you ask AI to show you a vision of a sustainable future where materials and energy flows are balanced this is an example of what you get I specified Boston and you can tell because if you notice on the road there are cars driving in opposite directions in the in the same path with human bikers so so we know it's authentic but what jumps out to you about this image is the greenery covering every building absorbing carbon dioxide from the atmosphere we never see this even though it's kind of consistent with every image we've ever seen about the future of sustainable cities and sustainable human life what's notably absent are solar panels solar panels which are typically vast arrays that I used to dream about as a high school student wishing to come to MIT and finding those solar panels on the side of the highways in my drive many of you probably passed those on your way here today that Vision has become reality but I'm going to tell you in a few minutes that it's actually breaking some of our sustainability systems goals we want our solar panels to be embedded in the material infrastructure and if you take nothing else home from this image is that we're actually very far from changing the way that we manage our energy and materials flows there is a huge materials burden associated with the transition to a more sustainable economy um and that brings major challenges that were poised to advance if we're going to meet the 2040 sustainable energy goals we need to extract four times the amount of material we have ever extracted on planet Earth this is ex exclusive of our steel and concrete needs this is just copper and manganese Cobalt and base metals that we're trying to use largely for our energy infrastructure our grid distribution and our Batteries Now the unfortunate reality is that in order to get those Metals out of the ground we need to build mines on this scale this is the grasberg mine it's on the order of 2,000 m across and over 500 M deep those Road cuts that you see the lines on the inside of the cavern those are driven on by tractors whose tires are three or four times as tall as I am now any of you who's ever tried to power something with a battery knows that there's no battery on planet Earth that gets you to the bottom of this mine and back out with a load of or on it you'd run dead in the first leg that beholden you to fossil fuel only fossil fuel gets you to the bottom of this pit and back out and it's that precise problem and related ones that means that all of our renewable energy goals tie us to carbon dioxide emissions we are beholden to carbon dioxide emissions until we rethink this problem and we rethink our materials demands associated with energy and energy production you can see this into intergovernmental panel for climate changes predictions for our carbon dioxide emissions that r line is the most likely scenario with the orange um drip lines around it projecting the uncertainty but what you notice is that we're actually not going to be very successful at decarbonizing and in fact meeting these two degre c curves requires that we be pulling carbon dioxide out of the atmosphere and at astronomical scale so if you note that green line that's the 2° C scenario um actually requires us to be removing on the order of 13 billion metric tons of carbon dioxide annually by 2050 now 13 billion metric tons is larger than the coal and oil and gas Industries combined and they had 200 years to do it we've got to start doing this within the next couple of decades it's an astronomically different approach to doing things in addition to decarbonizing our energy source and fortunately for you there are some great innovators today who are going to tell us a few of the strategies that we can get to do this first you're going to hear about dematerializing our energy production it's a really exciting Innovation out of blad bullit's lab so you'll finally get to hear him talk to you about his science and engineering Innovations then we're going to hear from Antoine alinor about how to more sustainably use the materials that we have already extracted let's not dig those pits or let's dig them in fundamentally different ways so that we can be more efficient about our materials processing and finally you'll hear from Rafa Gomez bombelli Who's going to tell you a little bit about how we might use advances in artificial intelligence to better design materials a priori and I'm hoping he'll tell you a little bit about the work he's been doing in carbon dioxide removal material Innovation which as I mentioned is critically needed to spawn an industry that doesn't exist within decades now how many of you have been able to see this beautiful piece of land behind the MIT Nano building okay a few of you but not enough later today when you're walking around campus go behind the MIT Nano building and you'll find this beautiful um piece of land that used to be the MIT junkyard you'd go there if you wanted to find bits of an engine which we all do we go different places for that now um this is named uh kindly by Vlad as the improbability walk it's named in honor of Millie dressle house the queen of carbon Nano science and there are two reasons for that name one is that Millie used to U Mentor her students on walks she was known as a as a great and supportive mentor and she would take them on walks during these discussions so this walk is specifically for that practice and for that reflection second improbability because Millie used to say that her journey to being MIT Professor was as improbable a journey as any other she was the daughter of Polish immigrants who grew up in the Bronx and made her way to be Institute professor at MIT one of the first winners of the national medal of Science and many others uh we were fortunate to learn from her now the reason that I this up here is that that challenge that I just described to you is improbable but it's not impossible and I'm excited today to welcome Vlad bovich Antoine alinor and Rafa Gomez bombelli to the stage to hear the next few talks which we'll discuss after thank you all right um it's indeed a big challenge to figure out how to decarbonize the planet um in so many ways one way to provide you energy is to drill a hole pull the oil out of the well and burn it and if you do that you can use every molecule that you extract and from it gain about two three four electron volts of energy per molecule once as you turn it into carbon dioxide and consequently start causing issues for the planet what we have thought of is to ask another question what if I can take that molecule I have extracted in the form of oil out of the ground and what if that molecule can over and over and over again very say molecule give me the energy I need how would that work well it would work if you take that molecule and actually incorporate it as a tin film that you do not burn but because it happens to be very dark meaning it absorbs light very efficiently it can be used inside a solar structure layer an active layer of a solar cell every molecule we right now burn in our gasoline engines could be turned into a solar molecule and if such that film of oilbased material can absorb a photon and give you one to two electron volts of energy not once not twice but millions of times and the only reason why it would stop is because there is a probability that eventually after a million times of exciting and pulling out the charge out of that molecule you will eventually break a bond and that Bond consequently will cause that particular molecule not to be quite as effective as it used to be in making the sunlight turned into electricity carbon based materials that we extract right now and we do not need to add additional ones could be turned today into carbon based solar structures and indeed if you look at those solar structures there is an opportunity to ask well what should those solar structures sit on well how about carbon how about carbon-based Fabrics that again can be extruded made out of that very same oil the same petrochemicals that you right now spend once pull them out burn them you're done with them right now today if we change the Paradigm the value of every one of those molecules goes up by a factor of a few million because every one of them can be that much more effective in actually making an active structure that could never be previously done the goal what I want to show you in the next few slides is indeed examples of doing just that here's a picture of a solar cell here is a solar cell that's in that picture it's a piece of fabric that happens to have electrodes on it active layers on it and if you apply two electrodes here and here and C you know connect it to something while Shining Light on it you will indeed generate electricity it changes the Paradigm of what a solar cell could be and as a result reshapes the way we think of how to deploy such solar cells it turns out that the cost of a solar electricity is not in the cost of the solar panel that's the cheapest part of the system what cost you two-thirds of the money sometimes more is installing that particular system on your roof wiring it ensuring it's electrically safe because all of it is done in the field by a person who needs to climb on a roof and wire or electrically connect without Burning Down The House the set of panels if all of that can be done in a factory where you start with a fabric you add to it some solar material and make it as big as you want you need not need to wire it anymore all you need to do is climb on the roof staple it to the roof and through that generate a Deployable solar cell that simply plugs into your home it could change a paradigm quite significantly the other part of it is is we have enough installers to actually install all the solar cells that we want today if I gave you all the solar cells the world needs to power itself today how long would it take us to actually install them and deploy them the answer is a very very long time simply the amount of electricians and those who need to pour concrete and put aluminum framing on your roof that's a huge amount of extra work that we simply don't have the labor for If instead I can make you the laborer because you went to Home Depot got yourself a carpet unrolled it on your roof stapled it and plugged it on the side of the house dramatically increases the number of Labor Force we have to deploy and provide the solar deployment in the way we want it but the key to it is no wiring needed on your roof the key to it is unrolling a very light format solar structure and indeed that is the vision that we aimed for the Silicon solar c as we know it has looked like it looks today since 1970s essentially since it was invented the ones that were installed on the roof of Jimmy Carter's White House looked pretty much the same as the ones that you can buy today and install on your own roof that is not progress that is simply making technology better and better by efficiency but the efficiency will not give us the cost reductions that we need it will give us a little bit we are at about 20% efficient cells we might be able to one day get 25 30% efficient cells commercially available one and a half time Improvement or you can reduce the weight 100 times and consequently avoid the wiring avoid installation cost dramatically and through that reduce the cost of what would it take to truly make an active structure that could be Deployable starting back in 2011 we worked with Professor Karen gleon in chemical engineering to use her chemically deposited calent bonded polymers conductive ones on top of pieces of paper to serve as a substrate on top of which to re evaporated carbon based molecules similar to the ones that are in your cell phones and are used for your organic LED displays oh and by the way oleds have been now around for 20 years they've been commercially utilized for over a decade they are stable they can be used for million hours it turns out halflife carbon based solar materials can be similarly stable even more stable and should be able to be used for even longer time it's easier to make a solar cell than a display and yet we make displays readily why do I say it's easier to make a solar cell well because displays everyone looks at and you want every one of your pixels to work for a solar cell if I have a pixel or two section that doesn't quite work no one notices the yields are a lot more tolerant to actual large area deployment and hence imagining ways of deploying them through high throughput manufacturing unrolling them like this or at MIT Nano here's a coating of a thin film of material these are not carbon based the ones I'm showing you print here although this can be carbon based as well these are so called per guides also an extremely inexpensive readily available material these are options right in thinking how do we redefine what solar technology looks like and we have asked questions and what would be the ultimate reduction in weight one can imagine we made back few years back an a solar cell that is so light that it can sit on top of a soap bubble six watts per gram of solar cell that's remarkable amount of energy production with one little caveat which is it's only two microns thick as a device and hence if you pick it up and you don't handle it very nicely you'll break it so we've realized well maybe that technology though can be added onto a piece of fabric after all my shirt doesn't rip very easily and if I add to it a solar cell that solar cell would be protected by the fabric underneath and yet fabric is light and breathable and so we came up with this a method by which we can peel off a three Micron substrate off of a piece of plastic why do we start with a piece of plastic I need something to grow my extremely thin solar cell that will sit on these three microns on top of and so you can imagine chemically Vapor depositing the three Micron substrate adding to it the active layers the bottom electrodes the top electrodes and then transferring that onto a piece of fabric gluing it with another four microns of glue to make an active structure that when you shine light on you will be able to get a dramatic amount of power out of it as needed after all at the end you will make a structure that weighs about 1 gram and is about 50 microns in thickness about the thickness of a human hair all for the sake of deployment being extremely rapid and you utilization of materials being extremely reduced but keeping the part that matters the part that will absorb light and give you the electricity you need thank you ask my colleague Elenor to join us next all right it's a pleasure to be here today um I'm going to talk very very quickly in 11 minutes about a adventure that started 4 4,000 years ago uh that is about Metallurgy that is about in that case copper and precious metal recycling or recovery um and I'm going to try to convey the idea that we perhaps need to change the chemical basis of our ability to recycle and extract Metals in the context that was very well presented before by desire um something you may not know if you are a producer producer or user of silver nickel manganese Cobalt Platinum um the recycling of those elements that are very you very heavily used in modern Electronics um is actually completely tied to primary production the tide might be on the price point because you're always competing with the price point of primary precious metal production but also comes in the recycling today the main facility that's going to recycle your electronic waste will end up being a copper smelter if you don't have copper smelter in your country you're going to have a hard time recycling electronic waste because the most of the value in the electronic waste are the copper and the silver now don't worry about the chart here it's just designed to be very impressive uh it's showing the amount of energy that is going on to recycle the silver that is coming from your electronic waste you can see how long it takes for the silver to come out of that supply chain and even when it comes out it usually comes out with gold a little bit of selenium and torium which means you need now to invest in another facility which economy depends on the four elements that I just mentioned now don't worry none of you are doing this most of the time you're shipping it abroad and you're waiting for somebody else to do that but of course the amount of energy and environmental impact of each of the box that you see here is actually responsible for some of the sustainability challenge that you've heard in the introduction now um in my group and many people around here at MIT including with the climate project are trying to think about using electricity with the assumptions that electricity can be made without greenhouse gases emission and with a mitigated environmental impact now if this is the case right if electricity is cleaning off and of course its price is good enough then most likely all of the materials challenges that we're talking about will be handled by electrochemistry and electrolysis now here is an example of a large scale 60 million tons per year electrolysis a manufacturing process which makes aluminum yes the aluminum you're using today is made 100% by electrolysis and yet it cost I think $2 per kilogram so we know that using electricity at large scale for materials extraction and separation is cost effective and it's supporting the economy and Society we live in now there are few technical details on that slide um is the fact that you don't want to separate the two compartments of your electrolysis reactor you may have done water electrolysis if you have not done so it's pretty straightforward to do in your own kitchen if you want but if you do water electrolysis you know you have an anodic product you see the oxygen gas coming out and then you see the famous hydrogen you know the expensive hydrogen coming on the other electrode and these two things have to happen but if you have a separator in the middle you have to pay for the energy cost of going through the separator and immediately this is not cost effective for most of the commodity metal production now there are other factors that are chemistry related there are factors related to manufacturing this needs to be a continuous process you need to constantly purchase electrical power and dissipate that electrical power as current in order to make the metal or recycled and get your metal back now there is a wisdom in electrochemistry and electrolysis at scale which says you should not do any pre-treatment to your feet stock that is something where there are discussions aluminum has the Bayer process which transforms boide into pure alumina aluminum oxide this aluminize 99.99% aluminum oxide right so that's a lot of pre-treatment that allows this technology to work how to do that is what I want to share with you today for other methods what we're doing in the lab is using Earth so geology shows us that for example if you look in the Earth crust where you have copper and the precious metals you have a natural separation between the oxide and the sulfides this separation between the oxides and the sulfides allows you to get today your ore to get copper for example this is an example here you may remember remember from high school the pyite calop pyite which is goldish and at the bottom you see the silicate which is quartz which is basically youran component so we know by Nature that if you achieve certain conditions you can separate chemically the precious and valuable Metals copper iron and sulfides versus deoxide now once you do that your entire chemistry is upside down again I'm not going to dwell too much on this but this is telling you that everything you know about ox side chemistry or waterbased chemistry is completely upside down once you do sulfide and sulfur based chemistry so that's that's the graph on the right here where you see that uranium is now very easy to make silver is much easier than copper iron is easier than copper and so on and so forth and that's what we've done in the lab in the lab we've basically developed a technique to transform any feed stock either a mix of ew or perhaps certain alloid or even minerals or waste from existing facility and we're basically substituting the sulfur with the oxygen so that we can then use this separation that I discussed before now we've done this here for the battery material you may know that the nickel manganese Cobalt batteries are oxides and there is a challenge to try to recover the lithium versus the Cobalt and the rest of it it depends on the market value and here we've shown that we can use sulfur to selectively Target the nickel and the Cobalt and leave behind the manganese and the lithium as salt that you can dissolve and we do that using sulfur and the processes that you're showing that is shown here now how do you now recover the Yellow Part which is the sulfide versus the Pink part you can use for example physical separation now suddenly we go back to the very small scale perhaps even nanoc scale interaction to figure out in what condition can you liberate the sulfides with the valuable Metals versus the non valuable oxides or sulfates once you've done that you have to go toward the metal you have to do electrolysis we have developed a unique technology molten sulfide electrolyses where no oxygen is involved in the system what you get at the anode the positive electrode is the sulfur yes the same sulfur from the previous slide which we paid for is now recovered by electrolysis while in the meantime at the cathode that's the bottom electrode that you see here we are making the liquid copper but we've also done it for other elements like ranium and um and and all of the precious metals above copper this is an illustration of the recovery of the elemental sulfur so it can of become a material circularity problem we are initially consuming sulfur to get rid of the oxygen but at some point we can regenerate some of those sulfur by electrolysis and you can recognize your your famous yellow material that is this Elemental sulfur now we heard about scaling up how to do that at bigger scale uh we've been fortunate to do that with support of the federal government here is a 5 kgr per day capacity reactor where we can do any transformation of a mix into sulfides so example of utilization include the rare earth you may have heard about the magnet big problem how we're going to get the rare earth we can separate them and sidiz them on the on the right reactor and then on the left reactor we have the electris cell where we make metals from sulfide at a scale of about 2 kg per day capacity now here is a picture of the outcome of the electrolysis at 300 G per hour scale you can see the liquid Copper at the bottom on the top you have the sulfide electrolyte and we have shown that this concept can be applied for the selective recovery of anything that is less stable than copper again this is important in the context of silver or gold recovery but also MIP denum uranium and perhaps The Rare Earth because now we can selectively design the process such that we recover first the most precious and valuable metals and then we go after the copper and then perhaps we go after the iron now we are trying at MIT in order to really demonstrate the viability of this we are trying to build a ton one ton 1,000 kilogram scale capability I think you follow the track right we start at you know 10 milligrams we do one gram we do 10 gram we do 100 grams the typical path forward is to do a startup after that you can talk to me about this experience I've done it for 15 years now we've decided that we should go to the Thousand kilogram first before we can really t and collaborate further with the with the market of Finance so with this you have my contact here and I'll be happy to stay later to answer any any questions thank you for your [Applause] attention I I this I used to have a question here but you know if you have a question as a title then the answer is no so I have my answers AI for chemistry and materials we're getting there um and I think I don't need you know Des that was that was a killer introduction I I really loved it um so I don't need to say it again sustainability and decarbonization go through new materials Technologies and you've seen it takes decades since something works in the lab it's like oh this work this idea has legs until it makes it to Market and it's used by millions of people takes approximately 20 years so when we look at the 2050 objectives they're actually the 2030 objectives because we need kind of need to lock in a lot of the science 20 years ahead of of scaling it up um and at the same time we've all experienced something that's been transformational in our day-to-day lives at sort of you know five to 10 year scale like it's been 10 years in alphago right like I don't know if you were a go player this this whole world blew up and it hasn't been 10 years it's been barely 10 years since then um and then over the last three to four we've seen image generation right we thought you needed to be sort of artistically talented Ed to produce some images turns out the distribution of images was surprisingly easy to learn right a computer could learn how an image looks like from a text prompt in a way that was completely unexpected five 10 years ago and then there's been this big big win in the Sciences with Alpha fall I mean we've seen it right two Nobel prizes in The Sciences went to AI one sort of to the AI technology in general and one to AI for protein exu prediction right and then you know if you're a scientist I am never writing a cover letter again I'm going to just ch GPT you know I've got my uh this is a high temperature super conductor an impossible thing maybe I've got my cover letter prepared uh to send to a journal if we ever discover this thing so what does this mean for materials right like you've heard it we need materials how how do we get these tools and point them to the sort of societal systemic challenges that that are ahead of of us and um this could look like batteries this could look like electrolyzers this could look like renewable Plastics this could look like decarbonization Technologies um what's different what's the same we're never going to have the same amounts of data right there is no way this the models the foundational models that we have access to today have been trained on all of the internet all of the text and all of the images that were ever taken even if they were copyrighted we we' already run out of data for for those models in The Sciences we don't have access to that scale of data and that's typically Silo you know Dow data is not going to be accessible to me uh if I work at dupon anymore right so all all this subtlety about data is very pervasive and I would postulate that we kind of know the rules of the game and and in the case of you know Professor Alor you just saw thermodynamics that's written in in in stone right like compared to that the rest of physics is stamp collecting in the words of rather for so how do we exploit that right such that when we do AI on chemistry and materials we get to exploit this sort of fundamental understanding of the rules and uh there's one more ler it's like okay somehow we need to coule our ability to evaluate hypothesis right like AI you know you ask CH GPT you know when I drop an apple do you think it will go down or up and it can give you a sense of what he thinks it will happen in the Sciences that's half of the job right coming up with a hypothesis is half of the job that happens virtually and then you need to test the hypothesis and that needs to happen in the real world right experiments need to be done data needs to be collected to inform your understanding of the world the way we traditionally do experiments is not the the most helpful way for the AI Paradigm to thrive and this is what we saw the big win Alpha fall protein structure prediction that was predicated on a carefully curated data set of 100,000 protein structures or 200,000 protein structures that have taken decades and and sort of tens of millions of dollars to collect so when the data is there AI can definitely shine but typically we need sort of these more data intensive pipelines that produce hundreds of experiments and and explore um millions of parameters search spaces and again how do we miss how do match that with the kind of conversation that Professor L was saying 100 gr you know a ton a thousand tons how do how would we interplate that with the creativity and scale of AI so I would say things that are ready to go that are happening today if you have structured Property Data Machan model has changed how we go from the structure of a material or the structure of a protein you're going to he about nanom medicines this has completely changed over the last 5 to 10 years and one of the reasons is like AI Works in that space structur property relationships are functions that machine learning can totally totally learn so given enough data what is enough data it's probably more than a thousand data points right if I have more than a thousand pairs of this x goes to this y a machine learning model can do it for molecules for crystals or or for proteins as we've seen AI can predict how to make stuffff right that's that's kind of very important too because it's not enough to say well this material could have very good properties based on the model I just train now I need to know how to make it um and making is a you know an interplay of thermodynamics and physics driven driving forces and uh recipes and uh cooking in a way that that's very datadriven so AI can go read the papers process the literature and learn how to predict the synthesis of something from precursors that are commercially available through a series of steps this can be molecules or materials there is enough evidence out there that you can have an AI process the literature and like a student learn how to propose new synthesis for a material based on the synthesis of similar materials that have made before and lately with the with the outcome of uh of large language models we've seen this orchestration of pieces this coming and going has become a lot more smooth right so we can have an AI that has a piece has an agent that reads papers I have an agent that um runs a physics based calculation to get some thermodynamic parameters that may be very important and has another agent that goes and produces a synthesis plan and maybe has an agent that interfaces an instrument and takes a measurement so we've seen these orchestration papers um and these stration examples come up as as things that AI can do for us and uh something that is very dear to me and relates to the sort of astronaut on a horse riding on the moon is that AI can propose new designs right that's called inverse design that's called generative AI um this has been demonstrated for molecules this has been demonstrated for proteins I don't know if you've heard but there is a protein design company they raised a billion dollar a series meaning that you know some people are are willing to put their money where their mouth is in in protein design and it's work to for molecular for Atomic crystals like how do we arrange atoms to make a material that does something what has not happened quite yet well AI doesn't do anything in the physical world someone needs to do something at a milligram at a gram at a kilogram at a ton scale this this hasn't happened yet yet right the connection of AI with this very meaningful technological transition hasn't occurred right AI cannot convince human beings to change what they're doing AI cannot physically act upon the world yet so if we want that to happen there's a next layer of value creation very much along the lines that that uh Professor Alor mention in their own in their own work and then AI doesn't have any context we typically train from a inputs to produce B outputs all this context of our competitors are already doing that this is you know somebody else has this patent we cannot trust the supply chain for selenium in our own processes all those things would need to be put into the model explicitly right AI cannot decide what's important AI doesn't have a context for how the world Works in general so because um I got about a minute left I would say we've done the uh the beauty of AI tools and computation is that very agnostic so these are things that can be deployed systematically across materials classes and applications so this is uh CO2 capture so can we do better than current Technologies this is this gray band is how strongly current molecules bind to CO2 this is good enough for flu gas because there is so much CO2 that you know being okay at binding is enough you capture most of it because it's very concentrated if you want to do direct a capture you need to bind more strongly than the current molecules for the economics of of carbon capture to be effective so of all the molecules an AI could dream which are the ones that kind of make sense well we need to be more strong than this gray bars well it turns out this family don't work this families work and before we make a single compound we have a sense for like what's even doable what's even possible now which of these are expensive enough well that's another layer of uh of computation I got 10 second left so I would just say this is a material class that Professor Plata and I both like zeolites they're used for cracking they're used for Doo if you drive a diesel car same game we can use AI to explore the combinatorial space of how to make them uh and with this I will thank our sponsors thanks the team and thank you so much folks for for your time right up to the stage andan and VD please join us on stage for a conversation um we have about 15 minutes I believe before the the end of the session um I like to say that there are many high privileges of being an MIT Professor um being able to train some of the best students in the world uh being able to get other MIT professors to talk to me um and of course interacting with people like you who are advancing the frontiers of Science and Technology in your day-to-day lives we are really excited to engage with you this morning we welcome your questions and the discussion um and and we'll go ahead and jump right in so I'll be keeping an eye on the audience for for any questions um from the room but I also have a few myself and um it's hard to know where to start actually you all um had such lovely context to share I think um I'm going to start at the top uh Vlad I find it interesting that you rely so heavily on carbon structures is unsurprising Right giving the focus of your work um human life also evolved to use car structures in in so many differential ways um what is your sense of this term decarbonization and how do you kind of think differently about about this and and what we actually need to be doing to refocus our efforts well U I'll start by simply saying one thing I really wish I told you at the beginning uh on the side screens there is a QR code you can scan it and you can type your questions and then the questions would also appear for us for the sessions and sessions to come pigeon hole is the method by which we can also get additional questions no literally right in front of me no I wish I have actually repeated it my apologies for not doing that to ask you a questionary and it is incredible to me to recognize that we do have the abundance of carbon and yet the only way we use it is either to eat it to make ourselves or to burn it um and in the electronics domains uh Millie dressle house was one of the Pioneers who have instructed us that if carbon takes certain forms it can actually give you electrical performance Optical performance that we couldn't have expected uh or at least we were not spending enough time on um the field of organic Electronics as first pioneered through the Practical use of A40 billion old olet industry really validated that uh there are more active electronic surfaces in a form of carbon based organic LEDs than of any other material out there silicon backplane maybe is the only other one that you can claim is there but there is more more amorphous organic being laid down than crystalline silicons has been laid down area wise and that gives us an opportunity to ask well for devices that do work on electronic principles but don't require an extreme speed is it possible to utilize carbon based materials to do what they do best which is absorb light and then from there on convert that light into slow moving slowish moving electricity but plenty fast to be able to be utilized for everyday use um I am very excited about carbon I do think that anything that's in the Earth's crust today easy to reach is the material we should touch and hence address your question on are we going to be in this Perpetual catch22 of needing to and spend a lot more energy to decarbonize the planet as we want it yes thank you um I failed to point out earlier that uh Vlad is a founder of three companies uh Antoine I think I think is on the verge or or I don't know are we allowed to say um and um but also worked in industry and for a mining industry before coming returning to MIT as did Rafa so the reason that I point that out is that these are folks who consider real constraints of of Economic and um and application space so feel free to push hard on them as you're formulating your questions because they can handle it um many professors live in the distant future we try to live in the the near-term impact of as well as imagining and ideating that future so uh to that point I want to bring up the concept of biom materials momentarily so Rafa nicely highlighted the advances that have been made in understanding predicting and describing protein folding structures you know when I think about my bones and I think about what this building is made of and my skin and what my fabric is made of again all formulations of carbon and carbon storage what is limiting manufacturer as you showed in some of your videos from being able to access the WID space of biomaterials and and all of you might have something to to add to this question so maybe Rafa I'll start with you yeah I mean I think the design space of biom materials is has been proven sort of completely expressive right there's lots of things the same building blocks can be put together to make sort of functional enzymes and collagen right so we can get mechanical properties we can get electronic properties we can get catalysis out of biomaterials right the most effective electroly um is the photosystem too it's an enzyme so I would say the language of biomaterials can express lots of function including structure the thing that's missing is that we don't have economic control of atomic Arrangements in the way nature does right the cost of making proteins the cost of making this highly engineer Nano structures the way life does is uh is just very high so as of right now you can make medicines mostly right because medicines can have sort of higher costs there is there's a lot of price that one can put into a medicine um but many other applications are limited not by the potential of biomaterials but by the cost structures and I think there is a lot of room over there to sort of industrialize um synthetic biology and biomaterials in general it's it's more I think the the potentials there the execution I think will need a will need technology that that doesn't exist today yeah thank you and could you just comment briefly on the role of robots you mentioned you know AI can't actually go and do in the physical world do you see robots as being able to tackle this challenge yes so this is this is a great question because theoretically there is no reason why armies of grad students and technicians couldn't embody AI you could imagine just like you know there was a guy moving the chess pces H when the AI suggests chess moves uh we could do that for the sciences and have armies of students doing things in The Labb turns out that's not how humans like to work right with we expect the AI Revolution for AI to do the minial tasks and us keep doing the more abstracted tasks not the other way around um and then human reproducibility you know when when you're doing sort of all these repeated laborious intensive experiments um reproducibility is an issue and in these two things robots are tireless right they they will operate consistently they won't lose motivations for instance negative data is very depressing you do experiments nothing works that's very depressing to a human being it's critical to have negative data to to train models um so in that sense uh the ability of robots to like I said work tirelessly and extremely reproducibly even if they're not faster than humans the tolerances are much much better than human reproducibility right we've seen I'm sure you've seen examples in your labs when two students you know one of them I don't know keeps the blinds open one of then keeps the blinds closed and the results don't line up for weeks until somebody realizes if everything's put into a repeatable process with a robot so I think that's what a lot of these sort of AI power science will most likely be enabled by or or cycled through Automation in the lab yeah thank you um I I want to just move uh right before we jump into the questions from the audience Antoine I wanted to ask about scale and so it's really clear to see how you know robots and AI help us accelerate Discovery and augment our own capab abilities um I I I really like the line that you used I think it was you know we go to the Earth and we learn from the earth um when we talk about planetary scale that is needed here uh just to put it into perspective people I just want to throw some numbers out there the the whole Plastics economy is around 400 million metric tons we think of that as a big huge um term that is small compared to the 51 billion metric tons of carbon dioxide emissions and that is small compared to the reservoir of carbon that is in the atmosphere on the order of 800 uh billion metric tons what you're doing with materials processing has to be done at enormous scale I just wanted to give you a moment to to comment on what the next steps are it's you know you said ton scale ton per what unit time and and what's the timeline um to getting to kind of meeting some important fraction of human material needs yeah this this is an important question so the the Paradigm that are out there for metal extraction today are based mostly on carbon hydrocarbon and gas solid reactions mostly and these things we can do them at very large scale like a blast furnace and all these high scale reactor if we're thinking about electrochemical methods we might rethink a little bit the logic of high volume because electrochemistry is a Surface phenomena and as such you would imagine developing reactors unit reactors which one of them would make make maybe 10,000 ton per year so it's like two order of magnitude smaller than the existing operation but you would multiply them very much like a solar array of of solar generation is a multiple of cell put next to each other we have to do the same thing but I think the scale has to be you know thousand times bigger than the bonus scale you can do for sale for solar generation for example so that's the Hope right now is we're discovering that in order to allow companies to be comfortable with those Technologies you need to show them like a module in a mod modle that is operating continuously for several days and weeks and months so that we can achieve what we heard about manufacturing because ultimately materials and metals production is a manufacturing process as well the metal that is coming out as very unique structure chemical composition and even physical features that if you don't achieve this you're not making steel or you're not making copper yeah yeah and I just want to highlight there these area requirements and volumetric requirements are enorm because of the surface area requirements of electrochemistry this is where nanomaterials are incredibly exciting because you can take surface large surface areas and Shrink them down you know to to the size of a coffee cup and hold them in your hand as Vlad nicely showed us so there's a unique role I think for nanomaterials to enable novel electrochemical cell deployment um and then the rest is is just multiplication so we can handle that all right um there's some very exciting uh uh questions uh coming down the line uh and and there's a hand in the room but I I want to make sure we have time to address this one that has a lot of votes um Hanuman are you in the room who submitted the would you mind reading your question sure I guess the question I guess had to do with with AI data access you had mentioned that it's currently accessed everything on the public available internet but you said there's Silo data by individual companies but uh I can see that there's a disincentivize for those companies to just all of a sudden give everybody access to their data how can we expand that access while still maintaining IP well it it's a extremely timely problem and actually for instance we've come back with Publishers Publishers used to be okay with automated access to their papers to their scientific literature and they're now holding it close to their chest so we kind of moved in the wrong direction with Publishers because they realize there's a lot of potential value that maybe they don't know how to capture but somebody else could so we've gone a little bit in the wrong direction with scientific literature over the last two years there's good examples in drag discovery particularly some some consortia in Europe where things are a little bit more toped down sometimes um and using tools um what's called homomorphic encryption which are computational tools this is an area of research of computational tools that allow you to still train machine learning models but on an encrypted data that can't be decoded back to the actual molecular structure that you're protecting right so everybody could train on sort of this Federated data sets but uh the models people don't have visibility to the actual sort of data because it's been encrypted in a way that still allows you to do machine learning but doesn't allow you to know what the what the data is so there are some examples around you know Federated learning ER consorti potentially pre-competitive research and homomorphic encryption but this is a place where sort of policy and uh economic interests of the companies are are sort of aligning right now there is no solve problem it's not just a technical problem uh and it's been worked out there is no singular answer yeah I F that back on everyone in the room to take that one home with them and it's a really it's a really great question and start to think differently remember how crazy we all thought Jeff Bezos was when he was going to ship us all our stuff for a hundred bucks a year and we thought this man's going to lose money but he got us all hooked on that right what can come by by liberating data um there's probably a lot to be had there okay so when uh I'm gonna a question to Vlad deployment cost of solar panels driven by install rather than manufacturing what part of that is physical install versus electrical safety um and as your solution mitigates the physical cost it's it's a great question I I guess I would answer it in U parabs I'm going to give you just a few examples to kind of consider one of them is the cost of a prefabricated home is half the cost of a home you built on site you can say a lot of money by asking to build yourself a prefab um why is that because of everything that needs to be done to make things just right is so much easier being done inside the factory than outside the factory result of it is that rather than installing your solar cells on your roof and wiring them on your roof pre-wiring them prior to deployment would be particularly powerful but the only problem is that today's panels weigh 25 kilos 50 pounds a piece 50 pounds is the limit of what a laborer should be asked to lift and as a result we limit ourselves to that size one by 2 m not 20 by 10 m that we could do if we could reduce the weight of the solar cell dramatically so the opportunity in redesigning the deployment would be dramatic second part is it turns out that everything in the world cost just about the same per kilo now that's a very strong statement and of course you can always come up with a way to challenge me on it I'd love to have longer discussions on it but the cost of items depends on the number of synthetic steps that you go through if you need to synthesize a very complex molecule that's going to be a very expensive one because it's going to take many synthetic steps and indeed the cost of objects scales with a number of synthetic steps avoiding the amount of matter you need to use will reduce your costs as well so combine the installation Paradigm uh when it comes to prefab versus not combine it with the reduction of cost and what's the right number to State in the overall reduction well I'll wave my hands and say it's going to be three times cheaper to install a solar cell if it's light uh compared to the way we do it right now am I correct on that number it very much depends on a whole bunch of details but I know that I'll need to use a lot less aluminum in order to reinforce my home and I'll be able to use a lot less concrete to prevent wind from blowing my solar cells away because that's the other thing we fight with the solar installation plus I'm going to open up areas we right now cannot which is Warehouse roofs that are not loadbearing and capable of carrying silicon solar cells but certainly would be available in millions of Acres quantities to take on extremely lightweight cells thank you other I saw a hand in the audience yes please andur so so is there any way to find out I so yeah I mean we're back to the sort of Nano level control that life has over matter that we can you can barely replicate so I would say that's because every cell is much smaller and energy efficient than individual pieces of a chip right like I think it's we're going back to the sort of atomic control that nature has over is that that we don't necessarily have over manufacturing I think it's it's another side of the same coin I I I guess I would you know you're asking a very deep question that could be answered in a number of different ways we certainly do not operate our brains at the speed that the chips operate their brains that we also happen to have a cooling system in the form of liquid that computers don't and have avoided it to avoid shorts uh but that would also speed us up and reduce the amount of energy needed there are solutions that indeed we are working towards and our colleagues at Lincoln laboratory later today will tell you a few stories on how to speed up and reduce the energy of computation AI requires a tremendous amount of energy because we choose to use the cloud we are choosing to spend a lot more energy than we should right because every time that you send something to the cloud and receive it back you're using so many more turnovers of chips than it would if you just happen to do all the work on your local computer your phone so there are choices we are making right now to live the lifestyle we do that correlates to the excess energy consumption we have but we also wish for Chad GPT to write our uh introductory letters and if we stopped wishing that yeah we use less energy but we'll be less productive so energy is needed for the productivity gains that Society wishes uh can we speed it up absolutely uh today's ships can be about 100 times less energy consuming as compared to what they are we just need to reinvent them from scratch and you'll hear some of those stories yeah that's a really nice um summary it's really fundamentally different from the way that we've been been thinking and using them so far um I want to uh turn so I've lost the order switched on me I I Antoine I'd like to turn to you and ask is the a question from the audience is the process for metal recycling broad enough to apply to General unnown feeds um as in a one siiz fits most or is it actually necessary to know exactly what materials you're looking to to recover in the extraction process yeah this is uh this is an important Point um again this is all driven ultimately by the Purity and the recovery of the valuable things that you get from your Stream So if you don't know what's in your stream you're not going to be able to optimize the technology now that's a manufacturing statement very well known in terms of Technology themselves we are putting these materials in conditions that are unknown I think it goes back to can we predict the thermodynamics in those conditions the answer is no and as such it is possible that certain feed talk because of the presence of minor impurities might ultimately become more favorable than other feed stock in order to get the maximum recovery and selectivity so the answer is that the principles apply generally to any fet stock that industry mining waste manufacturer or waste generator might have but of course the the the devil will be in the details and you can only get access to those details once you have enough material to be processed I want to illustrate this with impurity right if you have 1% of silver in your stream and 99% of the rest is not silver how long is it going to take us in our laboratory to show you what happened to the silver atoms how many grams of silver do I need to make until I can see them and say yes this is marketable silver for the London Metal Exchange and and and in that sense it's going to be tied back to whatever fit stock you're giving us and it is quite clear that some of them might be more commercially interesting than others even though down the line the economy of scale will make even the less profitable one at some point profitable yeah um okay so I'm gonna um pivot us for a second because I've I've heard it as a theme in a few different questions that are asked um but it's not been stated explicitly and the the statement that i' I'd like to throw on the table is that e e economics and prosperity are very tightly tied to CO2 equivalent emissions and as countries transition to being more postr excuse me um what is the word I'm looking for as prosperous thank you I haven't had enough coffee this morning so um as countries are engaging in that transition they're inevitably relying on Old Technologies not necessarily leap rocking to what we know is the best this comes out as a theme in cloud computing and the use of that for artificial intelligence calculations it comes to us as a theme in our materials um Recycling and and Recovery uh and comes to us as a theme as we think about how to meet the energy demands of those growing societies um Can Can you comment for a minute about how you imagine us having that sustained economic growth as we try to transition societies around the planet um without the carbon based emissions or equivalents associated with energy production so I guess I'll uh highlight the energy initiative work that was done a few years ago through the Tata center where the mission was can you frugally engineered technology that can be adapted by the other people in the world Beyond us living in a very comfortable environment where we are today if everyone in the world tried to live at the energy intensity of us in United States we would not be able to contain we would not be able to support 8 billion people we would be able to support only one billion people at the level intensity of consumption of energy as we have right now and so can we frugally engineer an option for the rest of the world and the answer is with that our work has been driven by coming up with a much more easily Deployable solar cell to me I think a individual there are billion of them who have no Road leading to their Village really doesn't care what's the efficiency of the solar cell that they are carrying on their back to Power their Village I think the only question they have is how many trips do I need to make to get enough energy to power my Village and there giving you more watts per kilogram me meaning giving your solar cell doesn't perform optimally like Silicon would but weighs 100 times less is a value to them that will allow them to go ahead and deploy it much more readily the challenge though of today is that still they do not have the batteries they will opt for lead acid battery in order to store that energy and hence although we have some solutions there need more we need to be
2024-12-25 15:53