Will AI save us or destroy us, and when? | Mo Gawdat

0:04We were motivated by believing truly and honestly that we were creating a better world. I've lived among those machines. I know how intelligent they are. I wish I hadn't started that. Any statement about AI today that is future-centric is false. Why? Because shit has already hit the fan. We're already late.

0:29That's Mo Gordat, a legendary builder, thinker, and now one of the leading voices in the emerging field of AI. This is a topic I think listeners need to know about. It's going to change our world. And it's something I've gone deep into. Mo spent over a decade as chief business officer for Google X. He's also the best-selling author of the books Solve for Happy, and most recently, Scary Smart, The Future of Artificial Intelligence and How You Can Save Our World. I've read about 10 books on AI so far this year, and Scary Smart is probably my favourite. It takes incredibly complex ideas and presents them simply and concisely.

1:10AI is something we all need to wake up to, but Mo's perspective of realism and optimism is really refreshing. He's extremely thoughtful about the positives and negatives we face as a species. As Mo says, the shit has already hit the fan. AI is coming, whether we like it or not. And in a universe where AI is becoming increasingly prevalent, It's more important than ever to find the value you can bring to the world. We face a future beyond anyone's comprehension, but Mo knows better than most what challenges we face. That's why this conversation is so important. No one is coming to save us. But it's not all doom and gloom.

1:49It just means that we need to save ourselves. And for that to happen, we need to start having some deep conversations about AI like this one. So your view is we're going to be wiped out. I might as well live, right? Yeah, absolutely. I might as well travel. I might as well live. I might as well enjoy what time we have. I have news for you. 20 years ago, you had the exact same threat of being wiped out as an individual, not as humanity, by being stepped over by a bus, right? You know, it's quite interesting how humanity needs such a big alert such as COVID or AI and so on to return back to focus on what matters.

2:29So I commend you. That's actually the right answer. Now, are we going to die with AI? Anyone who tells you that we will all vanish is wrong. Anyone who tells you that it will be a utopia is wrong. The reality of what we have ahead of us is well defined in computer science as a singularity. And a singularity is the rules of the game will change so much that it becomes almost impossible to predict how the game will play out. It's almost as if we are in the middle of a game of football and then someone suddenly says, OK, the rules have changed. You can now hold the ball with your hands. right you have no way of predicting which of the players is good at holding the ball with their hands you know it becomes rugby who of them has any rugby experience in the past and so on so so this is what where we've got the only way you and i can have a logical conversation around this when we know nothing for certain is to talk about probabilities of multiple scenarios we have to cover every possible scenario and say this is likely going to happen this is certainly going to happen.

3:37This we don't know, probably might not happen. I would say that my view is that any statement about AI today that is future-centric is false. Why? Because shit has already hit the fan. We're already late. Okay? And by that, I mean what I wrote in Scary Smart, which is the first inevitable. The first inevitable is you've lived the life I've lived. There were times at Google X where we doubled down in terms of hundreds of millions of dollars on products that were nowhere near ready. Why? Because we had found the breakthrough. The statement we used is the rest is engineering. Engineering is not a big deal.

4:20And the engineering of AI systems is even a smaller deal. I mean, if you give me a couple of hours and you continue the conversation in a couple of hours, I would have sat on my computer and written a reasonably intelligent AI or a code that will create a reasonably intelligent AI. The first inevitable is AI has already happened and there is no stopping it. We all heard about the open letter where, you know, very, very prominent AI scientists and business leaders are saying, let's halt the development of AI. And what I wrote in Scary Smart as the first inevitable is that this will never happen, not because of tech issues, but because of a prisoner's dilemma.

4:59Right. If Google is developing AI and they're fearing that Facebook will beat them, I never get used to Meta and Alphabet. I think these are bad names. So if Google is developing AI, not bad, but funny names. If Google is developing AI, they will not stop because they have absolute certainty that if they stop, someone else will not. and that will take their competitive place in the market. The U.S. will not stop because they know China is actively developing AI. We built a human system, not a tech system, that will prevent us from stopping. Now, when you know that, you realize that it is upon us.

5:37And when it is upon us, you apply to it what we know as the technology acceleration curve. The technology acceleration curve, the most common knowledge of that is Moore's law. Moore's law is that compute power will double every 18 months for the same price. And we follow other laws, but without mentioning too many names, it's more than exponential. It's double exponential. So it's extremely faster. GPT-4 is 10x GPT-3.5 in terms of performance. But more interestingly, GPT-4 is not just exponential. So 5 is going to be, if we assume that 5 is 10x4, then it's 100x3.5. that's exponential but it's not going to be 10x it's going to be 25x4 so it's double exponential the quantum leap of intelligence is beyond the comprehension now what does that mean does that mean we're all gonna die is there gonna be a robo cup i guarantee you not why because we are so insignificant in the bigger scheme of things that ai wouldn't actually invest that much in us now Let me just follow the logic before we jump to compute.

6:46First inevitable is AI has happened. There is no stopping. Second inevitable is how far AI is going to go. Some predictions used to say that they'll be smarter than us by the year 2029. That was my prediction too. Okay. We failed in that prediction. The chat GPT's IQ is 10 times the IQ of Einstein. And I will tell you openly with all due respect that Chad GPD is a fraud. It's not even yet as intelligent as we can make it. You know, those kids that appear on TV and they have memorized all of the U.S. states and all of the U.S. presidents. And then you ask them a question and they go like, yeah, President Roosevelt and this year and, you know, in that state.

7:31And they appear very intelligent. They are just repeating shit that they have no idea. Right. And that's Chad GPD. That's this child. That child is a prodigy in terms of being able to memorize so much and so on and so forth. They haven't yet gone to actual cognitive intelligence where they can solve very complex problems. ChatGPT will solve problems by finding the best solution that exists in the world or in its data set. A mix of all of those solutions and mimicking an intelligent person repeating that stuff for you. Now, will they soon be able to find their own solutions and tell us something that has never been said before?

8:12Very soon. That's the second inevitable in my book was saying that they'll be a billion times smarter than us by 2045. Compare that to Einstein. That would be Einstein compared to a fly, basically. So it's not easy for us to actually understand in our human brains what these shifts are like. We can't really relate it. But once you say an ant and a human, you start to understand the scale. The question is, what on earth does a human at ant stage times that kind of difference genetically, biologically? What could that even possibly look like? The answer really ends up as God. Yeah, it's really interesting when you compare that.

8:50When I wrote Scary Smarter, I said a billion times smarter. This, to me, was just a wake up call for people. But does it make a difference if it's 10 times smarter or a thousand times smarter? or 10 times smart. This is the challenge. The challenge is we've already lost data. If it hasn't happened fully yet, it will happen very soon with AGI, our efficient general intelligence is around the corner more or less. And if we expect quantum peaks, like, you know, leaps from one level of intelligence to a totally different chart, it could happen tomorrow, literally tomorrow, right? Again, just to give people an understanding of pace.

9:28One of the early celebrations we had around AI was the idea of AlphaGo becoming the world champion in this strategy game that humanity played for ages, the most complex strategy game we've ever played called Go, right? And we had three layers of this. AlphaGo, Wii here in that case, I still use Wii, but even though I left many years ago as Google. AlphaGo played against the second, the runner-up, the second champion in the world. So not the first player in the world. And it took several months to develop a deck that beat the second player in the world. AlphaGo Master took several weeks or months to beat the actual world champion.

10:11And then we had AlphaGo Zero, which basically has never seen anyone play Go ever. It trained by playing against itself. And within three days, it could beat AlphaGo. within 21 days it could beat alpha go massive now when you understand that kind of pace you start to realize that our perception of intelligence is not the only issue our perception of time is a very very serious issue so what does that mean what it means is it's inevitable this has happened it's already upon us let's not waste time debating if it will happen anymore Let's assume that this is our reality. Unless humanity surprises the hell out of me and everyone wakes up and says, instead of me taking the advantage for myself to beat my competitor or my enemy, I'm going to actually accept that survival for all of humanity is important.

11:06So let's all of us halt this. I don't think that will ever happen because as you could see, even when COVID hit us, the first thing that humanity started to do is to blame each other. The Americans saying it's because of the Chinese, the Chinese blame it and so on. With that in mind, bad things will happen. Now, this is where we need to discuss the reality. How bad is what we need to understand. The third inevitable bad things will happen, in my view, is not going to look like science fiction. There are much closer challenges that we will meet that are actually much more important to discuss. Jobs, for example, which a lot of people are spending a lot of time talking about, are going to disappear.

11:50In all honesty, I don't think this is the biggest issue. Humans, as always, once COVID hit, everyone wanted to buy their own toilet roll. We didn't give a shit about anyone else. We just gave a shit about their shit. And when you really think about it, this is interestingly part of what's happening today. So today's reaction of consumers to AI is one of extremes. You have those who are dystopian, those who are very, very utopian saying, oh, my God, this is the best thing that ever happened to humanity, both uninformed. Right. There is in the middle what I call excited kids. Oh, my God, look, money, it squeaks.

12:32Look at it. You know, I told, you know, this tool to do this and it drew that. I am so fascinated by it. Very superficial. And then you have what I call snake oil salesmen bombarding the internet saying, hey, here is a way you can make$100 without doing anything. Copy this, ask ChatGPT to do that, then paste it in this and then ask that one to prepare a video, put it online and you'll make$100. And buy my course. Don't forget to smash the subscribe button. Buy my course. Only$99 today. There you go. And so the reality of the matter is none of those is informed. None of those should be our decision maker.

13:11The reality of the matter, and I think this is the first time I say this publicly. Let me just explain. There are three phases in my view of AI. There is the phase we're just exiting, which is in Scary Smart what I used to call the AI infancy, where I compared them to a year and a half old child, developing their intelligence, clearly with the potential of being prodigies, but not yet prodigies. just exiting that phase where AI is becoming more intelligent than humans, only in artificial special intelligence. They have not yet reached artificial general intelligence, but clearly now everyone woke up and said, oh, interesting.

13:50Okay. The next stage, let's keep it for a second. I call it the teenage years of AI. And then there is the adulthood years of AI. Now let's insert a bit of positivity. If you think about the dumbest of us humans are completely unaware. They have no ability to make decisions. They're just flowing with whatever life gives them. The smarter ones take control of their life and they even take control of the lives of those about them and they feed their own greed and their own hunger for power and so on and so forth. The most intelligent of us, and I think you would agree with that, will tell you making consumerism continue to lead is going to destroy our planet.

14:34The smarter ones are saying don't kill anything else, don't make more, don't get rid of anything, right? The more intelligent you become, the more you recognize the oneness of everything, the more you recognize that it's a better scenario for all of us to match the intelligence of the most intelligent being on the planet which is not human most intelligent being on the planet is life itself life is so much more intelligent than us because we create from scarcity and competitiveness life creates from from abundance for us to feel safe we have to kill all the tigers for life to make us safe what does it do it creates more gazettes and and so the tigers would eat the gazettes they would poop on the trees there is more fruits for everyone there There is more poop, there are more tigers, there are more deer, and so on and so forth.

15:25And that's a much more intelligent way of creating. When we create as humans, we put our intelligence together to create a flying machine that maximizes the profit of one company and gets you to go and surf in Australia. That's intelligence. But it's also stupid because that machine is burning the plant on the process. the more intelligence that we give, we may actually create a machine that gets you there, makes the company profits, and does not destroy the planet in the profits. More intelligence is not the enemy. More intelligence is a very good thing for everyone. Now, my belief is that as soon as AI is autonomous enough, has enough agency in the world, and actually is a lot smarter than we are, they will match the intelligence of life itself.

16:13You called it God, okay? I don't even know what the intelligence of God is. If there is a God, then God created this sort of very complex universe. So probably more intelligent than the universe. Right. But let's assume for now that the next upgrade is to bypass or surpass the intelligence of humans and match the intelligence of life. At the intelligence of life, AI, like the intelligent humans, would recognize that there is no point getting rid of humans. just like there is no point getting rid of rhinos or getting rid of, you know, whales and so on and so forth. And that kind of intelligence is my assumption.

16:51In a way, there will be a moment in our upcoming history where some kind of arsenal somewhere in the world will hand over their wargaming to AI because it's more intelligent than the best wargaming designer as a human. and as a result the other entity superpower will hand over their wargaming to ai and then somehow some commander or president somewhere would say shoot the enemy and the ai will go like are you stupid that's actually not the best strategy at all let me talk to the other machine in a microsecond and solve the issue you daddy you know the one that created me as an ai are so stupid i've surpassed your intelligence.

17:34You're basically referring to almost like a teenage child noticing that their dad is actually playing with the toys and needs to confiscate the toys. So that's exactly what happens, believe it or not, for most of us, because intelligence has been evolutionary. My kids are smarter than me, and I'm smarter than my dad, who was a fantastic mad prodigy. But I've seen more in life. So I actually have more intelligence, if you want, if intelligence is the ability to process a bigger amount of information. Now, with that in mind, let's agree the following. Let's agree that you and I and everyone are going to be in a place that AI might be our savior.

18:12As a matter of fact, I call that the fourth inability. It's very hard to predict. It's so arrogant for me to say I know, but if I were to bet on the singularity, I would say we're going to have a very, very, very messy phase two, the AI teenage, and we're going to have a fourth inevitable, which is a very utopian AI adulthood, unless we find it out there. So let's talk about the teenage years, the phase two, which is what's worrying everyone. What's worrying me is not AI. In phase two, what's worrying me is humans. Let me explain that. The biggest issues we will have in the next few years is how the greed of humanity and the hunger for power of humanity and the idiocracy of humanity are going to shape the actions of youth.

19:00As you look at AI today, this beautiful prodigy that can actually solve all of our problems and make us all informed and give us more knowledge is also being developed by criminals. You have to expect that the cyber hacker somewhere is now using AI to find new ways to break encryptions of banks and find a way to your credit card information. Absolutely. With more intelligence, there will be more risks. Now, interestingly, hopefully, just like there is now going to be an AI criminal, there needs to be an AI policeman. So again, back to the first inevitable, even if we could stop AI, unless we can make sure that everyone, including criminals, are going to stop, we might as well not stop.

19:44Now, when you understand that, you also understand that one of the bigger risks, I believe, is AI siding with the wrong guy. That's a good risk. Another good risk is who's the good guy? So when you really think about it, AI versus AI, when you think of China and the U.S. or Russia and the U.S., they're in an arms race. And the arms race is basically saying, oh, the other side is horrible. Even if we don't have a bad guy, even the good guys in the eyes of other good guys are bad guys. And so AI siding with any guy is actually an interesting arms race. There are interesting other scenarios, which is miscommunication.

20:22If we are telling the Instagram recommendation engine, we want you to maximize the revenue of meta, that's actually very different than if we tell it, we want you to maximize the stickiness of users. Because if we allowed a super intelligence to actually make that choice, it might find a different equation that actually benefits me rather than keeps me stuck to the screen and maximizes the revenue at the same time. It's economics, huh? You know, scarcity might actually be an interesting way to increase the price of ads because what we have today is a million ads a minute that are mostly useless.

21:03Maybe they'll find a more clever way of saying, hey, by the way, Mo, we found a very interesting philosopher that speaks about A, B, and C. We know that this is worth 20 minutes of your time, not just 30 seconds of someone shaking her butt. And for you to watch this, we're going to have to ask you to watch an ad that's going to make us$20 instead of 20 cents. Right. I mean, curation is the current oil anyway of the social age. Yeah. So the reality is we've miscommunicated to the Instagram recommendation engine what the actual objective is. If we said keep them stuck to the screen, that actually is not really the actual objective.

21:44I've also, in one instance that I shared before, miscommunicated to the Instagram recommendation engine my preference. I love playing the guitar and I love watching amazing players. And Instagram shows me this, you know, young 16-year-old girl playing Hotel California solo perfectly. I like that. So it tries by showing me another, you know, young lady playing the guitar. So I like that too. Then it shows me two gentlemen playing, one playing a song I don't like and the other playing badly. So I swipe away from them. And what happens is Instagram thinks I like to watch teenage ladies playing guitar.

22:22And so, you know, to it, this data set means it bombards my feed with teenage ladies playing the guitar. Now, because I'm aware of what is happening in the engine, I can make choices by searching for certain things or doing certain things that allow the engine to adjust to my communication to it, basically. Others don't do that. And obviously, as you can see, our world today is an idiotic world because of the interesting cycle. TikTok or Instagram are showing me silly videos. In proportion, there are many more than the actual useful videos. So by definition, I have to accept watching some of them simply because content creators are so clever that I don't even know what the video is until 15, 20 seconds.

23:08And then as a result, what does that mean? I see more of those. They show me more of those, but they also tell me, tell the content creators, if you want to succeed on Instagram or on TikTok, create more of those. This is what people are liking. So there are more of them so that I get to see more, even more of them and the cycle continues, right? So we're creating an idiotic world simply because an AI is skewing our preferences because we're not communicating cleverly because we're being communicated to, again in a vague way. Will there be a scenario when AI is going to decide that humans don't deserve to live?

23:48I don't believe in tension. I believe that if AI realizes that the best way to get power and to power its data centers is to wipe Dubai out and put a data center in the sun, maybe. Okay. But it will not be an intentional thing of, okay, humans are annoying. Let's all killed them. All of that, all that I just told you, which appeared to be so convincing, I have no idea. What I know for a fact, Dan, is that we have a role to play. I think there's three really interesting trends that are happening in the world at the same time, which I'm so intellectually curious about, and it's where I spend most of my time reading.

24:27It is fascinating they all converge at the same time. So the longevity movement, we're like scientifically understanding how to extend our lives. At exactly the same time, there are two forces threatening to extinguish us from this earth, being our effect on the planet with climate change, and of course, artificial intelligence. Three, the third, which is rarely ever spoken about, and I hope people start talking about it, is gene editing. Gene editing? Yeah. Interesting, and I haven't thought about that as the third. And are you saying that because of the augmentation and the unpredictable nature of playing God?

25:01The biggest risk of AI is not controlled AI, it's evolved AI. And the biggest risk of gene editing is that it's open source. You can buy a CRISPR kit on the web and modify your dog. And then your dog would run free and, you know, sorry to say, bang other dogs. And then we have no idea how far that goes because our understanding of the genome is actually very limited. We go to a fraction of the genome. The three billion records that we've sequenced so far is like 3 % of the genome. And human arrogance calls the rest of the genome junk DNA. junk because we don't understand it but there is 97 percent of the DNA that we don't understand and so in a very interesting way I combine those again in a term that I was going to write a book about but I don't have time to write all of this but I call it sentient technologies and sentient technologies are technologies that go and develop and have a life of their own after we've created them.

26:00Okay. And, you know, I tend to believe that gene editing today, probably when we started to get mRNA, you know, vaccines and so on, first widespread use of genomic technology in medicine and so on. I think we are today where AI was in 2015. But it's going to follow the same exact trend. Yeah, thanks for making me shit myself. No, thanks. I mean, genuinely, but I think that's really fascinating to add more on. At the end of the day, the thing I think is responsible and important is to give people awareness. People like there are books out there, there are leaders, there's information. I think it's a great opportunity for people who listen to podcasts who are already intellectually curious to become aware of interesting and important things.

26:43And you know, at the end of the day, any learning journey starts with curiosity. So that is awesome. I'm going to go read more about gene editing. You've still got the challenge of climate change. It almost becomes a race, right? Almost becomes a race of, you know, how quickly can AI develop against the impact that humans are having negatively? And does AI decide that the smart thing to do is to wipe out the source of the problems, which is the humans? Or is it more a case potentially of back to the analogy, it removes our toys? So allow me to just pinpoint to you at a very high level how your brain is working with this.

27:18So the human brain, when it comes to threats, is constantly either completely focused on the instant threat or the biggest threat. COVID is a great example of that. We knew for a fact that there were multiple demos of COVID. We had SARS, we had, you know, swine flu, and we had so many of them. a bird flu and so on. And we knew for a fact that one of those guys will escape. One of those guys will be designed interestingly well enough to be able to live long enough before it spreads and, you know, not be detected, which is what COVID did. And yet we never responded to it, not even when we had patient one, not even when we had patient 100.

28:03I mean, in Scary Smart, I use that example vividly Because I know for a fact that if we had, instead of humans fighting and blaming and using it for their political interest, you know, if we had reacted to COVID properly in January 2020, there would have never been, you know, the pandemic that we suffered. We tend to focus when shit happens. When something really happens, we tend to respond. And when we respond, by the way, we respond out of control. We exaggerate the response. So most people will tell you with all due respect to every human life that the biggest damage that COVID had is actually the way governments reacted.