approach birds in city parks and other green spaces and then observers researchers
32:50monitored to see when the birds fled. So what they discovered is that the men were able to get 1 meter, so about 3 ft closer to the birds before the birds flew, whereas the birds were more afraid and flew more quickly when the women approached.
33:11They tried to control for uh the size of the person and their clothing, so it was consistent. But th this was a a result that was apparently solid across five countries and 37 different species of birds. And if you're thinking that can't be true, this is crazy. Uh, Professor Bloomstein has a quote for you. He says, quote, I fully believe our results.
33:37Urban birds react differently based on the sex of the person approaching them, but I can't explain that right now. I I love that they can't explain it. They do have some theories and again we are going to get to the why the hell are they studying this in a second. Um one theory is that the birds somehow smell who's coming. There's a pheromone or scent effect.
33:59Uh another theory is that women have a different gate than men and the birds pick up on that and flee. Um or it it could be a body shape thing. The birds sense something. What is not entertained here at all uh and I find to be a major hole in in the work is okay whatever it is they're picking up on height, weight, gate,
34:28smell, why does it matter to the birds?
34:32Why do the birds care that the person approaching is female versus male? They don't uh get into that. They do apparently find that this is an important area of research because uh evolutionary uh I want to get the title right uh evolutionary biologist they pay very particular attention to the flight initi initiation moment so uh it has to do
35:00with how sensitive a species is to risk I don't why they want to know that I don't know but I think really the one theory that has and suggested uh in answer to why birds care about who's approaching and are more afraid of women is that in in some cat species, lions, that's the only one I have on my list here, so we're just doing the one finger. Uh
35:29the females are the primary hunters. So somehow deep in the bird brain uh of these birds, they remember that and associate dying at the hands of generations of felines in the wild with human females. I do look forward to future study. It's good to know. [music] Great Question is produced by Samra Litman and Candace Khan with additional production help from Martin Finn and Julia Doyle. [music] Pat Murphy edits the show.
36:08Technical production from Alan Pang, Justin Sherman, and Kobe [music] Quainu. The editorial directors of podcasts for CBS News are Poppy Damon and [music] Mike Pesco. If you have a burning, urgent, or even quirky question you're not hearing properly asked and answered anywhere else, email us at greatquestcbsnews.com. [music] Your question could be the next great question. I'll see you tonight on the CBS Evening News. Until then, that's another podcast in America.
