I'm motivated to study this because of uh climate change and actually this all started back in 2009 when Dan arisu who was the head of the national renewable energy laboratory and who also is a new Mexican um came here with a group of people from the national renewable energy lab and said hey can you guys help us think out of the box and after a few hours of talking with them it became clear that the whole problem of adapting to climate change hinges around technology which Technologies are going to cost what when and so ever since then I've been working on this uh first with a group of people here some of you a few of you may remember ban NJ uh who by the way disappeared nobody knows where he is now um long interesting story um but okay so this is just the de ongoing development of that work now it's it's clear that we need better guidance in the energy transition uh I would argue the track record so far is poor um standard integrated assessment models which are the main Workhorse tool people use for this compute an optimal transition path based on a given climate emissions Target so that's the question they're asking and um I think it's a peculiar question to ask because it really only makes sense if we had a global dictator who could say this is what we're supposed to do and um if you believed these arguments about that you know we we have to make a tradeoff between well-being and transition that might make some sense but um people don't do that and so models are untestable you poke at these guys and say look your model hasn't been doing very well they say oh well it it it's not designed to answer that question and just to illustrate here's a picture of projections that have been made um uh so so here is the actual cost with the black dots of um photovoltaic solar energy um these red and blue lines are predictions made by various models of that type the the black line there is the trend line from 1980 to 2020 and what you can see is like here's a forecast made in 2000 for 2020 and they were off by about a factor of 50 okay so they were off by a factor of 50 and they were too high on the cost and they've been making forecast since even before that and they've always been too high way high and they're high again and again and again and again and I used to make the joke that the people at the International Energy agency somehow they can't plot data on log lmic scale because when you plot the data on logarithmic scale the trends are apparent these by the way are the projections they made for improvement rates that were reported in 2014 but mostly around 20 thou uh 2010 for how much were going were were was solar energy going to drop in cost over the next decade and you can see there were about the order of 3,000 different projections none of them came close to the actual value which is that black line now why are these models so off well to do this I already said they're asking the wrong question but they also have to make ad hoc assumption so these models can get quite complicated I mean the remine model at pick probably I would guess thousand person years have gone into building it um and they have to assume things or they assume things when they make them like they they put in solar photovoltaic and they use a learning curve I'll describe what that's about in a minute and they um uh but then the model says oh do nothing but invest in solar energy well oh that can't be right prices can't get that low so they put in floor costs so these green lines are examples of floor cost in various models of this type the black uh line again is PD system cost you can see it just repeatedly punching right through the floor costs that they assume were impossible to achieve um okay now so our model modeling philosophy is different we we want to ask answer a different question we want to try to predict what will actually happen and how can we nudge it and I don't mean nudge in the sense of psychology but what can we do to incrementally make things better we want to make testable falsify predictions one of the problems with these other models they're completely unvalidated they just make projections they don't call them predictions because they don't want to be too assertive and so we test everything out of sample we try and understand how good the predictions are and we want to build reliable models and we're actually going to do things much more simply now those of you who were at my talk last year or who were at my talk last week will have seen this picture doesn't hurt I thought I'd put it up again because some of you weren't at either of those and um and so this is the history of the global Energy System over the last 140 years and on the left we see the costs on the right we see deployment and we see the different kinds of Technologies so say that Dash black line up there is the cost of Coal Energy um the no sorry the brown one is Coal Energy the black one is oil the gray one is gas you can see that those they wiggle around a lot but there's no obvious TR Trend um in contrast some recent uh contenders like solar PV the orange one um uh hydrogen p2x fuel meaning hydrogen-based fuels the green one um wind that blue and there they're all coming down in cost in a fairly regular way in fact solar photovoltaic energy costs one 10,000th now of what it did in 1958 when it was first de loyed in the Vanguard satellite um and then over on this side you see uh the deployment of these energy sources that green line across the top is traditional biomass the dash Brown Line is coal the black and gray lines are oil and gas then the Blue Line Is hydro the pink line is nuclear the dark green line is biopower blue is wind electricity purple is battery and orange is solar solar PV and then the green one is um again p2x fuels so again something remarkably different is happening you see that historically coal and gas came up exponentially but at a at a slow rate compared with what solar and wind and batteries have been doing um it's also notable that nuclear energy initially shot up at a fast rate and then plateaued out um okay so that's just background and and in a sense what I'm talking about in this talk is projecting these kind of curves so how do we project the cost curves and how do we project the deployment Curves in fact I'm not going to just say project I'm going to say how do we predict the future of these and how do we do it in a probabilistic manner so we understand how good our predictions actually are Jim what's useful useful energy ah so that's a it's a bit of a tricky thing because you have to make some assumptions but if you just ask about what's usually called final energy that's how much energy is there if you could actually get every Jewel of energy that's in it to do something useful useful energy takes into account typical conversions so if you're running a um uh normal you know gas fired or sorry uh internal combustion engine then the efficiency is about 25% so you take the final energy in the gas and you divide it by four sort of like a free energy it's it's kind of like free energy but it's a little little rougher than that because it really depends on the way it's being used but we did that even though there's some ambiguity in how you do it because we wanted to have an Apples to Apples comparison people there's a lot of confusing crap in the literature where people uh talk about final energy which is kind of irrelevant now I'm going to so somewhat zip through the first part of this talk talking about cost because I've been working on that for de decades and some some of you at least have heard it before so but because I don't want to get to the the topic of the title I gave but I still want to say something about it because I think it'll interest those of you who haven't seen it um so first of all in predicting it's worth noting that techn though you might think oh technological progress impossible to predict because it's Innovation innovation's unpredictable by its very nature yes but the rates of progress are highly predictable as Gordon Moore showed we'll get back to that but just to illustrate here's the cost of various things over 20-year period in the United States ranging from hospital Services which you know over that 20-year period went up by more than a factor of two to televisions which at least if you adjust on a quality control basis dropped by uh 95% in other words they went down by a factor of 20 in cost huge differences in how these different things behave also notice they're very persistent televisions didn't just suddenly drop they drop every year um and we're back to Moore's Law which is quite a remarkable thing you know and originally Moore made the statement that the density of transistors was going to double every 18 months he later on adjusted it say every two years but since then that prediction has been remarkably good though we've always known it was going to hit a wall around now when um things actually reach the quantum scale so it's unclear what the future of this law is uh it seems like it's got to stop um and so this just shows you what I said that even if the Innovations are unpredictable and actually I remember being in a conference here David was in the room um Doug was in the room and it was on Innovation and invention and so forth and there was a guy standing up here um who was an actual chip designer and he said you know you guys you all think Mo's law it's like god-given it's just going to happen every year and I can tell you every year we we'd feel like we were facing a brick wall and then we'd pound on it and eventually we'd figure out a way to solve the problems so so well Mor's law you know some people say it's a self-reinforcing prophecy I don't believe that first of all it obviously wasn't before Moore wrote it down why did he write it down because it was happening before he noted it and and secondly well I just think it's caused by other things um though it is certainly extremely useful and this was driven to me home to me by Alvi Ray Smith my fellow New Mexican who is you know one of the brains underlying Pixar and he said at Pixar they had all the technology ready about 5 years uh before they could actually make a movie and they said look we just have to wait we'll we'll polish it up and make it better but we just need the machines to run faster and as soon as that happened they made Toy Story so so they were able to plan in other words around that and in fact Mo's law did mean that the rest of the industry could plan assuming that CPUs were going to be faster in a certain date etc etc it's extraordinarily useful to have a law like this now uh there's also rights law which is actually much older it it's a little bit different it says what Wright said observing he was in the aviation business he said that any given airplane Factory every time its cumulative production doubles the cost of producing the airplane drops by 20% so works good for airplanes works good for tons of other Technologies it also works on the global scale as well as on the level of individual factories um so different law and it's it's it's now a power law which actually means that the rate of progress is slowing down in the sense that you have to have exp it's every exponential increase gives you a 20% drop okay now so we extended this we developed a probabilistic method for forecasting technology costs based on historical data we did a preliminary version of this here at SFI with baa and and then we extended it to do it better better um with franu Leon and myself and then some other collaborators and we tested the method by making 6,000 forecasts for 50 different Technologies it worked pretty well and we applied it to three scenarios for the green transition so again I'm going to pace through this a little bit fast you can always stop me and ask questions but our forecasts look like this so notice we're not making a point forecast And since this forecast is kind of trivial you can see here's past data for solar PV starting in 1980 um you can see it's Wiggly but we draw a line through it we actually just go from the first point to the last point and from that last line our model says well these are the probabilities of the different outcomes where there's a 5% probability of being out in the white zone and each of those bands correspond to a 5% probability so it's a probabilistic method and we actually tested the probabilistic method right so yeah in other words we believe we have a pretty good idea of how good the forecasts are and to illustrate we actually wrote the paper where that last circle is in 2013 that was at least the most recent data and you can see the subsequent data points and they're you know well within the band that we predicted they should be within um uh you know we have here we we had we then used this method to look at actually 14 different Technologies IES some of the ones like fossil fuels we just used something simpler because they don't follow this rule um the technologies that do follow these kind of laws are somewhat special but anyway we we built some scenarios to ask about what's the cost of the climate transition be and is there a way to get there that's cheap so we looked at the key Tex of solar PV when batteries p2x fuels we extrapolated the deployment Trends we ass we phased out fossil fuels over the next 25 years that's an assumption I'll come back to it in a minute and we used p2x fuels for energy storage so we stored enough we assumed the World by the end of this 25-year period was going to have enough stored hydrogen that if the sun completely stopped shining so it it's just night all the time no wind blowing we could run the planet for a month on the stored energy not clear exactly how much you need maybe you need even more than that because of inter but it's still a pretty fair amount we also dealt with all the hard to decarbonized sectors like heat shipping Air transport and so on and we did all those with p2x fuels in other words you make um you use solar energy or wind to make electricity to make hydrogen to make fuels and then you burn those fuels for those applications and we tried to make very conservative cost estimates because we didn't want anybody to accuse us of you know boosting our favorite techs so here are the three different transitions that we looked at uh your labels are gone but there's the fast transition here the slow transition in the middle and no transition on the right or slow trans very slow transition you can see we enormously ramp up electricity which is shown down here and that orange band is the role solar plays the blue band is the role wind plays the green band is the role hydrogen p2x fuels play and all the other stuff is you can see the fossil fuel is getting phased out in that plot in the upper right upper left corner so basically we assume in about 20 to 25 years these take over now is that plausible in our paper we showed these three plots you see solar wind batteries and p2x electrolyzers we drew a red trend line it's plotted on semi- lar scale you can see solar so far and actually if you just extrapolated it exponentially it take over quite a bit faster than we assumed we assumed in that upper blue line that it starts to taper off right now but but and then it tapers off even slower uh no even faster in the slow transition or the no transition um okay and this is showing um the re the top panel the white space is nonfossil fuel and so you can see by 2040 non-fossil fuels are largely taking over the blue one is the is the fast transition so in the fast transition most most energy is being provide not being provided by fossil fuels by 2040 these these predictions are all probabilistic so here you see the fast transition slow transition no transition so there's still quite a bit of overlap in there I think we've been cautious in making sure our a bands were big enough uh that we're going to stay within them and this shows you the relative costs over here so if you say take a 2 and a half% um uh discount rate then we save the order of 10 to 12 trillion doll so we save a lot of money in other words by doing this why because since we go down these learning curves since we Renewables let's just go back to that earlier plot if these Trends continue you see solar in particular wind at a slower rate by 2040 or 50 they're much cheaper they make energy much cheaper than it's ever been and because of that big savings we end up actually saving money by making a rapid transition now I want to point out that um oh and wait let me show one other thing the plot over on the right here shows greenhouse gas emissions due to energy generation and you can see by 2060 they're at zero they're actually pretty close by 2040 um now I just want to beat my chest a little bit in 2010 I published a forast in nature that um the cost of solar PV would be cheaper than cold fired electricity and um I quote The Economist from 2014 four years later solar power is by far the most expensive they thought solar energy enthusiasts were nuts and yet my prediction was right although people it really irritates me again and again they say nobody could have predicted this and I go all you had to do is look at the data I didn't have to be a genius to make this prediction I just had to look at the data and believe the data which nobody else seems to have done or few others did but but not the main people in the IP CC and so forth and by the way this just shows the track record of the simple method we're using these bands in each plot are the 90% quantiles so you can see and making forecasts at different points in time for solar wind batteries p2x electrolyzers you can see the forecasts for p2x electrolyzers are still kind of shaky based on that data alone that's because the technology has not been deployed that much yet on the other hand that technology is a lot like a lot of other technologies that we have lots of data on and and so it's the behavior we're predicting is is in line with that namely it should improve at a fairly decent rate um so you can see these forecasts all did pretty good for all four Technologies and okay I'm going to skip this slide now I'm doing pretty good with time um so what about diffusion everything I was talking about until now was about cost and it was under the assumption that we we the diffusion that is the deployment of these Technologies happens at a certain rate and then we got cost predictions depending on the rate at which that happens but to deal with climate change we really have to do this pretty fast so you might immediately ask which of those three scenarios that I showed you is most realistic so you know it's I suddenly realized a couple of years ago that nobody has done this problem right um so question is can we predict how quickly these will diffuse thus how quickly will the transition happen well let's just look at the historical data a little more closely so here we see solar PV and wind we see the deployment since actually 1976 for solar 1984 or so for wind and semilog scale we draw a line from first point to the last point for reference or or uh linear regression and you can see that well it's kind of Wiggly but roughly speaking these have both increased exponentially uh for the last 45 years and uh and at a pretty hefty rate the order of 35% per solar and a little less than that per wind and if you s at the curves there's maybe some hint that they're starting to Plateau out but it's not clear and notice solar in that period of time it's increased in in its deployment level by a factor of a million six orders of magnitude so huge and and by the way since we're on this slide this is an important thing to keep in mind people go oh well you know there's still not that much solar and wind it's only a few per of global energy generation but exponentials with high rates of growth are tricky it's easy to let them fool you because they're small and they're small and they're small and then suddenly they're really big so these are about to appear on the stage in a major way but you go well how do we know they're not just going to flatten out it's totally possible okay well so to do that we start getting into technological diffusion Fusion if you look in the literature there's a long literature saying the deployment follows s-curves and so in other words there's some kind of s-shaped curve for example the logistic function that I plot down below where in this function you have that the deployment at some point in time is equal to the ASM totic deployment level Time 1 plus expon exponential of minus K which is a rate constant time T minus t 0 t 0 is a location parameter that is where the S curve is and so you can take data and fit these kind of curves to the data and and there's a whole kind of story incubation rapid growth maturity blah blah blah um now people have tried to use S curves to forecast what solar and wind are going to do and here I give you some examples uh let's maybe start on the top left panel so there's four different studies that are colored in the dots are the actual deployment levels for uh electricity Generation by solar energy um those gray lines are fits that were made at different end dates so in other words we go along and pretend the data is you know we're stopping in 2015 we make a Fit Stop in 2016 we make a fit so we make fits every year just the only thing that changes we add one more data point every time we do the fit and you can see they're all over the place so obviously this doesn't doesn't work very well um but you'll also notice the four published studies that I'm going to beat up on a little bit here are all well below that curve I'll come back to that in a minute and you know some of these made big claims these guys Kramer and Hugh hey I spoke at a conference at Oxford in 2014 maybe and I was saying you know look we're solar energy is going to get really cheap etc etc they said oh you're completely full of [ __ ] uh it's just can't diffuse it can't take over more than 1% of the Energy System why because of blah blah blah blah blah about you know there's been no example that did that I and that got published in nature um so all right then here's a couple more studies um similar kind of story and there's wind wind's a little different a few of them actually are on the high side but okay so it's easy to to um generate foolishness by just using Lea squares fitting on this this data and here's another example that that was at National level Global level data this is National level data where things get worse there we see Austria where we see two fits one made in 2019 one made in 2012 you can see they're both utterly different and you can see the data starts to get a lot bumpier and in fact if you you know delve into it you can ask what happened there well um Austria had a generous policy here favoring the deployment of um this one is wind and then they remov the policy and then they instituted another policy and so you can see these curves get bumpy at the national level somehow though when you average a lot of countries together it kind of Smooths out and just another example of the garbage that's been published um this is a paper by chirp at all um and they're claiming that when you look at the historical data doing a lot of mumbo jumbo that um you can see that it's just utterly impossible that solar and wind are ever going to provide the energy because they had actually a good idea the idea was let's look at the leaders the early adopters let's see what how fast did they adopt and what level did they Plateau out and let's assume the others are going to be kind of the same and then we can extrapolate from those leaders not a crazy idea except the difference between the leaders and the laggers is not big enough to really tell the difference very accurately the data is super noisy and you easily get absurd fits like they say oh they include only the cases where things have reached maturity and whether or not something's reached maturity like changes Year bye it'll first reach maturity indicating it's achieved more than 50% of its deployment level and then next you oh nope wrong didn't achieve 50% and back and forth all right and we actually have now written a paper criticizing all this stuff I hate writing papers like this because you know it's not pleasant the people that you criticize get extremely angry but we kind of had to do it because back to Moore's Law it's planning for the pace of the transition if you believe these guys then we shouldn't be investing too much in solar and wind because they can't do the job they're not going to get there quick enough etc etc ET so it's it's a sort of high stakes thing and as I urge them we should leave our egos behind unfortunately that hasn't happened um but but so we go through and and and just show unless you're very careful to to not do overfitting data selection bias and make other AD up assumptions you can lead to you can get unreliable conclusions and I just wanted to make it sure that it was clear that back here I mean notice these predictions have been completely falsified because they predicted the ASM totic level was here and it's already here um okay now this is a hard problem and there and there's a good reason why nobody's done it right yet several good reasons first of all the data is terrible it's hard to come by a few you know fantastic Pioneers have gathered some data and we've made heavy use of them but it's still the data sets are pretty bad the time series are also short um you can see this let me actually go back and make this point here you want to forecast these curves based on this data this data goes not quite but almost to present I guess it goes up to 2021 but the thing is in an s-curve notice there's not a lot of bending in an s-curve things start to bend away from exponential growth but because well let me just say this this function in the limit where T is small it's just pure exponential growth because this term is negligible right it's only later that that this term becomes comparable to this term and that's when things slow down but just imagine trying to predict where this is going to deviate it's pretty hard to do that okay now the growth is inherently non-stationary when we're in costs we assume there's some process that's like Mo's law or a generalization of it or rights law or a generalization of it there're stationary Moors law in particular but growth is inherently non-stationary because well you know there's uh you know childhood middle age and maturity um the deviations from exponential growth are small in early stages I just demonstrated that and as I'll demon demonstrate a minute the noise is heterosis meaning the size of the amplitude of the noise changes during the life cycle of the technology the noise is heavily autocorrelated meaning if you're above the standard S curve you probably will stay there for a while the the noise fluctuations if you get a big one the next one's probably going to be big two and the fitting is strongly downwardly biased that's why we saw most of those forecasts being on the low side now so how do you think about this um you know I remember I was making a presentation for trying to get funding to think about this stuff and and one of the people I I said well you know um he said well we know why solar energy got cheaper and I go we do and he said yeah it's because of the Chinese and I go well yeah but it was getting cheaper at the same rate before the Chinese even thought about making solar energy you go but so I find over and over again people have a there's always a story with solar energy okay we used it satellites poles then you know a bigger Niche because people started putting little widgets on their lawn and and uh blah blah blah you know the Germans put in feed in tariffs and then the Chinese stepped in so that's the theory I would call one damn thing after another in contrast the logistic equation which I suggested should fit these curves um if we were talking about say bacterial growth we could write down exactly what the bacterial doing how they reproduce the rate at which they die we can solve those equations which are roughly you get a differential equation like this there's a little death term too and we can come back to the modifications maybe in the questions and so that's a deterministic process and once you've observed the bacteria in the petri dish long enough you can predict what their future population's going to be fairly well now I do want to emphasize this work as preliminary so let's begin by making a model based on the logistic function we took the logistic function and we've done one thing we've added this term over here which is the noise term and the noise process is now an autor regressive of order one process meaning if if that makes the noise persistent and this parameter row adjusts how persistent the noise is if row is one it's so persistent it's like a random walk and if row is close to zero then it just reverts back really fast and we find by the way it's quite persistent row should be like 08 um so and and then this noise here is just a gaussian IID noise so that's the model we're going to investigate and then let me see what the next slide is and then we can do a trick that physicists love of making a collapse plot so we to we gathered data on 47 Technologies when I say we I mean my graduate student Ben vagen vort and um uh with some help from Brendan Tena and Leonard uh bartner um and and all the people who collected the data that they gathered originally but we had gathered from a lot of sources so we have 47 Technologies now and we plot them in non-dimensional coordinates so let me go back and explain that so non-dimensional coordinates you see um K the parameter K has dimensions of one over time as it must because it's multiplied by something that has dimensions of time similarly L has dimensions of number of things so we plot them in units where L is by definition one p is all time is measured in units of K so if things are slow then we we compress time if things are fast we expand time so that everything's on the same scale and we set we locate them all in the same spot and so when you do that you see you get an interesting collapse plot this is plotted in what's called a Fisher pry transform meaning we take this sigmoidal scurve and you make a transform that flattens it out so now in in Fisher pry coordinates the growth is linear um and and so you can see that over in this part you see what looks like a plausible straight line but in this part you see all the Technologies are looking pretty different so when we first did this we said aha because the literature talks about stage one stage two and stage three technological growth roughly you know early middle-age maturity and so there we have a way of determining where stage one is because we can detect roughly where this transition from this really fast Behavior at the beginning often in a you know sort of preul commercial stage and then when it gets on an remember this is a fish in Fisher pric this is roughly speaking exponential growth and and and then flattening out into maturity so anyway thought great now we can isolate where stage one and Stage gives way to stage two for all these Technologies but then we realized that there's another hypothesis oh we can make which is called a GT's growth process so in a GT's growth process the derivative is some rate constant K times the log of LX * X in contrast remember the logistic one I did before is looks like this now you know so somehow it seems like there's at least some suggestion that Technologies don't this is maybe not quite as good is a scom process in bacteria it's pretty easy to understand why this is true the KX part is just because that's exponential growth and then the one minus X is that as the bacteria fill up the petri dish they have less and less food so they can't grow exponentially anymore but so this is some hint that the gumert process may be better um by the way these are just to illustrate four energy related Technologies and maybe I should mention in this set we have we have all kinds of stuff uh some of them you might even question whether we should call technology so we've got canals railroads internet all that kind of stuff washing machines cars nuclear weapons ammonia synthesis we even have um nuclear missiles and monasteries so um they all seem to follow very similar S curves now one thing you realize when you when you once you make this collapse is aha there's some universality and in fact back to my question of how much is just one damn thing after another and how much is the terministic growth well we kind of see it here the signal is that black dash line the noise is the variation in all those gray curves and the answer is that the bulk of the variance is explained by the deterministic function this is true even though you know because back to Sol and when people go oh but you know there's all this we have to have all these policies in place we have to do this we have to do that and it really depends completely on the policy how can you even make a forecast without that well all these other Technologies you look at railroads start reading about the history of railroads there were huge battles over rights of way very political thing um uh a lot of corruption etc etc etc but so so it's partly signal and partly noise but the signal Act is more powerful than the noise now okay let's put this and make forecast and um so here's for solar PV now first let's ask the question what this suppose you have a strong PRI about where it's going to land you for whatever reason believe that solar is going to land let's say the current um uh the current level production which let's see this would be a the power sector in 2022 so that's where it's going to land that's all right no wonder yeah much better um that um okay suppose you believe that's where it's going to land then you can ask what's the probability for different outcomes given and here we're assuming a logistic function and well look up at so the our prediction would be we'll get to 80% of that ASM toote in 2037 in other words 13 years from now solar energy will be generating 80% of the power in the grid and those bands there show you one and two standard deviation variations so the prediction is not perfect but you're at a 95% downward prediction would be 2036 and upward there would be 2030 2040 right so surprisingly small band why you start plotting these curves against each other you see they're not that bendable we actually tried applying functions with more than three parameters and by the way already with three parameters it's extremely difficult to get a fit but but you know you don't actually need those you get just as good a fit with these simple three parameters because once it's in motion there's a lot of inertia in this deterministic growth laws um now the plot on the right shows the preliminary forecast for solar PV capacity without assuming a prior and this is I emphasize preliminary again because our forecasts are going to change because we're tweaking up the method and I could come back and say a lot about all the work we've had to do to deal with all those problems um but there's our preliminary forecast and we're actually forecasting that it's going to land more over in here and um so you can see the plot for yourselves yeah so that ASM tootes around 16 terawatts um now I want to leave room for questions I want to emphasize I'm not saying policy doesn't matter in this is all just going to happen inexorably policy does matter just look at the Wiggles in these curves a lot of those Wiggles were probably policy driven um so policy matters um on the other hand the good news is we've got a lot of inertia going uh we do need to deal with things like grid expansion fossil fuel politics land use is a big one because I'm leaving out the 25% of emissions that comes from agriculture mostly cement but I I still think we're to see energy get cheap and it's going to happen fast um okay it's not just that okay we uh it's cheap low volatility of energy prices you could see the way those prices bump up and down it's actually very bad for the economy when Energy prices are fluctuating all the time you can't anticipate what they're going to be a lot of things you want to do depend on what energy prices are going to be and so we should have much less volatile Energy prices for reasons I can explain if you want better energy security for anybody that has sun or wind which is almost everybody low pollution uh much lower pollution lower environmental impact in general and much more sustainable and most importantly no greenhouse gas emissions so now you might say oh but you're just extrapolating time series that's true that's all I've done we're building models that do other things I think I already showed this slide in my talk last week um we by the way looking at the labor transition this is showing that we actually have a big the good news about the transition is that we're going to get a lot more jobs it's actually a job increasing thing at least for the next two decades the bad news is it's a job bubble because a lot of Renewables depend on construction and construction you do the construction and then you're done and so it's it's going to continue to grow at 20 or 30% per year until it plateaus out and energy growth only grows at about 2% per year though it may start growing a little faster once we get there because it's going to be so cheap um so we're building just as a you know a trailer for future research uh agent-based mod of the energy and power sectors the agents are real energy and power firms we have a massive data set showing for the last 25 years 90% of the power plants and oil fields and everything in the world together with who owns them and and data about costs balance sheets of companies so we're putting that together to build an agent-based model where the agents are the energy and power firms every year they may or may not invest in new assets like they might drill a new oil field or they might build a solar farm and then we simulate the market every year because the landscape changes we simulate the market we find prices and quantities sold and we estimate the profits of each of the companies and we predict the future cost of Technologies we do it year by year and now we have a policy laboratory because we're going to do this for the whole globe we just got a $4.6 million NSF Grant Hallelujah and um and so we're going to do it for the whole globe for um uh you know solar wind oil gas coal and hydro so that we can really look at the entire Energy System how the pieces compete with each other and put in policies and how see how the system responds to those policies yeah yeah um don't so I'm a little bit confused your message before a little bit was that policy doesn't matter look at no these things we only care about the Technologies we could just do the Plus without paying attention to all the corruption around railroads etc etc but now you're saying that um actually the main motivator is to figure out which policy to use yeah so I didn't say policy doesn't matter I'm saying that a lot of the action is just in the technology itself but policy does matter because as I said there's a lot of fluctuations around these curves and and you could see actually quite graphically on the Austria one I showed you back here look at Austria to just to see that policy matters again there's a favorable policy here the policy gets shut off there policy gets reimplemented here so policies do matter it's just that and and all this data included lots of policy stuff in the past some of it more than others so it matters but it's not it isn't everything that's my point okay so question that and then I'll shut up um you said you are not going to assume that we have a dictator for how the world right things bummer um but uh then you also said that those policies at the level of Austria for example that when you then look at the entire world all that washes out so who would be implementing the policy that you are investigating yeah so every country has their policy actually in the United States frequently states have policies even cities can have policies and so we're going to do this as best we can given how complicated the policy landscape is but we want to build this model for several reasons one is because I want to demonstrate to the fossil fuel companies that they're about to go out of business unless they get on board quickly uh I think that will help accelerate the transition um uh but but but in general we want a policy laboratory where we can play with it so we could say let's assume Germany has this policy the US has this policy China has this policy what happens and we can now play around with policy scenarios that may be quite complicated um so hopefully that answers your question and so anyway this is the way this model works we we um basic elements are the firms and the plants we you know have an investment module The Firm some firms go out of business we update the costs we run the energy Market we go around the loop and we're already doing this for Texas power grid um we picked Texas because Texas urot as it's called you can see there's a little bit of Oklahoma in there and you're missing part of West Texas but uh it's its own power Universe it's an isolated power grid so it's we can model it and and we can we've seen that we already have a pretty decent model we can predict things about the generation mix um we can predict things about energy prices the predictions aren't perfect yet but um they're I I think it's all quite doable uh now I think that's the end of my talk thank you yes question is in your proposed model is the policy endogenous or is imposed in our proposed model you mean the one I was talking about at the end yeah it it's it's imposed the policies are at this point exogenous now my student Penny m a student I mean my student as of eight years ago Penny M who's now at the World Bank has um been um studying policies and I think there is some hope for somewhat endogenizing policies because what she did is get a big database of policies and classify the policies according to types and look at the time history of the policies so that she can actually understand which policies are feasible for which places when and so with that coupled to this we may be able to really uh go even deeper sorry I'm just any other questions Andrew do you have some intuition for the gobert's growth process where you have now the difference of the lws versus the actual difference between the so you have in the gos gr the logs right yeah logistic you just have the difference of the two yeah do you have a good like analogy for the intuition behind that of what no I I would love to have some um you know I um I it's it's a good question and I I don't I just we're thinking about it quite hard but it's a great question Sam thanks for great talk down um uh in uh you mentioned canals and monasteries yeah uh and is it I mean but you also described the inertia and so on as a technological inertia well clearly the case of canals is not primarily you're talking probably is it canals in us actually globally globally yeah that that's driven by public policy uh and and and whatever monasteries are driven by it's not technology so yeah is it just a quirk that that they also work well do that suggest there's something else going on that's making the S curve happen that shouldn't be called technology yeah probably the latter I think there is some Universal process back there um and by the way we might have a hard time distinguishing logistic from gobert's technology by technology because the data is pretty noisy but but you know on one hand yeah canals are heavily driven by policy but on the other hand there's a period where people went wow canals they started digging a lot of them at least in the developed countries and and um and then actually you can see that as canals are before they fully reached maturity railroads come in and out compete canals a lot of the legislative and you know competition was between canals and railroads um so and and of course monasteries again I I I don't know it's now we're talking about cultural diffusion and I think it's just that when you have growth processes with a limit and where where there's a possibility of some kind of replication process you know you make one Monastery and people go wow I want to join a Monas too and they go off and start another one so somehow it's infectious in some sense so I think some of the qualitative elements are probably very similar I just Sam I think it's David talking about confusion not the invention yeah this is diffusion just a fusion and I think for invention your criticism would hold right it wasn't even a criticism not a the observation yes cares monies canals integrated circuits that's a common mechanism that's a that's a mechanism of persuasion WRAL inertia here because you have the same yeah whether it's a monastery or whether it's a sh photov voltaics then clearly that analogy doesn't hold that I mean that's not exactly a criticism I'm I'm using technology in a very loose way that's that's I should have made that clear so I would call monasteries are in some sense a social techn technology um nuclear missiles are te okay that's obviously technology but canals are a technology sentences you said about people catching on and so on yeah that's a social technology yeah there's something similar going on yeah yeah yeah call technology but whatever it is I yeah we have we need some nice name for these things whatever they are that seem to behave the same way it's incredible that found that out Melanie so you should like the price of nuclear is prettyat yeah I'm wondering if that's more a policy issue or a technological issue and do you do you think that nuclear has any place in this yeah yeah so first of all it's it's not a policy issue because lots of countries have nuclear power they've had very different policies the French have been super gung-ho the Koreans um but in no case the Koreans managed to get nuclear power to come down at a rate of 1% per year French have not managed to get it come down and and in the US it's went up by a factor of three so I could attribute the factor of three to regulation but the fact that in almost exactly the same period solar PV went down by a factor of 10,000 and nuclear was at best flat tells you there's very different Technologies people are trying to understand what what is why is it one things this way and one things the other way well nuclear reactors are bespoke things they're not really manufactured although the French and the Koreans both tried to standardize them but uh they're not very modular whereas things that come down tend to be things like integrated circuits that you just print or um that you can manufacture in a highly modular way but the evidence for this is still pretty loose and so it's hard to predict AR priori from just characteristics actually the main theory that gets quoted is one that J mcer and I developed uh on and Sid uh uh where we you know we built actually on earlier work by Stuart Kaufman and Jose Lobo where they said what happens if you just it actually goes back to John MTH who was actually ironically the inventor of rational expectations so also made a dart throw model of technological change and uh so he was willing to go either hand and um so Muth made the dart throw model Jose and uh Stewart Etc improved it and we improved it another notch we improved it so that we could look at actually how is the modularity of the technology affect the rates and we were able to show that more modular technology should have faster Improvement rates so a couple of people have tested that the enough hand waving it's true but it's not highly predictive because the air bars are pretty big and modularity is hard to measure and so forth yes question is it possible to go back to where you compare the different predictions you said one prediction goes this way and the other one go yes so which prediction did you want to look at graph oh is a gra I don't remember which graph but I'll see but but describe the graph a little more because I I've got a lot of predictions on this gra yeah it had a pink and a green pink black dots pink and black dots okay let's see this one you know yeah maybe this one yeah this these ones I can't see anything what oh sorry that we're not projecting anymore what happened what I think Auto switches up well I can can remember this is it easy could could somebody switch it back on just what there a little the remote control okay it is a question about what was discussed before yeah say your question out loud and maybe eventually we'll get it we'll get we'll get the camera back yeah so but what's your question it's a it's a question of what was discussed before which is um how to how to how to combine inertia first order inertia with the S curve right which I if if I underst right well the S curve already incorporates the inertia but let me let me ask a question maybe you were you meant to ask which is there is some combination of these two things and actually one thing we're trying to do is to understand how how how do policies affect things why is it not showing the right screen it's St screen sharing it what I think it's not screen sharing anymore yeah okay it's also probably your second desktop so okay well we may just have to give up on this because so unless somebody can come fix it because I can't deal with it right now but but we are trying to understand how effective our policies and so we're collecting a lot of data to try and look you know do the do the cause effect thing what kind of policies produce what kind of effects and so there's a serious problem to do but that has to be combined with the inertial component that can I ask my question sure yeah yeah it was brilliant question that I would have asked well I knew you would have so so um you made a point I I will just describe what I saw okay yeah you made a point that this curve here's the real curve and here's the other curve and it's the difference in years was eight years yeah and so if you think of if you if you have these Technologies and it's kind of first order inertia but then you're trying to predict when they turn over saturate M yeah you have to separate out I think the inertia the first first order component of that and ask when during developmental time do I gain enough information to predict to responsibly or ACC yes no that's yeah let let me I'm happy if you want to no no go ahead finish the question right so my qu so that's the L my point is that that the question then becomes a long time when things change at what time can you can you predict l correctly right no that's one of the main things we're doing in our paper and that and if you start thinking about that you realize that L may change you may have very little information about L and L May so that becomes then a issue of of competition of the co you know the costs actually so these are it's kind of an assembly process that has to do with the first order um diffusion things but also with policy so I'm wondering if you even separate out L about eight years and maybe just say say how how can we kind of predict the future L whether that's red yeah so I should have been a little clearer we we are one of the key Innovations we've made to solve this is to actually use proper basian methods taking into account all those effects I listed and working very hard to get rid of this huge parameter bias but the question we're asking is not just let's make a prediction right now but as a function of where we are how accurate is that prediction and those aor bands that I showed you are saying right now we think that's how accurate it is if we if I had gone back and made a prediction 10 years ago the air bars would have been much wider because we're getting to the point where we can just begin to detect the deviation from the exponential 10 years ago it would have been invisible so so when the paper comes out we'll be doing exactly what you said now separating out the policy part from the from the you know deterministic part It's Tricky but but clearly when you look at the national level you can see the bumps and the curves caused by the policies and we want to understand better how to influence those bumps in the curve so the opposite reading of your talk is that all these terrible predictions we show us didn't matter because investors and scien got together and acted in a way that in effect falsified them and yet you seem to be expending a large amount of energy on making better predictions so why why because go back to what I said about Moore's Law and Ali Ray Smith it was critical that he could say in five years investors you I'm asking you for 100 million bucks finish this thing we've got all the technology in place mors law will deliver it for us in five years that's huge okay and and in fact I would the you know the fact that the predictions have consistently underpredicted deployment of Renewables you know predicted excessively high costs for the last 20 years has slowed them down if it hadn't been for that we'd have more than we have now so so I'm not saying everything is you know just pre-ordained and it's going to happen regardless of what we do at all it really does matter but but the technology and as you know and and and its usefulness and so on dictate a lot about how things are going to go then nice to see it's I mean I would I would say that probably a lot of it depends on actually articulating a mechanism because otherwise people just have to have their faith yeah yeah so that when you can refer back to some whatever segmented production cycle uptake all these things that that's the thing that convinces people yeah just don't smart but I don't know yeah so I mean you know that but I no I understand think back when we were seeing the logistic winning we thought great I can explain that you know I had a nice little story there's I wrote a draft there's two paragraphs with a mechanism but if Gumpert it's as really better as it suggests it is although right now what we're just doing right now is doing the head-to-head forecasting competition to see which one forecasts more accurately GTS is a little bit tricky because it's a stiffer equation in some sense and so um but if it's gumper it's then I'm a little puzzled know if if if it was logistic I could tell a story that's kind of like bacteria you know for instance on the consumer side the more people have it the more people see the oh your friend has it now I need to get one um but you saturate the market at some rate one minus X so it's kind of the same story as the bacteria yeah I mean I guess one thing I wonder about is how much of the underlying Innovation processes production processes kind of washes out in the noise like like what's the princip I mean yeah you know much better than I like generic principles of Technology development that have some hope of adoption a gain a foothold had you know again depending on what kind of Technology it is how does that feed back um into the cycle uh you know in in this sense transistors are kind of easy yeah um yeah whatever the physics of it was pretty well understood um but here there I I think I mean there are a lot more moving parts and to the extent that you could abstract different Industries but du Tails back yeah be interesting I don't well because it's it depends on so many other things so it's as you say the grid is completely essential what and as talks about it's complex but you've made a decision that in a way from what I hear because you're saying well the the the models are wrong and hence people are not investing so you want to have the right models so that people invest in in in which is great but it's still an opp cost it still means they might want invest in other things yeah so you're making a decision that that's what's worth investing on but it means that there are other things that could be invested by all means happen so is it NE by all means and let me just tell you the thi
2024-08-19