Support for this episode comes from click up. We lose an average of three hours every day, switching between all our work apps, but you can get them back with click up a flexible platform that brings all your essential tools in one place we can prioritized Six collaborated docks check with your team and track goals, so companies like Oberon web flow use click up as their mission control Centre for placing every other after we're using before cook up even Aren T used to help you save one day week and get more done. It's completely. Customizable is free forever, so try click up today. A click up dot com, slash the weeds support for those at this hour comes from indigo at INDIGO brings together leading companies committed to activating the full potential of agriculture to address the climate crisis in partnership with innovative businesses, farmer, scientific partners and informed advocates. We can re. Imagine agriculture for the benefit of people and the planet. Farmer, Casey, Bryant Bamberger says our industry hives biggest platform.

Change our environment and working towards a better future, learn more it. INDIGO Ag, Dotcom, Slash, Reechoed, my guess is weak is calls you, but she is actually box writer part of the future perfect team I've been getting more interested in the kind of stuff TAT Kelsey covers in the world of rationalism, effective, altruism, these kind of things will people try to really think to rigorously what are the most important problems in the called it and why do they matter? The one I have always been stuck on is the level of interest in this community in artificial intelligence and its possible threats humanity. I suppose you have been talking for a long time- but have we should sit down and talk this through outweighs? Finally, you know put the date on the calendar decided. We should do it up with record the conversation after everybody's benefit. I think it's really interesting, I'm not sure if she's totally convinced me, but you know there is a lot to learn here so enjoy.

Hello welcome to another episode of weed on the box. Vidia podcast network, I'm Mathew Glacius, my guest today, Kelsey Piper is a staff writer with boxes future perfect. Welcome to the show Kelsey thanks excited to do this. We've been talking about doing this for like a year yeah, so I have been I've been interested in the kind of areas that you work again in the sort of you now ideas around rationalism, effective, altruism. I think you know there's there's some concepts that I struggle with, as, like personal ethics like I need to be better to animals and things like that, but what I don't feel like I've ever quite gotten, intellectually, is like. What's up with artificial intelligence, we hear that from

of people, because you know effective, altruism, tourism, like malaria nets, treatment of like parasitic worms in poor children, even factory farming. You know it's a little farther. He opened a thing was people I kind of like yeah factory. Farming isn't great, am not proud to be part of that, and then there's a focus on future technologies. That inherently is a little bit more speculative because their future technologies- and that has a lot of people going. Yet why? Why is this a major global priority? Why and also because Read said, there's a a difference in moon right. Is a lot of this stuff is like ok, just take a deeper right, and, like really try to say, like look what man sprite like weakest. Of course, just we fight in politics all the time about like a merit. And human beings in America and other people. We now

I think, if you say to most people just like on its face, isn't actually severe poverty in Sub Saharan Africa like worse than whatever it is you yelling about, and then I got that like that. That seems that seems correct, whereas, like these science fiction scenarios, which I think would get the audio a guy, I loved a terminator too ass occasion recite my favorite Skynet funding bill is passed. The system goes on line on August for ninety ninety seven human. I moved from strategic defense. Skynet begins to learn at a geometric rate. It becomes self aware to fourteen a M eastern time August, twenty nine th in the panic they try to pull the plug. Skynet fights back. Yes, yes, yes- and this is what it's about right is artificial intelligence will get out of control and pose an existential threat to humanity. I'm so

Here then am I gathered also like I do. I do love that movie, but, like you suffer real, so I dont think I risk looked much like her later, and I do think there. I risk work has been sort of damaged by the fact that yeah there's all this crazy cipher, where, like the robots, develop a deep loathing for humanity and they come, their guns and then they shoot us all down and only one time travel. You know that's ridiculous, and so of course it that's what people are thinking up when they think about the effects of eye on society in other gonna, be like that's ridiculous. I do think, there's like less of a divide between I and the other areas that future perfect focused design than it looks at first glance, though, like ok, we distribute malaria nets because that stops kids for dying of Malaria Sub Saharan Africa that straightforward. But then we also fund malaria, vaccine development and that's a lot more speculative that saying: ok, the technology doesn't exist, but we think ten years down the line, some of the time

now. If you will exist- and you know we have to lay the seeds now, because ten years down, the line would really want that malaria vaccine available to distribute to every person in Sub Saharan Africa thats where it takes, and so I think, a lot of what future perfect does is trying to say where's the world going, how the new emerging technologies both solve our problems and also agree our problems. In a case like climate, and stuff like that, and you know, try and get a head start on what the situation is. Gonna be ten years from now, instead of always being sort of reactive in our addressing global problems. Sudden Joe, you said, you said that you know errors. You then probably doesn't look like

These kind of science fiction stories, which you know is interesting because it is so it so deeply embedded actually in our culture. Ride like the very first robot story is about robots, staging a revolution, and you know taking over the world by catch him back and ass. He had a previous work, which was about genetically engineered super salamanders, who also do the the exam. Same thing. Additive is eventually get refine. So there's like clearly been this anxiety. Like long predating modern computers, like mankind, will create some kind of artificial mines that will turn against us and that we should be very, very concerned about this, and yet it is true that we have like blundered forward towards greater harvest.

Intelligence without actually in any way. Acting on this cultural concern, there were all sort of familiar with. I think, because it sounds so ridiculous, spread that, like ok, the chest playing machine is I now get inside

me or something I think, there's a disconnect where actual ay I so far mostly does narrow harmless things. You know it winds, video games, it writes reasonably good text. It makes deep fake videos of people singing songs, they didn't saying and there's disconnect words like how does that become a serious threat to civilization inside myself, that by saying, like, oh the robot wakes up and of the robot, it's us for whatever reason or the robot. Once the copper reasons humans want to conquer lie, that's like over anthropomorphized seeing eye. Nobody who works in the field expects that to happen. I think, actually, a better model for the reason I is dangerous is not what it may. I is a conqueror warlike. What if they I is like Philip Morris, the cigarette, what if it's just applied by a company that wants to make a lot of money, and it does that in a way that socially destructive and so actually want to step back and sort of expand on this. I think.

With every new technology, new powerful corporation new state barest forces that are pushing it to act in ways that are destructive for human flourishing destructive for the planet destructive, but the global order, whatever of and there are forces, acting to pull it towards being conducive to him flourishing conducive to the planet. Continuing to be you know at least intact enough. We can live on it, make money on it.

Conducive to there being some global order that produces peace and not nuclear war. So, like I start a business pumping led into the atmosphere making a ton of money, might businesses bound for the world because, like causes brightened at me, the world has various mechanisms that might not be stopped in my band lead it my tax light it might give the people who live near my factory threat to sue me for developmental blaze, I caused in their children. It might write, outraged, export aid in the press about me with your bad for PR investors downright my company by employees and shareholders board members have moral qualms about what I'm doing like. None of these are perfect mechanisms, but I do something bad and then we have various regulatory democratic, economic things that sort of pull us back in line, and I am I concede and stop pumping let in the atmosphere. I might a greater regulatory package that disadvantages me but disadvantages my competitors even more. I might of two states make it look like. I stop pumping let into the atmosphere, but I am still doing or I make it look like

that is great for children's developing brains by taking some studies, we saw the status with tobacco companies. They become lanes of people, they tried every tactic and booked get keep doing that, and eventually we brought them like a little bit more like right. So aid is a set of powerful new technologies to do things in the world, and many of the things that they do are not by default, going to be in line with human flourishing, not because they hate us not because there are trying to kill us not because they woke up in any sense, but they might invent products like tobacco, better, profitable and addictive, but kill people. They might maximize engagement on social media platforms at the expense of the functioning of our democratic society. They might invent brilliant new financial instruments to make their inventors

of money, but unexpectedly destroy the whole national economy a couple years down the line. Those are things that computer ADA decision process and are also doing right now, even though currently I is really limited, which I want to get its more later, but in this world with very limited, I we have computer, aided decision processes that are doing things at audits, human flesh and we have various ways to push back. You know the Assisi can send you to jail. The government could bring an anti trust against you, Well, well, protest! You and bad actors can do a lot of harm and it can take decades before mechanisms for keeping a Czech catch up, but fundamental. Our mechanisms for keeping them in Czech exist. There are forces that are bringing them in line with what humans want and the phenomena from a day? I is that we're gonna get way better at solving a wide variety of problems in new increase the ways using avenues. We haven't thought about and the solutions are gonna be too complicated, Freddy Human to understand a designing drugs and it's hard to guess the mechanism of action because their unlike any drugs we had

March, designing financial instruments and are so complicated, the finest people sure what their buying and selling they I design speed, campaigns for products and they work really well, but the product people sure where they work so well of you. So I think this naturally usage and sooner sue man a little bit why? What is going on now with with artificial intelligence which you know. It's always a little surprising to me right. So it's like, I have seen, there's been a big trend lately to these not like super convincing, deep fakes but like they take photographs of famous people and could have done more for even sing songs and it's like it's kind of need. Its answer were important, but what I would I think actually like I did not realise- is that the thing that makes that possible right. That is, artificial intelligence. That's that's! That's the tool that does that work, yes,

and there's a weird thing here where, like, when you talk about what I can do, I engineers are kind of like we really can't do that, much we spend all day like tweaking parameters to get like slight changes and performance. We can make pictures of them sing and stuff like that. So, on that sort of micro scale, like minimal progress, is being made maybe a week to week, but if you zoom out a little bit and look at the state of ai compared to like the state of AI,

I have ten years ago. It's it's an insane deference. It's an extraordinary efforts, like there's tons of things that ten years ago, people were saying. Maybe this is completely impossible- that lack now a grad student Hindu over a week. You know I remember when I was in college, which was a while ago now, twenty years ago, you know I was in a class and they were saying something about neurology and to the point they were making. Is that, like the brain, has a special area to detect other human faces so, like we see human faces really really easily, and the kind of like tossed off thing was that this is like an unsolvable, a problem, and it wasn't like a strong claim about a eyes. They were just observed that, like I could not do that and now totally came by. Yes, there has been a ton of progress in that in things like reading taxed and interpreting in things like generating taxed in things like watching us

singing and coming away with a summary of what happened in the scene. You just see a are able to do things that you'd know five ten or twenty years ago, people were say: maybe this is completely impossible, for I am sorry the issue is that computer programs sort of it has that that the compounding quality? were you can work for years and years and years making progress and seems extremely incremental. So it's like we ve been going thirty years on this and just now like, what's a face in what isn't a face, but then, once you get, there is like one more year: two it can scan your whole library of photos and tell you who your cousin sway and, like it accelerates, really really asked, and so you can be reading a book. That's only a few years old and its wildly outdated. Yes,

exactly right and there's also commercial dynamics that contribute to accelerating it like when you can't even get it to identify a human face in a picture no company in spending much of their budget on that, because, while Father and once it can do that, then billions of dollars and thousands of bright people are poured into can better figuring out what it can do next. So, as I makes progress, more resources get pointed at making that next step of progress So you see super non incremental growth and I soon becomes, I think, a difference from the tobacco case where the lead case, I'd, because an issue that were always dealing with is that you know people come up with some new idea right like I have this way to make the gasoline work better by by putting lead and right and their turn out to be a lot of problems with it.

An idea, but as terrible as like the story of lead was across twentieth century industrial history, it mostly got better over to read it? I will we developed a better understanding of the lead situation started regulating it started, regulating it more strictly more countries start using unleaded gasoline. We ve done some clean up because, like the lead, technology itself did not evolve faster than our capability to keep up with it. I think that is a completely key thing is: is the technology avoiding faster than the resources that people have Sort of understanding technology bring the technology, unlike in the case of twentieth century technologies like lead like tobacco, there was an invention, it had lots of harmful effects, but I feel basically option

stick as a society. We learned more, we acted on what we learned. We phased out the bad thing we made progress, whereas if the pace of innovation is fast enough, if we're building on new innovations, much much faster than our capacity as a society to bring them in line to learn about them to understand. What's going on and react to it, then I think we have problems, so that's a great place to take a break, and then I want to draw a sort of philosophical distinction here. Support for the show today comes from the economist intelligence unit or yeah you. For short, they help the business, financial and government sectors to globally strategizing, navigate and ever changing landscape, which is important to most of us too. Right. The award winning data analysis helps. Clients understand, world today and prepare for the challenges of tomorrow. Not all its clients, seize opportunities and way risk effectively, even governments. Why in the EU to support key policy decisions so to help your company understand the world and prosper within it? Visit ye? I? U dotcom, slash weeds, that's ye! I

dot com. Slash weeds support for this episode comes from America's leading beverage companies who are working together to reduce plastic waste in our environment. Not all plastic is the same. America's beverage companies are carefully designing. One hundred percent recyclable plastic bottles, including the cap's their bottles, are made to be remained and their investing in community recycling programmes to help get more bottles back, so they can be turned into materials used to make new bottles that completes the circle and reduces plastic waste. Please help get every bottle. Bag learn more at every bottle, back dot, Org, so part of what this points to re is the difference between fine. I harms like even really bad one straight, but did or still bound rhetoric as many people as died from tobacco. You're talking about lives, cut short,

but still people heavy smokers like grow to adulthood. They have children of their oh and plenty of people smoke heavily without dying, like I guess it's genuinely very bad, but like no society was ever like brought to its knees by people smoking. Right, like civilization, does not collapse. If risk of. I were the risk that I would invent the ex tobacco. You know that what would be some resources to address, but it wouldn't really merit the sort of very strong terms that a lot of people, but the eyes in which his like we are going to lose our civilization. If we don't get this right and the idea that there is a very limited list of things that pose that kind of genuine existential YE, I think that's true. There is a difference between,

like lead, that kill tons of people and make our civilization worsen in various complicated ways cause harm that were still on doing and things that just are not necessarily compatible with any kind of human future. Yes, so I mean among kind of normal people. Climate change seems like the thing that, if some he wants to name you like a source of existential dread right that that's like a pretty like Normie american liberal concern, and you re Emmy written about this for spread at and you'll be. Was it that's probably wrong on the you're talking about

a serious but bounded set of harms from our of warm or climb. Yes, that's my best estimate is that climate change is going to cause, probably hundreds of millions. That's that's an unimaginable tragedy. Obviously I think it is different. Then you know something that is incompatible with continued human civilization and I think most people instinctively agree. Those are different, like people who think that climate change is going to drive humanity extinct in your tends to be endorsing much stronger measures to deal with climate change than people who think you know it will be a bad, but not in that kind of way, and that seems right to me if I thought climate change is gonna drive, humanity, its think. That would be a favour if you don't doing what

It takes spanning planes and banning cars tomorrow. Maybe that causes ambassador upswing in poverty, but you know better to live in poverty than to not have the chance to exist at all where, since I think that climate change will kill a ton of people but be survivable for civilization, I tend to be more like ok. How do we minimize total human suffering here, which involves lots of climate mitigation, but not every possible climate mitigation that concept to higher costs in developing countries? Well, so it's it's a big problem, but a normal sort of balancing set of considerations, still is again a sort of puzzling area, because so, if you were gonna, tell me that speak all you know doodling away, machine learning to kind of make our photo apps work better and you don't get the auto folk S. Son staff- and you know we're transcribing- are calls fester did this is posing a potentially

existential threat to humanity. Should we just like like shut it all down like this, gains actually of machine learning and artificial intelligence will not zero. Obviously, like there's a reason, companies investing money in this, they seem almost a little trivial like what. Why are we just like rounding up the computer scientists you, don't put them in jail smashing the machines with our crowbar is like like. What's what's Goin on Cyprus, definitely under selling the benefits of potential machine learning, research like I think, climate change probably won't kill us all, but I think, as we had into the twenty first century and the acceleration of technology, what it can do? I think there's gonna be a lot of emerging technologies that significantly threaten our way of life and like our ability to continue as a civilization. So I,

I think you know pandemics can get a lot worse than this pandemic that we ve suffered through the last year, and I am not totally sure that the world could sustain having a pandemic like this every decade anyway, let alone ones that are much worse than that so basically not in favour, and I dont know anybody who is in favour of like trying to hug her down, maintain like our twenty twenty one tech level and addressed whenever future problems come our way with that and then the other thing is this: on a purely pragmatic level, if the, U S, government announced that a recent was illegally, I research would happen elsewhere. Lots and lots of people are doing research if there were some horde of total global agreement to slow down on on research research, I think then I would be in of of saying: let's do only the kinds of I research that help us understand how these computers work and help us like get results we want. And slow down on on research avenues that art back, but that global agreement it is completely unattainable, it's absolutely

and it happened. You alienate alot of researchers who are otherwise broadly on board, with trying to do safety and get this right appear like well. We could just banned the hoping and we can't to spend the whole thing I think that the answer is like cried magically. We kept them whole thing and trying to would mostly just make sure that it is happening somewhere farther away from where we can do understand ability, work to make the computers like do more of what we want and less things that we don't want, and then the other part of it is like. I think that we need to make things better than they are now in order to be a quick to overcome. The challenges of the next century makes me sad as usual, this more more good science fiction that scram straight. This is in in doing the Butler Irian Jaya hide eliminate Artificial intelligence, sinner nor romancer is deterring police who

stand, realistic right and especially when you ve got a multi planetary civilization like it doesn't take that much to do I, research and and over time it's gonna take a less and less in the way as computers get more powerful, it's cheaper to two areas, Dr, so you just can't have an extremely sophisticated civilization with good computers that doesnt work buddy research. I have that kind of coordination over planets. We don't have that kind of coordination between DC in California it it's so I went to bed. I think that the paradigm for a lot of this stuff you know serve implicitly, is late twentieth century like nuclear nonproliferation, yet which has proven to be very challenging, but more than a little effective, but the thing about nuclear weapons. Meant is that involves he's like very large macroscopic objects that can be easily observed. Yes, you can.

Look through a satellite and be like hey that looks like a nuclear weapons programme. What are you doing? You know, and you can't as much do that with or now ok right now, building state of the art machine learning Systems requires a lot of resources. Much less than building a nuclear bomb, but still alive. Resources in a lot of computing power that only a few places have access but that's one of those things that is rapidly changing right, like computing power, is getting much cheaper. Lots of people are buying extremely high and computers. You know, if only to mine, crypto currency and so more, More people personally have the resources to do. I work. That's just gonna get more and more true overtime when computers obviously have been getting more sophisticated for a while now that continue, as they are relatively small, easy to conceal You can build after every. You can just have a room full of computer equipment in a basement somewhere like even a lie, and nobody would

oh, so so on an international perspective right, because I do think this important right. If one country wanted to institute some kind of very sharp limits on Digital Technology programme. There would be no no good way to assure that. It's not happening in China happening in wet wherever out straight, and that is important, because obviously one of the things you can do is like forgone weapons were done. Military applications it's not all singing FED chairs. Yes and the military is very concerned with this problem, and I think there is a recent National Security Council report on a high in the task they sort of calm down on is we want this to happen in America, not China, and so we should be

keeping immigration policy in shaping our technology investment policy in order to make sure that the first day I happens in the USA, China- and you know the chinese government- is very terrible. The Chinese Communist Party is awful. I don't want them to have eighty I, but I think, if you are thinking of the problem is just like. We need it to happen here and not in China. You're really under rating the scale of the problem, because if it happens here and we screw it up, that doesn't say bus ride, gizzard, that's the sort of the Vienna see report view of it is just kind of life Well, if we're making like a better plain engine and why

then why we want the best plan engine like the Europe that the concern is that the point of really advanced day. I is to do things that people could not do, and that means that the creators wonder certainly understand like what the computers doing for the yes, and even if we ask the computer for it, a computer could be really good at coming up with new drugs and not really good at explaining in human terms like what all of the effects that new jobs will have be good at explaining all the effects that jobs will have, but not adequately. Incentivize should do that like producing, and accurate report on the drugs is just not one of the things that is designed to do, even if it would be capable. So what what's the application to drugged, because this you you mentioned mentioned a couple times, Turkey,

Sir, I've talked to researchers were sort of Gung HO on this transformational. I Development Programme, one of the big things they say is like what, if we can end cancer, what have we can end aging? What have we can just and involuntary death? obviously we're getting into some pretty ambitious stuff there, but, like human medicine, has come a really long way and if you can supplement that with super human medicine, with tools spurred drug development and invention that we have come up with yet that super grid, and would somebody also take this and be like develop a better cigarette? Yes, absolutely summit will do, but I'm not stupid

equally worried about the better cigarette, and I think drug development is something that we want. A superhuman effort pointed out is this: what protein folding his for protein folding his project about Mitya, so for decades, pharmaceutical companies have been trying to figure out how to predict from a protein slight line of instructions, how it will fall to us. The super hard problem, and last year Google's deep mind day. I department blue everybody out of the water and won the competition by a mile, Withan, hey, I that can predict protein folding. This is gonna, be serious. Offer drug development by itself, you know- that's not going to change

Think as there is a lot of other steps, the drug development process, but it doesnt seem implausible to me that you know at some point. We had a nice that can spit out of a line of dna and say like print this, and it will do you no cure that person's cancer did. This is one of these six. Why? I was feeling I can maybe fake my way through it, but I don't sincerely understand what the subject is. I know that there is this protein folding problem and that its hard and that there was a breakthrough in it and that there was a lot of excitement about that. Is it like what? Why is this so hard like? What is the well? What is the nature of the issue, so you trying to think of a good, so you got that the dna right and it's just a see tee he whatever some sort of strength and with proteins. I think you ve got other sets of instructions, and then you can figure out. Ok, if we have this year,

encodes codes for this series of amino acids are like little protein blocks and you would think that would be the whole problem. Okay, now we know all of the amino acids. We know how they line up, but actually the way that they are, acids will blob together into a protein, it's kind of like okay, you know if you fold a piece of paper, there's like a million ways, you crumble in your hand, there's like a million ways. It can crumble in your hand, because every little ridge of it can crumble in like a hundred different ways. And for amino acids for every sequence? There is a specific way. It will crumble, but its very very hard to look at it and just guess what is the way it will crumble, because it involves the interactions of like two hundred, frank things: they'll have electrons that are all pulling on each other in different ways, so you look at that sequence and then, when hop researchers or before need my top programmes tried to guess how will crumble they're, usually wildly and with you. Did was stayed through. I dare say through a lot more resources. At this end, they can with reasonable reliability, predict how it will crumble.

This is really important because how it crumbles determines which bits of it are on the outside, which means which bits of it you can maybe bind to with receptors. Are things like that and it predicts lake which pieces of it are going to be the most functional ones, versus weapons are kind of buried and I'm not going to affect the overall function, so you ve got to figure this out. There is you know if you do occur, The Turks coach there's no trillions Italians possibilities. An eye is reasonably good. Now at saying: oh it'll come politeness and and surrender its basically, it's like the the physical shape of the proteins, is critical to what they actually do exactly, and this is a not as funny as some of the things that go on with images, but like has a really powerful sort of real world application, and then it

the question becomes like what do we do if we are concerned about this are going badly act as way now it's like I don't know you can have conferences are people could do takes, but it's really challenging to regulate like happens on people's computers, especially because the regulators themselves are probably not going to be at the cutting edge of technology. So I think one thing work, for us Here- is that nobody actually wants to do nay. I system that like proceed, something contrary to human values and makes everything way worse, for nobody wants that. So if we can develop tools that make it easy to characterize what I was doing, even when it super intelligent or things that make I you know, differentially super intelligent like you, you can work ports are today, I that will be the excited about, and parts of it

That allows for pushing back, allow for correction, allow for keeping up with the pace of eight changes to the world and addressing destructive in destabilize excited. So I think from a research perspective, there is like lots of promising making a more understandable making I easier for us to like interpreting, respond to work that needs to get done from a policy side, it's hard, because I know you're a policy person, and I, like most people in the government like what we think I'm serious, what we do about it I've I think there should be more focus. I think there should be a focus on in the regulators up to speed in this field and, like hiring top tell it, even if that means you have to pay a lot of money for it, so that you have people thinking about. I who really

We get a high and understand recent developments, but I dont think the: U S should try something like a massive ban on this technology can fund promising research. It can actively try to discourage research. That's just aiming at eighty. I without a clear plan for interpret ability, understand ability, but a lot of that in order for it to go right, the people making the rules have to really get and right now I don't think there are very few people who really get it. So I think you need to start with that. You need to start with more people in government who are deeply

often the process of making and get what's going on there. What's he gonna break? It reveals a governor of hours in the day to get everything done. It might be because you're missing out on three of them. Where does the hours girl they probably fell into a deep dark? A bits opens up when we switch between work. Apps, add those three hours to all the protecting family, miss out thanks to at home, distractions, disorganisation fatigue. It's no wonder the days feel to shop work should work and with click up it does put up as a flexible productivity platform. That brings all your work into one place. That's all you're, chats, apps docks and tasks, one central S Place Mission control companies like Hoover and Google use cook up to

their days, more proactive, managed projects, people and calls for effectively both for teams of all sizes and industries. Hookups pleasingly fast features of one thousand plus integrations became must have for anyone wanting to track manage to tackle their work in one place. Cancer hours back with clear up trade for free. Today had cleared up the palm slash, the weeds, if you're, a gig, worker or self Floyd there's some good news about PPP loans, you might want to consider millions of self employed workers may qualify for up to fifty thousand dollars in one hundred percent forgivable alone, You might be one of those millions as the leader, PPP loans wobbly can help you find out. They ve helped over three hundred thousand small, businesses across Amerika. Get a ppp loan. Funds are limited, so apply now wildly dot com, slash, vocs and see if you qualify for a ppp loan That's w! Oh, I m p l why dot com slash veo Ex Wobbly,

not a lender terms and programme rules apply. I think one of the most basic sets of concerns that people have about artificial intelligence and have had a long is a bow to impact on on the labour market and on you know, displacing people from jobs, You know, like detects text, generating a that exists. Now they could not do my job but they're pretty good. And their way better than their precursors were a very very very short time ago? And it's not that hard for me to imagine just being able to write up news stories. You know that really soon it if you could create the right inputs and things like that- and you know I mean,

I don't know like most people, I don't wanna be out of work and a z might might mass specialise skills thrown away Erika producer. He was telling me about some possible new. I audio editing tools that people are trying to do you. And like it, if it all seems kind of kind of alarming, it will be greater cure cancer, but not to just like replace everything that everyone does yeah, and I think this is again a thing about. Peace like a guy is going to change a lot of how the world works, and if it's doing that gradually, then you know we can adapt and maybe figure out like a way for that transition to end up being good for humans. Almost every previous technological advance hasn't. Let

mass unemployment, its light to, like you, know, people finding new things to do in a higher productivity society. I certainly I want a idle work that way too. I want, or maybe I want, mass unemployment, because we have ended scarcity and cannot spend our time. You know, I don't like mass unemployment in the kind of society have now certainly you get things are slow, and I we have lots of mechanism, casino, democratic and commercial, and so for this sort of make things mostly good, and the problem is, if things happen too fast to sort of make things mostly by an so if they I is happening too fast, then I don't think we have any good mechanisms to make things mostly good aside. For me, I alignment research, which is very much in its infancy and the people doing at are not totally sure will solve its own problems, but at least flake. I think we should be funding more of it and I certainly hope it does so thats what I wonder ran immediately. You know we ve taught you couple

how about some problems with trying to do like hard bans on any I development, but its sounds from a lot of the sort of tour of issues here like if there were. Some way to throw sanded the gears and slow things down, that that might be good which is contrary to how I normally think right like it seems like there's a lot of areas where, if we could make progress happen more quickly, like tat would be really good. That's because I'm always assuming you're if we're talking about developing a vaccine in a month rather than a year, so like on a time frame that human beings can comprehend verses. You're talking about Wall Ito, yet you you don't. I guess we'll have progress at such a runaway pace,

it's like, we can't even see what's happening or consider a response to it, and I I struggle with this, because I am very much in a techno optimist. I very much a person who well like everybody about how economic growth is the best thing that has ever happened for people. I do believe that, and I We believe in the potential of aid to continue that story instead of just tragically ending it. But to be honest with you, if I could trust ended the gears what if things were happening slower. I think it would have a better chance of going well for humans. I think that we need the mechanisms for making things good to be faster than the mechanisms for making things power and right now they had used the phrase airline into early when what is needed one if you please So I alignment is research into how to build a eyes dead, pursue our goals and communicate about their goals behind you know improve themselves

whatever else say, I've made might do in a way that is compatible with human flourishing. It like promotes the goals of people, that's very broad, it A lot of I research is specifically about understanding. The exact capabilities of exist in systems which is a surprisingly heart problem right. You develop like you, did he three is an eye that produces text. It can right. You know, okay articles? It can right, ok, product reviews, awe and it can do math, sometimes if its prompted the right way, but not if it did other ways the lot of HIV alignment. Researchers like what are actually DVD three is capabilities, because all it does is it generates text that it thinks is the appropriate response to it just ask them, and that makes it hard to figure out. What does it actually know? Does it know something if it can prove That is a response, sometimes so what'd. I alignment recesses about understanding what s know. No an understanding with their capabilities are

then a lot of it is about getting a eyes to report. To us on their capabilities in their uncertainties and then may I make research is about trying to build. Humble eyes like a eyes that believe that we know more within them about our ultimate goals and priorities, and therefore They should be motivated to cooperate with us. Laid thanks to us and do things at our pace, because they believed that we know more than he's a hard thing to make any. I believe, especially if you think that eventually we're gonna try and build this into an I. That is much much smarter than us, but those are all things that Voluntary island research. What what does that mean smarter than us? Ok, yeah! That's it! if our question intelligences lots of things, but if you think about intelligence has lagged overall cap city to like look at the world making princes and then form plans that work. Ah, then, like clearly people,

Are smarter than animals? When we look at the world, we figure out more about the world than the animals do an hour plans in the world work better. And you could similarly imagine something that, like looking at the world, makes better inferences than us. It figures out war about what's going on and its plans work better than ours and I think that's like be useful definition. Superintend it's like things that can understand war by looking at the world and things his plans work better than earth. So I mean I feel like in a sort of domain specific way, that's easy for me to understand why you like that. You know right now there are self driving cars and that technology exists, but also there are people driving cars around. There are also a lot of car accidents and, like theoretically you

make cars the drive themselves without does accidents? Does it be super intelligent car drivers compared Day Human Cartwright, Russia? That would be great, but it doesn't exist, but at the same time like my computer can multiply high numbers much faster than I can. So I'm like a better, a driving that my computer, but I'm worse and multiplying, share, but what use is really a totally general eyes So I think it's kind of a generalised, like I think fit compared to my four year old. I am better at making inferences across her really wide range of areas, an better forming plans that actually work across it really wide range of areas, and we take my four year old versus a cat for old. Is better at forming inferences and better at making plans and like the cat, is better at sea things, the count as they better walking on fences and the cat is better at light, falling and landing on its feet. But I still feel like

There is a meaningful says in which I can say: ok, but the cutest smarter than the cat, and I'm smarter than the kid. For now she can be the chest. So's will see how the slots right, but beat me a chess, and yet I think it still makes sense to say you know I can form a lot more but more of a picture of the world for reading about it, just through additional experience, and my plans tend to go better because I dont like burst: open ice Paxton and eat the interiors like she did yesterday. So I think there is a general. City here I think. Obviously there are narrow versions of it and we might want to build their versions of it like it might be that the smart thing to do is to only Doug narrow eyes, only build eyes that are super intelligent at drug invention and not super intelligent anything else. If you could do that, that would be cool, but I do think that it makes sense talk about a sort of general capacity aware You look at stock market data and you do better than anyone else at predicting how the stock market will do and also you look at a presidential debate.

You do better than anyone else at predicting which candidate will be more popular and also you look at a marketing campaign. You do better than everyone else at predicting how that marketing campaign will persuade potential producers of your product. You can imagine the capacity that helps it with all of us thinks I'm in this is interesting. I think a lot of people of resistant, not just like the artificial intelligence case, but in the general case to the idea that sort of general competencies exist even on genetic, reducing clearly, but children right are adorable and waterways, almost any kind of task like a six rubber, outperform

for you I'll ride like this. No, yes, they are better looking at the world and making plants that work, and you know I understand that there is a lot of reluctance. You know given the history of sort of intelligence in lake various awful social engineering programmes. I understand that there are some hesitation there, but six euros are better at making. Plans for girls and grown ups are better at making plans and six year olds like does it just facts about the world? Rightly, we ve gotta be able to talk about general capacity. Make plans for anything that makes sense. Ok, so let's say you know somebody lessons in his pipe cast their convinced there more concerned about risk their adding to their list along with pandemics, and whenever asked the boy, what should they do? Remedy the other good that we were talking about malaria nets right and the great thing about bad is like you know, you can just like right. A check to the net people. Vaccine research is hard but like

Some people have researched lots of vaccines, their existing funding streams for that there are ways to do so viewed vaccines to people. So you know you: can you can call your congressmen and tell them they should put more money into the global fund or INA whatever else, but why should we do yes, I think, if you workin, I that's easy. You know Paul how Cristiano diameter do great I'll wine that type work you should get familiar with what they're doing in there see where, if interference, I researcher, where you want to build on it, but like most of us, are a researcher spend. What ways it's like vaccine, research, where some people are gonna, be specialists and figure out how to do it and then Nobody else can mostly help by, like you know, having a good enough understanding of the area that the government is incentive, have a good understanding of the area and to pursue policies that are good like a lot of our problem with vaccines. Right now, you ve written about how there's not a lot of demand among people for better vaccine policies. We don't have better vaccine policy, may think. There's price index.

We're going on with a right. We don't have a lot of demand for better ray. I policy so we only have good I policy to the extent, like the defence Department, considers it Annapolis security issue and they take it really seriously. But I dont want all this work to happen. A defence department. I think that creates its own problems so what more people can do is like look out candidates, a platform is: ask those candidates, lake What do you see as the role of the government in a line of work create an environment where people feel like they need to understand? I'm. You know well enough to answer told you Questions maybe that environment higher smarter people at the decision making a first baby it doesn't. I don't know, I think policy is definitely a hard labor for something like this, such as policy reviewed site. Maybe this naive, but I think it is very easy to understand from a distance like who were the people working on developing vaccines,

versus who are the people making things worse spread liquids would seem like some weird about the air I feel to me. Is that, like I would see nobody's gonna, say: oh I'm doing reckless research right like people, people believe that what they're doing is helpful or a lie about it and since its like the key for I research, is more. I research it tried to tell he's doing good wives, I'd say it's hard for me to even tell whose side I write like what like like what it like. What am I saying here yet? No, I think that's what you ve a really big problem even among people who take alignment really seriously and agree that we need to solve it. Some of them think that others, a firmer, are being too reckless and pushing forward a little too fast. So There is no anywhere a list of people who are doing a good job, and then there is the fact that

think I alignment research, which is a very specific thing about this fast pace, future technology and how to make a deal with things. Get sort of combined in a lot of me, was minds with bike ethics obey I research, which is more about like shit. We help the chinese government build surveillance systems, should it can use to do genocide it. We should not do that, but it's not really an eye alignment issue, its just like a basic ethics. A few that day, I component is really very important and then, like algorithmic bias in hues. Which are fascinating, unlike their whole, on topic of discussion. But you know because we felt it a bunch of racist stated. This criminal justice system is racist is like again a pretty different category of issue from a whining future super intelligent systems and so a lot of places will say: oh we're doing AI safety research they're, like oh, tell me about your ai safety, research and they're, doing great work on algorithmic bias or on not selling China tools that can use to do genocides and I'm glad they're doing that. But

doesn't make it hard to figure out who is doing alignment racing wait. Maybe because again I mean these sort of finite scale. Problems right, like you, don't you you might not want to sell the chinese government paper right there that they're using in concentration camps right, I mean any like horrific human rights abuse is made up of a bunch of stuff, right and lights for general reasons like you, you might want to sink. And entities that are doing that you don't wanna, be complicit. You don't wanna have labour in your supply chains for your company is unlike air. I could be part of that story, but just in the way than anything else would be Exactly it's nothing specific to eye, and I think a lot of attention ends up there, because it's more specific its market, creates more abounded, it's easy to say: oh yeah, we specifically took these measures to ensure we work complicity, any human

to be assessed. We're done now we're not implicit in any human rights abuses, whereas with alignment research there sort of an endless like. Are we making progress on the women? Are we making part an alignment faster than we're making progress on capabilities because a lot of great research organizations I felt like the answer- is yeah you're, making progress in alignment you're, making faster progress on capabilities, it's much harder, and so I think it's easy to sort of try and solve a different problem right. Well, especially on Friday, I mean we were talking about good guys and bad guys right and it's. I don't want to say easy, but relatively easy to say that, like okay, like tat the the piercing government, the date they are bad, does strike, and so then, ok, you don't want to help them right, but that's different from is the research programme beneficial or not, right or or just neutral, and an that's like that? More challenging issue right is like what high Whitworth avenues of research

are actually helpful, and I think I can answer that in terms of specific avenues of research. I can say I think, research into measuring the capabilities, our systems, research, into making our systems report out what they're doing more carefully research into understanding our systems, that's good, but, like most labs, do some of that sum. Some capabilities, research and many researchers are doing both in a lot of researchers who do care about alignment field, but some of the capabilities research is necessary in order that have the kind of systems they can like. Ok, jpg, three that the tax day I we ve been talking about that capabilities research. It is an area that has new capabilities, but we also learned a ton from it that people are applying to their away. So you could see this reminds me a little bit of the light.

We have to make the super virus in order to understand how to prevent pandemics, yeah and, as I say that I am like. Oh man, it via security. I have no sympathy for that and unlike stop building the silver virus- and Maybe- and I start building tomorrow- a eyes, but there is certainly a lot of awareness that it build this murder. I somebody else will build this Marty. I, and there is also just you- know- the people working on this stuff love, making there s, murder and seeing what they. Do I need some somebody at the wheel, Han Institute of Iraqi was sang, but we don't mutate the bad virus looking to do it in four Dietrich. Yes, that's up when you can't, let that be it escapes right, like you. Just shouldn't. Do things that run the potential to make the world vastly works, even if you think that you're, a girl, ideological opponents are doing it or wherever, but that's a very hard call to make in its very specific to an individual situation. I don't know I'm against gate a fucking research and biology,

I guess to be consistent. I should probably also be against it. May I briefly something I struggle is also something that I do the web is that this is a sub species of the general issue. In intense zero sum. International competition leads to dangerous. Policy choices. It is right that, like when everybody gets into, if we don't do it, the Chinese will mode about everything it. It's not even a sorry, that's wrong as an analysis.

The situation, but like you're out of good option once that's what you're telling yourself, whether it's about by a weapons or I or nuclear weapon it. You know anything else right that, like it's important, I think to actually try to have at least the possibility of international cooperation on big issue is because otherwise the logic of better here than there can can take you really far yeah, and I think you know, with nuclear weapons we we were certainly in a situation where it was better for us to build them than any of the people we were at with We ve been better for the world. If we had somehow stayed out of a situation where we are, we ended up having a nuclear arms programme during World WAR. I definitely if there's any,

I could do to sort of cool international tensions. I think that would help with eye. But you know: cool international tensions isn't exactly a thing where you can point to the people who are doing that? Yes, it's hard. While we will now have to think think more about it Hopefully somebody out there listening, has some better ideas for policy going forward. I know that's a lot of the audience here. Michel have Ado trade. Do try to trying to bring ideas to the people. I'm so thank you so much Kelsey Eggs is always two responses are producer. Economic is, unusually back on Tuesday.