Learning Library

← Back to Library

AI‑Driven Materials for Climate Mitigation

Key Points

  • Stacey Gifford, an IBM Research scientist, frames her work by asking how it impacts the world, leading her to explore how AI can address the urgent challenge of climate change.
  • She emphasizes that climate change is fundamentally a chemistry problem driven by rising CO₂, and that mitigation—through new low‑carbon materials and chemistries—is the preferred strategy.
  • AI‑enabled discovery can accelerate the development of green‑chemistry processes, energy‑efficient designs (e.g., aerodynamic structures), advanced batteries, and carbon‑capture/utilization/storage materials.
  • By using AI to design and screen novel compounds and materials, researchers aim to create “sponge‑like” membranes for CO₂ capture and convert captured carbon into valuable products, directly supporting climate‑mitigation goals.
  • This “chemistry edition” of AI for climate showcases how data‑driven material science can rapidly expand the toolbox needed to limit global temperature rise to 1.5 °C.

Full Transcript

# AI‑Driven Materials for Climate Mitigation **Source:** [https://www.youtube.com/watch?v=92yV9afmc38](https://www.youtube.com/watch?v=92yV9afmc38) **Duration:** 00:11:43 ## Summary - Stacey Gifford, an IBM Research scientist, frames her work by asking how it impacts the world, leading her to explore how AI can address the urgent challenge of climate change. - She emphasizes that climate change is fundamentally a chemistry problem driven by rising CO₂, and that mitigation—through new low‑carbon materials and chemistries—is the preferred strategy. - AI‑enabled discovery can accelerate the development of green‑chemistry processes, energy‑efficient designs (e.g., aerodynamic structures), advanced batteries, and carbon‑capture/utilization/storage materials. - By using AI to design and screen novel compounds and materials, researchers aim to create “sponge‑like” membranes for CO₂ capture and convert captured carbon into valuable products, directly supporting climate‑mitigation goals. - This “chemistry edition” of AI for climate showcases how data‑driven material science can rapidly expand the toolbox needed to limit global temperature rise to 1.5 °C. ## Sections - [00:00:00](https://www.youtube.com/watch?v=92yV9afmc38&t=0s) **AI‑Driven Chemistry for Climate Solutions** - Stacey Gifford outlines how artificial intelligence can be harnessed to design new materials, positioning AI‑enhanced chemistry as a key tool in tackling climate change. - [00:03:03](https://www.youtube.com/watch?v=92yV9afmc38&t=183s) **Challenges in CO₂ Utilization Technologies** - The speaker outlines how converting captured carbon dioxide into products or storing it underground faces hurdles such as high energy penalties, material toxicity, limited stability, and elevated costs that hinder large‑scale adoption. - [00:06:11](https://www.youtube.com/watch?v=92yV9afmc38&t=371s) **AI‑Driven Literature Mining for Hypothesis Generation** - The speaker describes using natural‑language processing and knowledge‑graph technologies to ingest and interlink vast scientific publications, enabling rapid, broad‑scope hypothesis creation—such as discovering novel carbon‑capture solvents—far beyond what a single researcher could manually achieve. - [00:09:18](https://www.youtube.com/watch?v=92yV9afmc38&t=558s) **AI‑Accelerated Membrane Material Discovery** - The passage explains how AI‑guided monomer prediction, high‑throughput simulation, and automated synthesis can rapidly design, fabricate, and evaluate polymer membranes for selective CO₂ separation, vastly increasing efficiency and reducing cost. ## Full Transcript
0:00Hi, everyone. My name is Stacey Gifford, 0:02and I'm a research staff member at IBM Research, 0:05and as a scientist, I often ask myself the question of, 0:07"how is the work that I'm doing impacting the world around me?" 0:10And I think that this is an important question to ask. 0:13So today we're going to try and answer it 0:15for one of the fastest growing areas of research, 0:17which is AI, or artificial intelligence, and 0:19one of the biggest problems that we have to address today, 0:21which is climate change. 0:23So we're going to ask the question of, 0:26"how can AI help solve climate change?" 0:35And, full disclosure, I'm a chemist 0:38so we're going to focus on how we can use AI 0:42to develop new materials to help solve climate change. 0:45So we're going to call this the chemistry edition. 0:48OK, so before we dive in, 0:51let's talk a moment about the problem at hand. 0:54So most of you probably already know this, 0:56but climate change is the observed increase 1:00in global temperature that we've observed over time. 1:03And if we look at the curve of the past hundred years or so, 1:06it looks something like this. 1:08And we know that we're on track today 1:10to hit a 1.5°C increase in temperature by 2040. 1:15And that's not very far away. 1:17And while 1.5°C may not seem 1:19like a whole lot of change, we know 1:21that this is going to lead to more extreme events 1:26like hurricanes, 1:29tornadoes, floods, etc.. 1:32OK, so we need to address this problem urgently. 1:35And at its core, this is really a chemistry problem. 1:40So the reason that we're 1:42seeing this increase in global 1:43temperature is due to rising 1:44greenhouse gases 1:46and primarily carbon dioxide. 1:49Now we can solve this problem in two ways: 1:51we can either mitigate it, 1:53which means that we stop it before it happens, 1:56or we can adapt, 1:57which means we deal with it once it's here. 1:59And from a materials and chemistry 2:01perspective, really what we want to 2:03do is mitigate. 2:04We want to develop new materials and 2:05new chemistries that can help solve 2:08climate change and stop it from happening. 2:10And there's a couple of different ways we can do this. 2:12So there are what we call low carbon technologies, 2:17and this includes things like green 2:19chemistry that can be synthesized 2:21with lower carbon emissions 2:23and produce things that are more 2:24environmentally friendly. 2:25It also includes things that operate 2:28with less energy requirements. 2:30So if you think of something that's 2:31more aerodynamic, it's going to 2:32require less fuel, which means fewer 2:34carbon emissions. 2:36It's also includes things like 2:37energy storage. 2:42And so this is batteries, 2:44which are really critical for 2:45enabling renewable energy, 2:47which is an important part of how 2:48we're going to address climate change. 2:50And then finally it includes things 2:51like carbon capture, utilization, 2:53and storage technologies. 2:55And these are materials that act a 2:57bit like a sponge or membrane that 2:58separate out CO2 from either the 3:00atmosphere or point sources like 3:02power plants. 3:03And then the utilization and storage 3:05part is, once we have that carbon 3:07dioxide in hand, we can either 3:08convert it into things that are useful, 3:10like drugs and new materials, 3:11or we can store it underground 3:12where it turns into rock 3:13and it's stable for millions of years. 3:15So these technologies all exist 3:17today in some capacity, but they've 3:19been limited by a couple of 3:20different challenges, and 3:22they can kind of be broken down into 3:23a couple of key areas. 3:24So the first is performance. 3:28And one example of this is if you 3:30have a material like a solvent or 3:32a solid absorber for carbon capture, 3:34usually you have to heat it up 3:36or pressurize it to release that CO2 3:38back so you can go do something with it. 3:40And that requires fuel 3:41and it requires energy, and it means 3:43that you're actually emitting some 3:45carbon dioxide while you're capturing it. 3:47And we want to minimize that energy 3:49penalty as much as possible. 3:51And there's a lot of room for improvement today. 3:54We also have issues with toxicity. 3:58So if you think about the materials 4:00that go into batteries today, 4:01many of these are very toxic to the environment. 4:03So while we're solving one problem 4:05and helping to address renewable energy, 4:07we're creating other problems 4:08elsewhere in the system. 4:10Stability is another problem. 4:15So we know that many of these materials 4:17break down quickly over time 4:19and ultimately this 4:21all leads to an increase in cost. 4:24And in order to really drive large 4:26scale adoption of these 4:27technologies, we need to bring the 4:28cost down. 4:29And at its core, all of these things 4:31are can be addressed by 4:33designing new materials and new 4:34processes. 4:36But if you think about doing that in 4:38a traditional laboratory setting, 4:40it becomes really complicated and 4:42almost overwhelming. 4:43So if we take one problem, right, so 4:44let's talk about creating a new 4:46solvent that's better than 4:48any other solvent on the market 4:49today for carbon capture. 4:52The solvent space alone has 4:54thousands of candidates that already 4:55exist today, even if we don't 4:57discover a completely new one. 4:59And then if we start talking about 5:00solvent blends where you're mixing 5:02two or three of these together to 5:03get better performance, now you're 5:04talking about millions of different 5:05combinations. 5:06And that's not even considering the 5:08processes and the operating 5:09operating conditions that these work 5:11under. So you can improve the pH, 5:13the temperature, the pressure, 5:15the concentration of CO2, all of 5:17those things have an impact on 5:18performance. 5:19And so now you're starting to talk 5:20about millions of different 5:22experiments that have to be 5:23performed, and this becomes 5:25an incredibly hard problem to solve 5:29on an experimental level. 5:31And this is really where AI comes 5:33into the picture. 5:34So if we go back to our 5:36high school science class and we 5:38think about the sort of tried and 5:39true scientific method, it looks 5:41a little bit something like this, 5:42right? So you have a question. 5:46You have a hypothesis. 5:51You test that hypothesis with 5:53experiments. 5:54You observe the results. 5:57And then you draw some conclusions. 6:02And then we iterate through this 6:04many, many times 6:06before we really decide 6:08on some fundamental conclusions from 6:10our experiments. 6:11So this can take a long time to do, 6:13especially if you have lots of 6:14different conditions that you have 6:15to try. 6:16And we can use AI to speed this 6:18process up and make it work better. 6:20So the first place that we can use 6:23AI is in the question and 6:24hypothesis generation part. 6:26So we can use techniques like 6:28natural language processing to 6:30ingest thousands of journal articles 6:32that exist out there, which is 6:33really sort of the foundation of our 6:35scientific knowledge. 6:37And start to link entities so 6:38we can link materials and different 6:40properties in their performance 6:41metrics and other properties 6:44and build knowledge graphs. 6:50And these allow us to explore 6:52the literature in a way that's much 6:54more meaningful than if you were to 6:55have a single scientist reading 6:56through a journal article after 6:58journal article. 6:59Because if you think about it, the 7:00scientists is going out to create a 7:01new solvent for carbon capture, 7:03they're going to read the carbon 7:04capture literature first and maybe 7:05the solvent literature, but they're 7:07probably not going to explore a 7:08whole lot beyond that because 7:09there's just too much to look at. 7:11But if you have a way of ingesting 7:12all of that literature and linking 7:14it up all together, then you can 7:16really start to make insights in new 7:17areas using other materials that 7:19maybe nobody's looked at before. 7:20And that's really where the 7:21interesting and exciting hypotheses 7:23come from. 7:25We can also use query technologies 7:28to start to ask interesting 7:29questions, so we can say, "what is 7:31the best performing solvent on the 7:32market today?", and it's probably 7:34going to tell us MEA because that's 7:35what's most widely used, but there 7:37could be other candidates that get 7:39linked up through those entity 7:40recognitions to help 7:43bring to light new materials that 7:45nobody's really looked at before. 7:47So that's one area. 7:48The second area is that we can start 7:50to predict new materials. 7:58And we can do this through 7:59generative modeling. 8:02And this allows 8:04us to 8:07set boundaries and say, "we want a 8:08material that has these properties 8:10or these performance metrics", and 8:12we can start to predict new 8:14chemistries and new materials that 8:16meet those requirements. 8:18We can also use data driven 8:20techniques like machine learning to 8:23bring in large datasets. 8:25So say we have a dataset on the 8:26performance of thousands 8:28of solvents. 8:30We can start to look at how we mix 8:31this together to actually predict 8:33new mixtures that can outperform 8:34what's there today. 8:36And then we can use other techniques 8:37like quantum chemistry. 8:42And this can allow us to predict 8:43single properties of materials and 8:45how they might perform and rank them 8:47against each other. 8:50Then we can start to look at their 8:52performance so we can assess 8:54them and 8:56down select. 9:00And this is really before we even 9:01get into a chemistry lab so 9:03we can use 9:06AI-enhanced simulation 9:13to predict how 9:15different materials will perform 9:17under certain conditions. 9:18So, for example, if we're able 9:20to predict new candidates 9:22of monomers that can perform 9:24that can create a polymer 9:25that we can use as a membrane to 9:27separate CO2 from nitrogen, 9:29we can simulate that 9:31membrane and then we can look at 9:33its permeability and selectivity for 9:35carbon dioxide versus nitrogen 9:36versus NOx versus SOx versus other 9:38contaminants. 9:39And we can do this in a way that's 9:41higher throughput than we could do 9:42in a laboratory setting, and it's 9:43much cheaper because we don't have 9:44to go create all of these membranes 9:45from scratch. 9:47And this really drives up the 9:48efficiency. So we know that this 9:50increases efficiency several orders 9:52of magnitude. 9:54And it makes a big difference on our 9:56ability to discover new materials. 9:59And then finally, 10:01when we actually do want to move 10:02into the real world and start 10:04thinking about creating things, 10:06we can use AI-enhanced 10:11synthesis 10:14and automated labs 10:18to speed this up. 10:21So using enhanced synthesis 10:23means that we can 10:25predict retrosynthetic 10:27pathways 10:30to predict the best ways to make new 10:32materials. And "best" 10:34can mean a lot of different things, 10:35right? So it could be that it's 10:38the cheapest way to make it, or it's 10:40the most environmentally friendly 10:41way to make it, or it's the purest 10:42way to make it. 10:43And then once we have those 10:44conditions, we can go into an 10:45automated lab and synthesize them in 10:47a high-throughput manner and then 10:48also test their performance. 10:50And really, what this means is that 10:51we're able to both validate 10:55and we can generate data. 10:58And that data can 11:00feed back into 11:02both our knowledge graph, so that we 11:04get a better understanding of the 11:05full picture of the scientific 11:06knowledge that's out there, and also 11:08into our models so that we can be 11:09better at predicting new materials 11:11moving forward. 11:12And so hopefully what you're seeing 11:13here is that the same cycle that we 11:15use for the scientific method 11:17is replicated in the AI-enhanced 11:18version of the scientific method. 11:20And I think what we're going to see 11:21over the next couple of years is 11:22that this process is really going to 11:24accelerate discovery and we're going 11:25to see significant impact on how we 11:27address climate change with new 11:28materials. 11:29Thanks. 11:32Thank you. If you have any questions, 11:33please drop us a line below. 11:35And if you want to see more videos like this in the future, 11:37please like and subscribe. 11:38If you want to learn more about AI for climate change, 11:41please visit the links in the description.