# Anthropic CEO : AGI is Closer Than You Think! (machines of loving grace)

## Метаданные

- **Канал:** TheAIGRID
- **YouTube:** https://www.youtube.com/watch?v=O8GQyncW7GY
- **Дата:** 13.10.2024
- **Длительность:** 1:01:17
- **Просмотры:** 56,420
- **Источник:** https://ekstraktznaniy.ru/video/14004

## Описание

Prepare for AGI with me - https://www.skool.com/postagiprepardness 
🐤 Follow Me on Twitter https://twitter.com/TheAiGrid
🌐 Checkout My website - https://theaigrid.com/

Links From Todays Video:
https://darioamodei.com/machines-of-loving-grace

0:00:00 - Five Key Areas of Focus
0:02:23 - Basic Assumptions: Powerful AI
0:02:52 - Defining Powerful AI
0:04:36 - Capabilities of Powerful AI
0:05:14 - Agentic Behavior & Physical Limitations
0:07:23 - Millions of AI Instances
0:08:37 - "A Country of Geniuses in a Data Center"
0:11:29 - The Singularity vs. Real-World Limits
0:11:48 - Marginal Returns to Intelligence
0:13:29 - Factors Limiting Intelligence
0:14:47 - Need for Data & Intrinsic Complexity
0:15:47 - Constraints from Humans & Physical Laws
0:17:41 - Timescales & Malleability
0:20:02

1. Biology & Health
0:21:09 - Potential for Improving Human Life
0:21:53 - Limiting Factors in Biology
0:22:18 - Experiments & Data Quality
0:23:17 - Historical Examples of Misinformation
0:24:50 - Intri

## Транскрипт

### Five Key Areas of Focus []

so Dario amod the CEO of anthropic recently released a huge essay in which he titles it Machines of Loving Grace how AI could transform the world for the better so essentially this is basically talking about how AGI super intelligence could transform the next 100 years and in this it's pretty insightful because it doesn't focus on the majority of things that other individuals are focusing on like the risks not that anthropic is not focusing on safety but it purely seeks to look at how good can the future get if we manage to get super intelligent AI right along with artificial general intelligence so in this video I'll be diving into absolutely everything that this talks about as well as looking at the future with as to what the predictions are for the future onwards like the 2030s 2040s and any subsequent years after that so you can see here that he starts out this essay by stating that in this essay I try to sketch out what the upside might look like with a powerful AI if everything goes right of course no one okay and this is absolutely no one even those at these top Frontier Labs can know the future with any certainty or precision and the effects of powerful AI are likely to even be more unpredictable than past and previous technological changes so if all of this is unavoidably going to consist of guesses so he's basically facing this with sating that look these are his best guesses but considering the fact that this is fundamentally new technology an entire new paradigm the accuracy could vary slightly he also says that I'm aiming for at least educated and useful guesses which will capture the flavor of what will happen even if most details end up being wrong and I'm including and he also says that I'm including lots of details mainly because I think a concrete Vision does more to Advanced discussion than a highly hedged abstract one so this is where he actually discusses exactly what's going on and there are five key things that this video is going to be broken down into that we will discuss so essentially we can see it says the list of positive applications of AI is extremely long and some of them he won't even cover in this essay this includes of course robotics manufacturing energy and much more but

### Basic Assumptions: Powerful AI [2:23]

he's going to focus on the small number of areas that seem to me have the greatest potential to directly improve the quality of your human life and there are five categories that he's most excited about which is first biology and physical health second neuroscience and mental health third Economic Development and poverty fourth being the peace and governance five being the work and meaning so all five of these are going to be covered in this video so that's where this thing is headed now some of

### Defining Powerful AI [2:52]

the basic assumptions for this framework because like I said before this is based on the fact that AI is going to be in increasingly powerful but he says to make this whole essay more precise and grounded it's helpful to specify clearly what we mean by powerful Ai and this is you know for example the Threshold at which the 5 to 10e clock starts counting as well as laying out a framework for thinking about the effects of AI once it's present so Dario amade the CEO anthropic CEO of Claud 3. 5 Sonic um company behind clae 3. 5 son it says that you know what powerful AI will look like and he actually dislikes the term AGI or when it will arrive is a huge Topic in itself I've discussed publicly and could write a completely separate essay on I will at some point and obviously many people are skeptical that powerful AI will be built soon and some are skeptical that it will be built at all but here's one of the biggest things that he recently said is that he thinks that it could come as early as 2026 though there are ways that it could take much longer but for the purpose of this essay I'd like to put these issues aside and assume it will come reasonably soon so basically from this he's basically stating that powerful ai that can start to directly impact your life in a very positive way could come as early as 2026 now as of the date that I'm recording and potentially publishing this video it is the 12th of October 2024 which means that is just around 2 years which isn't that long if you think about it stating that look maybe not God level AI but powerful AI is going to be coming rather soon and he explains exactly what he means by powerful AI it's going to be coming you

### Capabilities of Powerful AI [4:36]

know in two years is quite surprising and basically this is focusing on the 5 to 10 years after that and right now he's going to get into exactly what such a system will look like what its capabilities are and how it interacts even though there is room for disagreement on this so by powerful AI I have in mind an AI model similar to llms in today's form though it might be based on a different architecture and it might involve several interacting models and it might be trained differently with the following property so here's where you want to pay attention because this is where they're giving the first kind of predictions to how AGI is and of course

### Agentic Behavior & Physical Limitations [5:14]

he dislikes the time AI but I'm still going to use that in this video because I think it's you know a pretty useful term for quickly describing how AI is going to be in terms of pure intelligence which is essentially how smart the model is it is smarter than a nobble prize winner across the most relevant Fields biology programming math engineering writing Etc this means it can prove unsolved mathematical theorems write extremely good novels and write difficult code bases from stract Etc that is insane I'm not going to lie to you guys that is absolutely insane stating that in terms of pure intelligence in 2026 we could have an AI that is smarter than a Nobel Prize winner across the most relevant Fields okay that's uh pretty crazy and an AI that could prove unsolved mathematical theorems um the implications for that outstanding which is of course what we're going to get into later but I just think that is pretty crazy now quickly I do want to say that even if this doesn't happen in 2026 this is still an AI model that could happen in the future but I think it's important to understand the timelines for which individuals are discussing this stuff so it also says that in addition to just being a smart thing that you talk to apologies for that it has all the interfaces available to a human working virtually including text audio video and keyboard control and of course internet access it can engage in any actions any Communications or remote operations enabled by this interface including taking actions on the Internet or taking or giving direction to humans ordering materials directing experiments watching videos making videos and so on it does all of these tasks again with a skill exceeding the most capable humans in the world that is I'm going to say again okay it does all of these tasks again with a skill exceeding most of the most capable humans in the world which is pretty insane when you actually start to think about the implications for this so we're going to have an AI that is going to be able to use all of these modalities but exceed some of the most capable humans in the world now it also says here that it does not just passively answer questions instead it can be given tasks that take hours days or even weeks to complete and then goes off and does

### Millions of AI Instances [7:23]

those tasks autonomously in a way that a smart employee would ask asking for clarification as necessary so this is where we get into the agentic behavior where we see that in 2026 onwards that these models are going to be you know doing tasks and the thing about this that most people might not understand is that some of these models whilst yes they're going to be smaller than a Noble Prize winner which is crazy to think this doesn't necessarily mean that these models are going to be able to instantly respond to exactly what you're doing because there are physical limitations such as time and you know the physical world is bound by of course physical constraints like the laws of physics which essentially means that even if we do have really smart AI things can't happen in a split second as some may presume so some tasks might take days or weeks to complete as certain experiments are conducted and as data is fed back into the system it also includes that it doesn't have a physical embodiment other than living on a computer screen but it can control existing physical tools robots or laboratory equipment through a computer and in theory it could even design robots or equipment for itself to use which is pretty incredible I mean imagine a AI designing robots for itself to use that that's pretty I mean the implications for that are just outstanding then it says here that the

### "A Country of Geniuses in a Data Center" [8:37]

resources used to train the model can be repurposed to run millions of instances of this model and if this matches the projected cluster sizes by 2077 and the model can absorb information and generate actions of roughly 10 to 100 times human speed of course this is what I was already saying before it might be limited by the physical response time of the physical world or the software interacts with so yeah this is some pretty insane stuff and I think the crazy thing about this is that like when we actually start to see how autonomous agents are working I think it's going to be the biggest mind warp that most people can't wrap their head around because right now you're seeing me you know move around the page okay things are going to get really weird I'm going to move around the page but you can see me move around the page you know let's say if you wanted me to highlight resources I could highlight that I could highlight this text but an AI model okay as long as the internet is working and you know the computer isn't laggy and the limits of um any software are being maintained and AI is going to be able to highlight every single piece of word summarize things in a split seconds okay of course there are certain physical limitations which it does talk about but this is going to be 10 times the human speed I mean sometimes you're tired sometimes you're fatigued sometimes humans get distracted imagine an AI who has one goal one Focus that is working that quickly on a single problem you can imagine how quickly that is going to be and can't imagine that like some websites and services are going to be AI agent friendly I think some are going to be completely not AI agent friendly because I think when we have you know Bots and stuff that already exist like there are already Bots out there that you know for example if you're trying to buy something sometimes you'll see that this product is completely sold out the moment everything goes on because Bots can instantly buy and sell those things but I think when we start to think about Bots that can do that but with real intelligence real problem solving and moving forward on the goals that could actually benefit Humanity we can start to realize why things are going to be absolutely incredible and of course when we have millions of instances of this AI doing the same thing it's going to be absolutely insane so it says that each of these millions of copies can act independently on unrelated tasks or if needed can all work together in the same way that humans would collaborate perhaps with different subpopulations fine-tuned especially to be good at particular tasks so this is what he's stating that in 2026 these kind of models that we're going to have they're going to be um you know not like General models for example you know with the 01 models though those models are reasoners but I think even within those reasoners we're probably going to have a model that's just focused on math focusing on you know reasoning for AI research one that's just focusing on conducting experiments and it's fine to just be specific at those tasks and this is essentially like what opening had as their goal which is where they say that you know they want to get to the final goal which is of course organizations where you're going to be able to essentially have organizations that are just completely air run so he says here we could summarize this as a

### The Singularity vs. Real-World Limits [11:29]

country of geniuses in a data center clearly such an entity which is an interesting way to describe it would be capable of solving very difficult problems very fast okay which is crazy but it is not trivial to figure out how fast two extreme positions both seem false to me first you might think that

### Marginal Returns to Intelligence [11:48]

the world would be instantly transformed on the scale of seconds quite like the singularity which is where people think that the AI just builds upon itself and on upon itself and just continues Contin TOS until it's so smart that it's able to just become a god basically um but of course he talks about the problem with this is that are real physical and practical limits for example around building Hardware um or conducting biological experiments even a new country of geniuses would hit up against these limits and of course intelligence might be powerful but it isn't Magic fairy dust like for example let's say it managed to you know even solve cancer let's say for example it would you know take a while to you know um you know get up the factories to build the drug to distribute the drug there are still physical limitations just based on you know the physical laws of reality like there is a speed limit the speed limit is the speed of light so there are those things like there is literally physical limits which is by even if we have rapid technical logical advancement it's not going to happen like all within a day as some of these sci-fi novels may suggest and it says here that secondly you might believe that um technological progress is saturated or rate limited by real world data or social factors and that better than human intelligence will add very little this also the flip side seems implausible to me as there are hundreds or even thousands of scientific or social problems where a large group of really smart people would drastically speed up progress especially if they aren't limited to analysis and can make things happen in the real world which are postulated country of geniuses can included being directed by or assisting teams of humans and it says I think that the truth is likely to be some messy mixture of these two extreme pictures something that varies by task in the field and is very subtle in the details

### Factors Limiting Intelligence [13:29]

and I believe we need new Frameworks to think about these details in a productive way so of course the economists often talk about factors of production things like labor land and capital and the phrase marginal returns to labor land and capital captures the idea that in a given situation a given Factor may or may not be the limiting one for example an Air Force needs both planes and pilots and hiring more pilots doesn't help much if you're out of planes and I believe that in the AI age we should be talking about the marginal returns to intelligence and trying to figure out what other factors are that compliment to intelligence and that becoming limiting factors when intelligence is very high and we're also not used to thinking in this way and of course we need to ask how much does being smart to help with this task and on what time scale but it seems like the right way to conceptualize the world with a very powerful AI basically what he's saying here is that like you know that the plane example is a really good example he's basically saying that look let's say you were a country that wanted to go to war but you only have three planes but you've trained a million Pilots it doesn't really matter if you have the world's best fighter pilots if you only have three planes so it's like you know once we have unlimited Geniuses or maybe not unlimited cuz you know physical limitations of data centers but once we have thousands and millions of agents that are really smart what's going to be the actual physical bottleneck that we cannot solve which is I think it's a really important question that I genuinely didn't even think about

### Need for Data & Intrinsic Complexity [14:47]

of course you can see right here um these are the factors which limit intelligence so it says intelligent areas need to operate interactively in the world in order to accomplish things and also to learn but the world only moves so far to cells and animals run at a fixed speed so experiments on them take a certain amount of time which may be irreducible the same is true of Hardware Material Science anything involving communicating with people and even our existing software infrastructure furthermore in science many experiments are often needed in sequence learning from or building on from the last and all of this means that with the speed at which a major project for example developing a cancer cure can be completed how an irreducible minimum that cannot be decreased further even as intelligence continues to rise which basically means that as the future yes things are going to move faster however there is certain limits to systems with how things move which means that even if you have everything moving at 100% there is a base rate at which things move out which cannot be improved upon which is

### Constraints from Humans & Physical Laws [15:47]

why when you think about the speed of these things happening you have to understand that like things are going to happen but they're not going to happen as quickly as you might think but they're also not going to happen as slowly as you think and they will happen faster than you think which is why he said it's going to be um quite a balance now of course he said the need for data sometimes raw data is lacking in and its absence more intelligence does not help so today's particle physicists are very ingenious and have developed a wide range of theories but they lack the data to choose between them because the particle accelerator data is so limited and it's not clear what they would do drastically better if they were super intelligent other than perhaps speeding up the construction of a b bigger accelerator so yeah this is a crazy point because when we think about you know the particle accelerator and the kinds of data that's there there's just not enough data to go around because it's like you know um even if we had the Geniuses to develop let's say they developed the most incredible theories like let's say we had for example a million particle accelerator theories we someone developed ASI and we had a million theories to test the problem is that the particle accelerator is so expensive it's qualy to build like you have people that are trying to all test their theories and there's the physical limits and slots and stuff so it's not as easy even if we have a lot more intelligence okay and that's something that you know wouldn't help the only thing they could really help with is of course speeding up the construction of a bigger accelerator to you know conduct more tests of course we've got intrinsic complexity some things are inherently unpredictable and chaotic and even the most powerful AI cannot predict or entangle them substantially better than a human or computer today for example even incredibly powerful AI could not predict only marginally further ahead in a chaotic system such as the three body problem in the general case as compared to today's humans and computers and basically this is a largely unsolvable problem I believe in physics I should probably include a video of this but this is just something where um they're basically saying that there are limits to uh how smart A system can get and the kinds of problems

### Timescales & Malleability [17:41]

it can solve but the point of this problem is that it's just chaos like there is just pure chaos so even if a system is super intelligent it's only going to be marginally ahead in some areas due to the com due to the chaotic nature of the universe of course constraints in humans many things cannot be done today without breaking laws you know harming humans or messing up Society an aligned AI would not want to do those things and if we have an unaligned AI of course we're back to talking about those risks many human societal structures are inefficient or even actively harmful but are hard to change while respecting constraints like legal requirements on clinical trials people willingness to change their habits or the behavior of governments and for example what they're trying to say is that even if let's say for example a super intelligent AI managed to figure out you know all of these like faster ways to travel faster ways to build houses you know there's legal there's going to be zoning requirements there going to be building requirements you've got like you know certain bodies that are just slow you've got like um you know all of these regulatory issues and hurdles because of humans essentially and these systems are just genuinely quite slow because there are rules and regulations not that there shouldn't be rules and regulations there always should be but it's just like you know when we actually think about how things move in the real world that I mean some countries are going to move fast with this stuff but some countries and overall we have laws and rules and regulations for what individuals can do and I'm sure like you know there are governing bodies for like the world like you can't developed certain levels of nuclear weapons and things like that for example you know examples of these advances that work well in a technical sense but the impact have been reduced by regulations or misplaced fears which basically means that you know humans are essentially fearing this greater than it actually is include nuclear power supersonic flight and even elevators which is uh pretty trivial so of course we've got the phys physical laws a starer version of the point there are certain physical laws that appear to be unbreakable it's not possible to travel faster than light as I said before you know the physical light limit I'm not sure what the exact number is but you can't travel faster than that it seems like the limit if you're someone who believes we live in freaking simulation you could say that that's the like game speed whatever engine speed um and of course chips can only have so many transistors per square centimeter before they actually become unreliable and computation requires a certain minimum energy per bit erased limiting the density computation in the world there's a further distinction between time scales things that are hard constraints in the short run may become more malleable to intelligence in the long run for example intelligence might be used to develop a new experimental Paradigm that allows us to learn in vro what used to require live animal experiments or to build the tools needed to collect new data for example the bigger particle accelerator or to within ethical limits find ways around human based constraints for example helping to improve the clinical trial system helping to create new jurisdictions where clinical trials have less bureaucracy or improving the science itself to make human clinical trials less necessary or cheaper and he says all of this to basically say that we should now imagine a picture where intelligence is initially heavily bottlenecked by the other factors of production but over time intelligence increasingly routs itself around these other factors even if they never fully dissolve like some things like the ABS absolute physical laws of the universe are absolute which means that they cannot physically be changed but they can be you know worked around and um lived with and the key question is how fast it all happens and in what order now with that framework in mind this is where we get into the first area biology

### Potential for Improving Human Life [21:09]

and health biology is probably the first area where scientific progress has the greatest potential to directly and unambiguously improve the quality of human life in the last century some of the most ancient human afflictions such as small pox have finally been vanquished but many more still remain and defeated them would be an enormous humanitarian accomplishment Beyond even curing disease biological science can in principle improve the Baseline quality of human health by extending the healthy human lifespan increasing control and freedom over our own biological processes and addressing everyday problems that we currently think of as immutable parts of the human condition so now here's where he's talking about how super intelligent AI could work with

### Limiting Factors in Biology [21:53]

the limiting factors of the physical world of course one of the things that you use to do in order to increase the rate of uh scientific discovery is that you conduct experiments and you conduct these experiments in certain ways and some of these um you know take certain times of course you know you've got clinical trials you we've got human trials we've got you know the FDA for certain things um and of course this is where we talk about experiments on cells animals and even chemical progresses and

### Experiments & Data Quality [22:18]

even chemical processes El limited by the speed of the physical world many biological protocols involves culturing bacteria or other cells or simply waiting for chemical reactions to occur and this process sometimes can take days or even weeks with no obvious way to speed it up for example animal experiments can take months and human experiments can often take years or even decades for long-term outcomes on studies and somewhat related to this do data is often lacking not so much in quantity but in quality there's always a Dar of clear unambiguous data that isolates a biological effect of interest from 10,000 confounding things that are going on or that you know intervenes casually in a given process and I think this is really true because we have to think about the fact that like you know for example one of the things that you know people are getting into today is the fact that what is going to be today's um cigarettes because previously um I'm not going to get into any of the policies or rules and regulations around what was going on at that time but if we look back at how

### Historical Examples of Misinformation [23:17]

cigarettes were marketed for example in the 1930s to 1950s tobacco companies used doctors and advertisement to show that cigarettes were not harmful um they were claiming that you know um Lucky Strike in their ads used to basically say that cigarettes must be medically better for your throat what I'm trying to get out here guys is that you know some things that we look back at uh we genuinely did believe that they were good for us but over time some things only happen over 20 years 10 years 5 years which means that these physical limits it's quite hard to test these theories in a short amount of time even if we have super intelligent AI which means that the Cycles upon which the iterations can improve is going to be a little longer than we think in some instances for example if you wanted to test what happens every 5 years on something and if you had four iterations to improve you could only improve four times in 20 years because 5 * 4 is 20 and that's something that most people are not missing and for example you know people are you know stating that you know sugar used to be marketed as a healthy substance we now know that this is something that is pretty awful okay like it feeds cancer you know it just does a diabetes skin acne just a million different things are wrong with sugar I'm not going to get into personal dieting and stuff like that the point is that you know there are certain things that you know arguably happening in today's society that will only recognize you know 60 to 40 years from now because certain things just happened over a really long process of time that we just can't see right now not saying that AI couldn't you know simulate humans and do those kinds of things but it's important to understand the physical limits of the world and of course this is where he talks about you know there's a million different factors that could intervene as well of course

### Intrinsic Complexity [24:50]

he says in part for these problems with data is intrinsic complexity if you've ever seen a diagram showing the biochemistry of a human met ISM you know that it's very hard to isolate the effect of any part of this complex system and even harder to intervene on the system in a precise and predictable way finally Beyond just the intrinsic time that it takes to run an experiment on humans actual clinical trials involve a lot of bureaucracy and regulatory requirements that are unnecessary and

### Bureaucracy & Regulatory Requirements [25:16]

delay progress which is why I think that you know there's going to be um probably some societies that are just you know moving at light speed whereas others that are quite regulated um are going to slow down and we've even seen this with like the EU where you know the regulation there has slowed down so much and I think in the future it's going to become more apparent and I think yes it is kind of good that you have areas that are quite slow and not going to be impacting all the stuff because of course we don't know what superintelligent AI is going to do we don't know if it's going to you know kill us or be great for us there are certain areas and of course certain experiments have to follow a certain guideline in order to remain safe but I do think that certain small pockets are going to just race ahead with this technology regardless if we like it or not so here's where we talk about where the specific acceleration is going to come from he says here to get more specific on where I think acceleration is likely to come from a surprisingly large fraction of the process in biology has come from a tiny number of discoveries often related to Broad measurement tools or techniques that allow precise but

### Acceleration from Major Discoveries [26:12]

generalized or programmable intervention in biological system there's perhaps one of these major discoveries per year and collectively they arguably drive around 50% of all progress in biologies these discoveries are so powerful precisely because they cut through intrinsic complexity and data limitations directly increasing our understanding and control over biological processes a few discoveries per decade have enabled both the bulk of our scientific base of understanding of biology and driven many of the most powerful Medical Treatments

### Examples of Key Discoveries [26:45]

and this is where we get into some of the examples of what happens of course crisper a technique that allows live editing of any Gene in living organisms since the origin since the original technique was developed there have been constant improvements to Target specific cell types we've also got various kinds of microscopy for watching what is going on at a precise level Advanced like microscopes fluorescent techniques special Optics electron microscopes genome sequencing and synthesis which has dropped in cost by several orders of magnitude in the last couple of decades we've got mRNA vaccines that in principle allow us to design a vaccine against anything then quickly adapt it and of course these became famous during Co and of course we've got cell therapies such as CTI that allow immune cells to be taken out of the body and reprogrammed to attack in principle anything why did Dario amod recently talk about those discoveries because he thinks that their rate of discovery

### Potential for Increased Discovery Rate [27:32]

could be increased by 10x or more if they were a lot more talented or creative researchers basically to put this another way I think the returns to intelligence are high for these discoveries and that everything else in biology and Medicine mostly follows from them and why do I think this because of the answers to some of the questions that we should get in the habit of asking when we're trying to determine returns to intelligence first these discoveries are generally made by a tiny number of researchers often these same people repeatedly suggesting a skill and not so random search second they often could have been made years earlier than they were for example crisper was a naturally occurring component of the immune system in bacteria that's been known since the 80s but it took another 25 years for people to realize it could be repurposed for General Gene editing they are also often delayed by many years by lack of support from the scientific Community for promising directions and you can see this profile of the ventor of mRNA vaccines and similar stories abound third successful projects are often Scrappy or were afterthoughts that people didn't initially think we're promising rather than massively funded efforts and it's not s suggesting that it's not just massive resource concentration that drives discoveries but Ingenuity and although some of these discoveries have a Serial dependence basically you need to have Discovery a in order to have the tools and knowledge to make Discovery B which again might create experimental delays many perhaps most are independent meaning that they can be worked on in parallel and basically what he's saying that there are hundreds of these discoveries waiting to be made if the scientists were smarter and better at making connections between the vast amounts of biological knowledge Humanity possesses again consider the crisper example which we knew about since the 80s and of course the success of alpha fod alphao at solving important problems much more effectively than humans despite Decades of carefully designed physics modeling Pro provides a proof of principle orbe it a narrow in a narrow domain should point the way forward so basically what he's saying here that my guess is that powerful AI could at least 10x the rate of these discoveries giving us the next 50 to 100 Years of biological progress in 5 to 10 years and you might think why not 100x well it is possible but here are both serial dependence and experiment times become important getting a 100 years of progress in one year requires a lot of things to go right in the first time including animal experiments things like designing microscopes or expensive lab facilities but he does think that we could get thousand years of progress in 5 to 10 years but skep but it's quite skeptical that we could get 100 years in simply 1 years basically certain design and experiments have certain latency which need to be iterated upon a certain irreducible number of times in order to learn things that can't be deduced logically basically saying that look no matter which way you slice it we're going to have to spend some time figuring out the problem because there's no way to figuring out what happens after this occurs basically he goes on

### The Compressed 21st Century [30:21]

to talk about clinical trials and stuff like that but he says to summarize the above my basic prediction is that AI enable biology and Medicine who allow us to compress the progress that a human biologist would have achieved over the next 5 to 100 years into 5 to 10 years and I'll refer to this as the compressed 21st century the idea that after a powerful AI is developed we will in a few years make all progress in biology and medicine that we would have made in the whole 21st century a truly remarkable statement but I do think that based on what he's saying here is that it is pretty possible because there's only a few small discoveries that you do need to make that's going to account for the majority of discoveries that exist within the field and although predicting what powerful AI can do in a few years remains inherently difficult and speculative there is some concreteness to asking what could humans do uned in the next 100 years and simply looking at from what we've accomplished in the 20th century or extrapolating from the first two decades of the 21st or asking what 10 Crispers and 50 CES would get us we all can see how the level of progress might happen if we get powerful AI so

### Predictions for the Future: Infectious Disease [31:27]

here's what we say in exactly what could happen in the future years we could get reliable prevention and treatment of nearly all natural infectious disease given the enormous advances against infectious disease in the 20th century it's not radical to imagine that we could more or less finish the job in a compressed 21st then we also have the elimination of most cancer I think this

### Predictions for the Future: Cancer [31:47]

one would largely be one of the best things to ever happen I think cancer is awful and currently when individuals do get that diagnosis it can be like a death sentence it talks about how death rates from cancer have been dropping 2% per year for the last decades thus we are on track to eliminate most cancer in the 21st century at the current rate of human science and some types have been largely cured for example some types of leukemia with car therapy and perhaps even more excited for very selective drugs that will Target cancer in its infancy and prevent it from ever growing and AI will make it possible for treatment regimes very finely adapted to the individualized Genome of the cancer and these are possible today but hugely expensive in time and human expertise which AI should allow us to scale we also have very effective prevention and

### Predictions for the Future: Genetic Disease [32:34]

effective cures for genetic disease greatly improved embryo screening will likely make it possible to prevent most genetic disease and some safer more reliable descendant of crisper may cure most genetic disease in existing people which is going to be pretty crazy of course the prevention of Alzheimer's we've had a very hard time figuring out what causes Alzheimer's it's somehow

### Predictions for the Future: Alzheimer's [32:54]

related to the beta ameloid protein but the details seem to be very complex it seems like the exact type of problem that can be solved with better measurement tools that isolate biological effects thus I am bullish about ai's ability to absolutely solve

### Predictions for the Future: Other Ailments [33:07]

it the improved treatment of most other ailments this is a catch all category for most other ailments including diabetes obesity heart disease autoimmune diseases and much more most of these seem easier to solve than cancer and Alzheimer's and in many cases they already in steep decline for example deaths from heart disease have already declined over 50% and interventions like gop1 agonists have already made huge progress against obesity and diabetes of course we've got biological Freedom the last 70 years featured advances in birth control fertility the management of weight and much more but AI accelerated biology will greatly expand what is possible

### Predictions for the Future: Biological Freedom [33:41]

weight physical appearance reproduction and other biological processes will be fully under people's control we will refer to these Under The Heading of biological Freedom the idea that everyone should be empowered to choose what they want to become and live their lives in the way that most appeals to them there will of course be important questions about global equality of access but I think the biological Freedom one is going to be pretty insane people just going to choose to be tall smart muscular individuals I mean it's going to be um interesting they're going to be like 9 foot humans walking around I don't know and that's referring to some other stuff as well and one of

### Predictions for the Future: Doubling of Human Lifespan [34:14]

these things that he says here is that doubling of the human lifespan he says this might seem radical but life expectancy almost creased two times in the 20th century from 40 years to 75 so it's on Trend that the compressed 21st would d double it again to 150 obviously the interventions involved in slowing the actual aging process will be those that were needed in the late Century to prevent mostly childhood premature deaths from the disease but the magnitude of change is not unprecedented so he's saying that concrete there already exist drugs that increase the maximum lifespan in rats by 20 to 50% with limited um side effects and some animals like turtles already live to 200 years so humans are manifestly not at all some theoretical upper limit I guess the most important thing that is near did might be reliable non-good harble biomarkers of human aging that will allow fast iteration on experiment and clinical trials once the human lifespan is 150 we might be able to reach escape velocity buying enough time that most of those currently alive today will be a will be able to live as long as they want although there's no guarantee that this is biological biologically impossible I couldn't even um speak properly while saying that because I couldn't even believe it like my my brain just got like melted that's insane okay he said that like once human lifespan is 150 we might be able to reach escape velocity so that basically those who are alive today might be able to live as long as they want which is crazy I mean that is truly a crazy statement but I don't think he's talking about you know individuals that are maybe like 70 or 80 right now I think maybe he's talking about people that are just growing up now maybe those 10 and under I'm not sure exactly what age range that is but I think this is something that is true because you know for example if you look at people like Brian Johnson if you haven't paid attention to this if you want to know about like life expect expectancy cuz this is something that people talk about quite a lot with AI like there's a common phrase right now um in AI which is just don't die and um I think it's rather fascinating and thought-provoking that some people are stating that you know if technology continues to advance in the way it does if AI to continues to advance the way it does maybe we might get to a stage where people just don't die because we managed to keep extending Life by 25% you know every couple or so

### Brian Johnson & Longevity [36:21]

years and so life extension just it just keeps on going on basically and um I guess in theory that does make sense but of course it's not no guarantee that's biologically possible but I am talking about Brian Johnson because this is someone that if you are intrigued in about I can't even speak properly CU it's just so mind-blowing if you are intrigued about how longevity works and all those kinds of things this is the channel I would recommend you can see in his channel Banner he says don't die um he's basically sold his company and all this guy is trying to focus on is just not dying he literally just focuses on um all of these different things and it's kind of interesting like he's his internal body clock I think is that like of an 18yearold or 24y old and like 40 so it's pretty crazy to see how long this guy is going to live for um and yeah I think this is going to be one of the channels that explodes as well because he's basically the most measured human they've got like all his biomarkers and there's a team of scientists running around the clock just basically seeing how long this guy lives for so I mean it's going to be a really interesting future to see if other people also adopt this kind of thing too now of course he says it is worth looking at this lick of course he says

### Societal Implications [37:21]

here that it is worth looking at this list and seeing how different the world will be if all of this is achieved 7 to 12 years and now which would be in line with an aggressive AI timeline it goes without saying that this would be an imaginable humanitarian Trump the elimination all at once of most of these scourges that have haunted Humanity for the Millennia I mean that's pretty crazy like that thing happening in 7 to 12 years from now I mean can I fathom that happening yes and no I mean it's kind of hard to imagine a society where people can live to 150 the average life expectancy there's no cancer there's no Alzheimer's there's a lot of uh Gene editing things that you can just do I mean it's going to be pretty crazy like trying to you know make this video and do that at the same time is it's pretty hard and of course we do have this that um as the ratio of the working age to the retired population will change drastically so this is going to be um changing how the population structures as well then of course this is point number two I do apologize for this long video but there is just a lot of content here and I do want to get through it all um we've got the Neuroscience in mind in

### Importance of Mental Health [38:20]

the previous section I focused on physical diseases and biology in general and didn't cover Neuroscience or mental health which is good thing because not a lot of people cover mental health just because it isn't something you can directly see but it is something that is real and does exist but neurosciences a subdiscipline of biology and mental health is just important as physical health in fact if anything mental health affects human wellbe even more directly than your physical health hundreds of millions of people live very low quality of life problems due to addiction depression schizophrenia low functioning autism PTSD psychopathy or intellectual disabilities and I know that uh many of these issues are pretty true spoke to some people with these issues it's a it's a problem that's quite unfortunate because it's like it's not like okay like you know I cut my leg I go I fix my leg or you know I get stitches it's like it's quite hard to um fix and diagnose which is it's awful it's really awful he says that I expect

### Four Routes of AI Acceleration: Molecular Biology, Chemistry & Genetics [39:10]

AI to accelerate neuroscientific process along four distinct routes all of which can hopefully work together to cure mental illness and improve function traditional molecular biology chemistry and genetics this is essentially the same story as general biology in section one and AI can likely speed up via these same mechanisms there are many drugs that modulate neurotransmitters in order to alter brain function affect alertness or perception change mood Etc and AI can help us invent many more AI can also probably accelerate research on the genetic basis of mental illness which would be amazing spine Grange neural

### Four Routes of AI Acceleration: Fine-Grained Neuro Measurement & Intervention [39:43]

measurement and intervention this is the ability to measure what a lot of individual neurons or neuronal circuits are doing and intervene and change their behavior optogenetics and neural probes are Technologies capable of both measurement and intervention in live organisms and a number of very Advanced methods such as molecular ticker tapes to read out firing patterns of individual neurons have also been proposed and seem possible in principle

### Four Routes of AI Acceleration: Computational Neuroscience [40:07]

Advanced computational Neuroscience AI can probably be applied fruitfully to questions in systems Neuroscience including perhaps uncovering the real causes and dynamic of complex diseases like psychosis or mood disorders which is really true and it stands the reason

### Four Routes of AI Acceleration: AI-Assisted Therapy & Coaching [40:20]

that AI could accelerate these as well both the development of new methods and helping patients to adhere to existing methods more broadly the idea of an AI coach who always helps you to be the best version of yourself who studies your interactions and helps you learn to be more effective seems very promising so in 5 to 10 years he basically says that my guess what will

### Predictions for the Future: Curing Mental Illnesses [40:39]

happen is like most mental illnesses can probably be cured I'm not an expert in psychiatric disease but my guess is that diseases like PTSD depression schizophrenia can be figured out and effectively treated via some combination of the four directions above and the answer is likely to be some combination of something went wrong biochemically and something went wrong with the neural network at a high level and that is a system's Neuroscience question that can be solved conditions that are very structural maybe difficult but not impossible basically stating that you

### Predictions for the Future: Genetic Prevention [41:08]

know certain conditions are going to be harder than others effective genetic prevention of mental illnesses seems possible most mental illness is partially heritable and genomewide Association studies are starting to gain traction on identifying the relevant factors which are often many and number and it's probably going to be able to you know prevent those via embryo screening simil similarly with the physical disease everyday problems that

### Predictions for the Future: Everyday Psychological Problems [41:30]

we don't think of as clinical disease will also be solved so most of us will have everyday psychological problems that are not ordinarily thought of as rising to the level of clinical disease some people you know are quick to anger others have trouble focusing you know they're often drowsy some are fearful some are anxious and these are going to be things that AI able to solve so it's going to be interesting to see how many different new areas are going to solve these kinds of issues and this is crazy okay this is the futuristic okay he says that you know one topic that often comes up in sci-fi depictions or AI that I haven't intentionally discussed here is mind uploading which is the idea of capturing

### Mind Uploading [42:05]

the pattern of dynamics of a human brain and instantiating them in software this topic could be the subject of an essay of itself but I think that uploading is almost certainly possible in principle in practice it faces significant and technological and societal challenges even with powerful AI that likely put it outside the 5 to 10 year window that we are discussing which is essentially 203 30 to 2035 so in summary for part two he says that AI accelerated Neuroscience is likely to vastly improve treatments for or even cure most mental illnesses as well as greatly expand cognitive and mental freedom and human cognitive and emotional abilities it will be every bit as radical as the Improvement in physical health as described in the previous section perhaps the world will not be visibly different from the outside but the world as experienced All Humans will be a much better and more Humane place as well as a place that offers greater opportunities for self-actualization and I think of course the improved mental health will of course solve a lot of other societal problems including ones that seemen political or economic and I think this is one of the things that a lot of people don't understand is that you know mental health is something that isn't taken seriously by many systems and I think that once there is something that can maybe not blanketly address the problem but severely help with how humans are functioning on an individual basis a lot of the other issues are going to be solved like I think most people don't understand that like so many people who are you know criminals and people think are bad people like those are some of the people who've had literally some of the worst upbringings and they've had like they just you know have like mental health problems they're more susceptible to depression schizophrenia and all of these issues and it's just like okay this person committed a crime you know put them in the jail so for 20 years they're awful yada y I'm not saying you shouldn't be punished for your crimes but it's like you know we're just surface level tackling these problems so I think that is also something to take

### Global Inequality [43:50]

into account now this is where we get into something which is uh pretty dystopian because this is economic development poverty and basically it talks about how um the development needs to happen for everyone because there is a large disparity in different worlds it says more broadly many existing Health interventions have not been applied everywhere in the world apologies for that and that matter is the same for non-health technological improvements in general another way to say this is that living standards in many parts of the world are still desperately poor GDP per capita is $2,000 in subsaharan Africa compared to $775,000 in the United States and if AI further in increases economic growth and the quality of life in the developed World while doing little to help the developing world we

### Moral Imperative to Address Poverty [44:32]

should view that as a terrible moral failure and a blemish on the genuine humanitarian victories in the previous two sections ideally powerful AI should help the developing World catch up to the developable world as it revolutionizes the latter and I think this is so true you can't just leave parts of the World Behind and that yes there are certain regulations and all of these kind of things of how things get implemented but like I think it would be awful if only certain parts part of the world accelerate further than others and I think that the Baseline of course while it's improving we need to ensure that there is just simply a base level that Humanity functions at like there shouldn't be certain instances where $2,000 is you know the GDP of a country and like you know people are living on less than a dollar a day like these things are just you know it shouldn't be happening of course this is where Dario amade talks about you know um the challenges faces you know in the private

### Challenges in Developing World [45:20]

public sector of course by pervasive corruption like I said before even if you try and help some of these countries unfortunately sometimes people are corrupt they spend the money elsewhere they don't Implement things that are going on it's quite hard to do that but of course he says you know I see significant reasons for optimism

### Reasons for Optimism [45:34]

diseases have been eradicated many countries have gone from Port to Rich and it's clear that the decisions involved in these tasks exhibit High returns to intelligence despite humans conit and complexity so here are the guesses that he makes for the next 5 to 10 years after powerful AI is developed on the economy of course you've got the

### Predictions for the Future: Distribution of Health Interventions [45:50]

distribution of Health interventions this is the area where he's most optimistic for the future is that diseases have been eradicated top down in campaigns for example small pox was fully eliminated in the 1970s and polio and guinea worm were nearly eradicated with the less than 100 cases per year the logistics of the distribution can be also greatly optimized and that AI accelerated efforts are going to become

### Predictions for the Future: Economic Growth [46:12]

more effective of course we've got economic growth can the developing World quickly catch up to develop world not just in health but across the board economically there is some president for this in the final Decades of the 20th century several East AFC Asian asent sustained 10% annual real GDP rates allowing them to catch up with the developed world now here's what's crazy okay he says that human economic planners made the decision that led to the success not by directly controlling entire economies but by pulling a few levers such as an industrial policy of

### Predictions for the Future: AI Finance Ministers & Central Bankers [46:41]

export-led growth and resisting the temptation to rely on natural resource wealth and here is where we're getting into some real interesting territory he says that AI Finance ministers and Central Bankers could replicate or exceed this 10% accomplishment how do we get devel Ving World governments to adopt them while respecting the principle of self-determination some are going to be enthusiastic about it but others skeptical so they're basically saying that look previously we've seen that some leaders can develop policies that do work well and do Advance the economy but how on Earth do we get that to be implemented if some of them are going to be focused on self-determination I think this is of course one of the most interesting things because I'm not sure people are ever going to relinquish power to AI or even use AI suggested policies I don't know if that is going to change the future I think if certain economies you know have an AI suggesting policies and their economies just racing ahead I think some other nations are also going to potentially use that as well and I think that's going to be really I wouldn't say dystopian but I think it's going to be interesting I mean if this thing is smarter than this why wouldn't it make the decisions that you know control where our society goes of course

### Predictions for the Future: Non-Health Technologies [47:44]

it says finally non-health AI accelerated technology such as energy technology transport drones improved building materials all of those are going to permeate the world naturally for example even cell phones quickly permeated subsaharan Africa via Market mechanisms without needed philanthropic efforts of course we have to talk about

### Predictions for the Future: Food Security [48:00]

food security advances in crop technology all of that stuff of course we do need food security for people it's kind of insane that we have some areas where people just waste food and are not getting enough food of course we've got mitigating climate change is going to be felt much more strongly

### Predictions for the Future: Climate Change [48:15]

in the developing World hampering its development we expect the AI will lead to improvements in the technologies that slow or prevent climate change from atmospheric carbon removal and clean energy technology to lab grown meat that reduces our Reliance on carbon intensive factory farming I wonder what people think about lab gr me I mean it sounds pretty dystopian but I don't know I mean it's kind of interesting some people say that like why not because it's like if it's just exactly the same and it's grown from a cow like what difference does it make and I mean they have a point there and of course the inequality within the

### Predictions for the Future: Inequality Within Countries [48:42]

countries there is a ridiculous level of wealth inequality within um the United States and I'm not saying that you know like tax the rich or do any of that stuff I'm just saying that like in certain societies there is a baseline of living that some people just don't have and I think that that's really awful I'm still think of course that you should be able to work hard and you know achieve things but it's just like the system needs to be designed so that people just don't fall through the cracks of course there's also a dystopian problem which is where the opt out problem so this is one that's like kind of cyber punkish but it's like one concern in both developed and developing World alike is people opting out of AI enabled benefits so I mean people are going to be skeptical of the technology when it

### The Opt-Out Problem [49:20]

does come here like some people like just nobody's going to like no technology is 100% adopted you know of course you've got phones and computers which are pretty much Universal but it's like AI is something that's kind of like anthromorph asiz in the sense that like people are going to think that okay it's a being so they might not take its advice but Dario says here that there could end up being bad feedback Cycles where the people who are least able to make good decisions opt out of the very technologies that improve their decision-making abilities leading to an ever increasing Gap and creating a dystopian underclass and some researchers have argued that this will undermine democracy and this would once again place a more blemish on ai's positive advances basically stating that look if you don't adopt some of the AIS and you lead to worse decisions you're going to be basically screwed because you're never going to be able to uh catch up uh I mean it's crazy but he does say that one hopeful sign is that historically anti-technology movements have been more B than bite most people adopt it in the end next we're of course going to talk about peace and governance

### Human Threats & AI-Powered Authoritarianism [50:13]

and he basically says suppose that goes well we get disease hampered down we get poverty and we get inequality significantly reduced and the Baseline of Human Experience is raised substantially humans are still a threat to each other of course we need peace currently there are just numerous Wars going on in the world I don't really want to talk about them but um this is something that is unfortunate humans are a big issue with one another and of course one of the things we actually need to do is to maintain the fact that AI powered authoritarianism seems too terrible to contemplate which is basically where you have no personal freedom and I would argue that this is kind of happening in some societies not going to mention any names but uh I think I think we know um some societies is just is becoming a little bit dystopian where people can track your every movements you know social credit scores that kind of thing it's not really going to be pretty good and especially when like in 10 to 12 years when you have super Advanced AI this is going to be something that we need to have avoid being overpowered by authoritarian and present I can't even say that word right I'm sorry um pre prevent human rights abuses within um these countries here's where he talks about his best guess that the way to do

### Coalition of Democracies & AI Superiority [51:15]

this is via this kind of strategy which is where a coalition of democracy seeks to gain a clear Advantage even just the temporary one on pal wayi by securing its supply chain scaling quickly and blocking all delaying adversaries access to Key Resources like chips or semiconductor equipment and this Coalition would on one hand use AI to achieve robust military superiority but at the same time offering to distribute the benefits of power powerful AI to a wider group of it of countries in exchange for supporting the Coalition strategy to promote democracy basically stating that they're going to be in a better position taking the same bargain as the rest of the world which is pretty interesting I mean I didn't really think of that one and of course those of you who are thinking that AI is going to be making the decisions for lawyers in terms of sentencing and all that kind of crazy stuff um he says I'm not suggesting that we literally replace judges with AI systems but that we have

### AI in Justice Systems [52:06]

the combination of impartiality with the ability to understand and process messy real world scenarios feels like it should have some positive application to Justice at the very least such systems could work alongside humans as an aid to decision-making and basically this would allow the system to be more transparent and transparency would be important in any such system and a mature science of AI could conceivably provide it and

### AI for Transparency & Bias Detection [52:29]

basically they're saying that you know Advanced interpretability which is where they look inside the model and see what's going on is where they are going to be able to you know see inside the final model and assess it for any hidden biases and the crazy thing about this is that is actually so true like if you do get Advanced interpretability and you're able to see inside the model you can actually see any hidden biases in a way that is simply not possible with humans like the crazy thing about this is that even you now watching the video you have your own hidden biases with how you're raised where you are the data that's been put into you the TV shows you watch all of these things create hidden biases meaning that you can't see them yourself and I know this for myself like I have my own biases I try to always remain objective but it's completely impossible to see those biases but if you can actually look inside of an AI and see what's going on you can see those hidden biases and you know try and remove them which is pretty crazy which would mean it would be more fair um one example of this I want to show you guys something really quick um

### Judge Bias Example [53:21]

there was this one that was like about Judges there there's some great areas about this research but it basically said that the authors of this peer-reviewed paper looked into more than a thousand rulings made in 2009 by eight judges they found that the likelihood of a favorable ruling peaked at the beginning of the day steadily declining over time from a probability of about 65% to near zero before spiking back up to about 65% after a break for a meal or a snack basically stting that look the judges were more lenient once they've had a snack or a break and maybe they get more agitated as the day goes on which is pretty crazy because that could mean that we have a situation where um you know AI judges aren't going to do that because they don't need a break snack they're always going to remain in the same complete State you know it's pretty

### AI for Consensus & Conflict Resolution [54:03]

crazy it says how in a similar vein AI could be used to both aggregate opinions and drive consensus among citizens resolving conflicts finding common ground and seeking compromise some early ideas in this direction have been undertaken by the computational Democracy project including collaborations with anthropic and a more informed and thoughtful citizens would also strengthen Democratic institutions of course the last one that we have here is of course work and meaning even if

### The Question of Human Purpose in an AI-Driven World [54:25]

the four previous sections go well not only do we alleviate disease poverty and inequality but liberal democracy becomes a dominant form of government and existing liberal democracies become better versions of themselves at least one important question remains it's great we live in such a technologically advanced world as well as a fair and decent one but some might object with AI doing everything how will humans have meaning and for that matter how will humans survive economically and this is one of the questions that I've been trying to solve because I think this is coming sooner than most people realize and people are not paying attention and what's crazy is that Dario Amad says that I think this question is more difficult than the others I don't mean that I'm more pessimistic about it than I am the other questions I just means that it's fuzzier and quite harder to predict in advance because it relates to a macroscopic questions about how Society is organized that tends to resolve themselves only over time in a decentralized manner and he gives a great example here he says he gives an

### Hunter-Gatherer Analogy [55:21]

example right here where he says for example Hunter gather societies might have imagined that life is meaningless without hunting and various kinds of hunting related religious rituals and would have imagined that our wellfed techn technological Society is devout of purpose basically saying that look back in the you know in the olden days the really olden days like you know areas when we just had sticks and running around hunting we might have thought that okay if people aren't hunting all day what on Earth are we going to do but we can now see that with technology we still have a lot of purpose and of course they might not understand how our economy can provide for everyone or what function people can usefully serve in a mechanized society which is really true okay it's quite hard to look forward and think how things are going to work but looking back we usually understand basically he's stating that you know there is a sign that there is clearly a lack of clear answers and this is one of the things I've constantly seen while researching this is that there isn't any clear answer to what people are going to do and I think it's something that needs answering genuinely like this is pretty crazy of course this is where he does talk about the fact that you know um I spend my time playing video games swimming walking around outside all of these generate zero economic value um of course it doesn't really matter that someone is better at those things he's basically saying that just because AI is better than those things it doesn't really mean that you know there's no value to you like you're still going to

### Economic Value vs. Meaningful Tasks [56:32]

enjoy them and of course he's basically talking about like there's an economic piece which is can you generate economic value in an economy versus the meaning piece which is like do you still have meaning in the AI economy which is where um you know are your tasks meaningful to yourself like there's two uh separate issues here's where he gives his predictions on this which I think is rather fascinating he says first of all in the short term I agree with arguments

### Predictions for the Future: Short-Term: Comparative Advantage [56:53]

that comparative advantage will continue to keep humans relevant and in fact increase their productivity and in some way level the playing field between humans as long as AI is only better at 90% of a given job the other 10% will cause humans to become highly leveraged increasing compensation and in fact creating a bunch of new jobs complementing and amplifying what AI is good at such that the 10% continues to expand and employ almost everyone which is crazy and in fact even if AI can do a 100% of the things better than humans but it remains inefficient or expensive at some tasks or the re Source inputs to humans and AIS are meaningfully different then the logic of comparative advantage continues to imply one area humans are likely to maintain a relative or even absolute Advantage for a significant time is the physical world thus I think that the human economy may continue to make sense even a little past the point where we reach a country of genuses in a Data Center and I think this is really true because there are certain things that humans in the physical world just simply do better than AIS just running around a computer and I think it's important to understand that things on a computer are going to become increasingly economically hard to generate value but I think that if you're doing something that exists in the physical world I think that's going to become increasingly valuable which is why samman said a while ago that like you know the jobs of plumbers and electricians and those kind of people might actually be going up in the future not saying you should quit your career and become a plumber but it's just funny how these things are starting to change

### Predictions for the Future: Long-Term: Economic Model Breakdown [58:15]

he does say here that however I do think that in the long run AI will become so broadly effective and so cheap that this will no longer apply at the point our current economic setup will no longer make sense those there's going to be a broader societal conversation about how the economy should be organized which is rather true okay um Dario am the CEO of this multi-billion dollar AI conglomerate not really a conglomerate actually um is stating that you know it's it's no longer going to make sense like the economy is going to break at some point which is really important for those of you thinking about the future because I mean um I'm I do hope that you know policy makers are really starting to take into account that there is going to be some point at which the economy breaks because at that point what are people going to do because is going to be not a lot for humans to do and I think um once AI becomes so cheap um it's going to really throw things off the rail which I mean I don't know it's it's going to be fascinating there really aren't any answers right now but

### Predictions for the Future: Potential Solutions [59:07]

of course he says um you know the solutions for this could be uh simple as a large Universal basic income for everyone although I suspect that there will be only a small part of the solution and it could be a capitalist economy of AI systems which then give out resources to humans based on some secondary economy of what the AI systems make sense to reward in humans based on some judgment ultimately derived from Human values which is completely crazy or perhaps humans continue to be economically valuable after all in some way not anticipated by the usual economical mod modules or models Jesus I can't even speak properly I've been speaking for an hour so forgive me and the crazy thing about this is okay and this is what I want to leave you guys with this that he says that um as with some of the other challenges we will likely have to fight to get a good outcome here exploitative or dystopian directions are clearly possible and have to be prevented much more could be written about these questions and I hope to do so at a later time but he's

### Dystopian Possibilities [1:00:00]

basically saying look dystopian outcomes can happen this is why I keep telling people okay like I'm not trying to be an AI Doomer I'm just saying that like if we look back at how much technology has advanced Society the fact that there is still dystopian elepants of our society where you've got like these billion dollar condos and there's like homeless people walking around you have to understand that like um just because technology advances in one area it doesn't mean that it lifts everyone up it does to an extent but some people fall through the cracks and if we get to a society where the economic model breaks dystopian outcomes are clearly possible which means that we have to make sure that as a society we fight to ensure that like we don't end up in a dystopian hellscape because it truly is possible um and the world is isn't as

### Dystopian vs. Utopian Outcomes [1:00:40]

bad as you think and isn't as good as you think it's somewhere in between so I mean let me know what you guys think about this I really didn't make to you know make this video that long I do have a bunch of um time stamps but please do

### The Importance of Fighting for a Good Outcome [1:00:51]

leave a like for this because this is I don't know if this is my longest video but hopefully you guys have enjoyed my thoughts and comments and stuff like that but let me know what you guys think about the future of AI I think this is going to be an interesting thing for the next 5 to 10 years it will be

### The Next 5-10 Years [1:01:03]

interesting to see if there are slowdowns if these things going to happen but at the end of the day he does say that this is basically when we do get an AI system that is pretty crazy um in terms of basically Geniuses in a data center which might not be as far away as people think