00:00:00.000 --> 00:00:01.480 [TEASER] [MUSIC PLAYS UNDER DIALOGUE] ANIRUDDH VASHISTH: From the computation   00:00:01.480 --> 00:00:02.440 point of view, 00:00:02.440 --> 00:00:04.440 we always thought that if somebody gave us,   00:00:04.440 --> 00:00:06.480 like, a hundred different chemistries,   00:00:06.480 --> 00:00:10.400 we can do a bunch of simulations; tell you, like,  10 of these actually work. What we've been able   00:00:10.400 --> 00:00:15.360 to do specifically for vitrimers is that we're  able to look at the problem from the other side,   00:00:15.360 --> 00:00:20.120 and we are able to say that if you tell me  a particular application, this particular   00:00:20.120 --> 00:00:25.320 chemistry would work best for you. In essence,  what we were thinking of is that if aliens   00:00:25.320 --> 00:00:30.015 abducted all the chemists from the world, can  we actually come up with a framework? [LAUGHTER] 00:00:30.015 --> 00:00:34.680 JAKE SMITH: If all of this work is successful,  in 10 years, maybe our materials design process   00:00:34.680 --> 00:00:40.680 looks completely different, where we've gone  from this kind of brute-force screening to an   00:00:40.680 --> 00:00:45.080 approach where you start with the properties that  you care about—they're defined by the application   00:00:45.080 --> 00:00:51.760 that you have in mind—and we use this, like, “need  space” to define the material that we would like,   00:00:51.760 --> 00:00:57.160 and we can use machine learning, artificial  intelligence, in order to get us to the   00:00:57.160 --> 00:01:04.128 structure that we need to make in order  to actually achieve this design space. [TEASER ENDS] 00:01:04.141 --> 00:01:05.840 GRETCHEN HUIZINGA: You're  listening to Collaborators,   00:01:05.840 --> 00:01:10.680 a Microsoft Research Podcast showcasing the  range of expertise that goes into transforming   00:01:10.680 --> 00:01:16.319 mind-blowing ideas into world-changing  technologies. I'm Dr. Gretchen Huizinga. 00:01:28.450 --> 00:01:29.920 [MUSIC FADES] 00:01:29.920 --> 00:01:34.240 I'm thrilled to be in the booth  today, IRL, with Dr. Jake Smith,   00:01:34.240 --> 00:01:39.600 a senior researcher at Microsoft Research and  part of the Microsoft Climate Research Initiative,   00:01:39.600 --> 00:01:46.160 or MCRI. And with him is Dr. Aniruddh Vashisth.  He's an assistant professor of mechanical   00:01:46.160 --> 00:01:51.640 engineering at the University of Washington  and director of the Vashisth Research Lab.   00:01:51.640 --> 00:01:57.240 Jake and Aniruddh are working on a project that  uses machine learning to help scientists design   00:01:57.240 --> 00:02:03.200 sustainable polymers with a particularly exciting  application in the field of the ubiquitous printed   00:02:03.200 --> 00:02:08.840 circuit board, or PCB. But before we get all  sustainable, let's meet our collaborators! 00:02:08.840 --> 00:02:15.280 Jake, I'll start with you. You're a self-described  “chemist with relatively broad interests across   00:02:15.280 --> 00:02:19.840 applications” and you've done some pretty  cool things in your career. Tell us about   00:02:19.840 --> 00:02:23.600 those interests and where they've led  you, and how they've contributed to the   00:02:23.600 --> 00:02:28.668 work you're doing now in MCRI, or the  Microsoft Climate Research Initiative. 00:02:28.668 --> 00:02:32.480 JAKE SMITH: Yes. Thank you very much for  having me. So I started, like most chemists,   00:02:32.480 --> 00:02:37.560 poking things around in the lab and learning  really fundamentally about how atoms interact   00:02:37.560 --> 00:02:44.280 with one another and how this affects what we do  or what we see at our microscopic level. And so   00:02:44.280 --> 00:02:48.840 after I left grad school doing this super-basic  research, I wanted to do something more applied,   00:02:48.840 --> 00:02:53.360 and so I did a couple of postdocs, first,  looking at how we can more effectively   00:02:53.360 --> 00:02:57.680 modify proteins after we’ve synthesized them so  they might have a property that we care about   00:02:57.680 --> 00:03:02.040 and then later doing similar work on small  molecules in a more traditional drug-design   00:03:02.040 --> 00:03:07.640 sense. But after I finished that, I wound up  here at Microsoft. We were very interested   00:03:07.640 --> 00:03:13.040 in one molecule in particular, one family of  molecules, which is DNA, and we wanted to know,   00:03:13.040 --> 00:03:18.920 how do we make DNA at just gigantic scale so  that we can take that DNA and we could store   00:03:18.920 --> 00:03:24.480 digital data in it? And because DNA has this  nice property that it kind of lasts forever, ... 00:03:24.480 --> 00:03:25.349 HUIZINGA: Yeah. 00:03:25.349 --> 00:03:29.480 SMITH: … at least on our, you know,  human scale, it makes a very, you know,   00:03:29.480 --> 00:03:34.360 nice archival storage medium. So we worked on  this project for a while, and at some point,   00:03:34.360 --> 00:03:38.640 we determined we can, kind of, watch it blossom  and find the next challenge to go work on. 00:03:38.640 --> 00:03:39.429 HUIZINGA: Interesting … 00:03:39.429 --> 00:03:43.520 SMITH: And the challenge that we, you know,  wound up at I'll describe as the Microsoft   00:03:43.520 --> 00:03:48.400 Climate Research Initiative, the MCRI. We  were a group of applied scientists from,   00:03:48.400 --> 00:03:51.680 like, natural scientist backgrounds within  Microsoft, and we said, how can we make   00:03:51.680 --> 00:03:56.780 a difference for Microsoft? And the difference  that we thought was Microsoft has climate goals. 00:03:56.780 --> 00:03:57.393 HUIZINGA: Oh, yeah! 00:03:57.393 --> 00:04:00.760 SMITH: Microsoft wants to be carbon  negative, it wants to be water positive,   00:04:00.760 --> 00:04:04.400 and it wants to be zero waste.  And in order to make this happen,   00:04:04.400 --> 00:04:11.040 we need novel materials, which really  are a macroscopic view of, once again,   00:04:11.040 --> 00:04:16.016 atomic behavior. And we said, hey, we understand  atomic behavior. We're interested in this. 00:04:16.016 --> 00:04:18.116 HUIZINGA: [LAUGHS] We can help!  We’re from the government … 00:04:18.116 --> 00:04:20.960 SMITH: Yeah, maybe this is something  we could help on. Yeah. And so here we   00:04:20.960 --> 00:04:23.700 are. We wound up with Aniruddh, and  we'll go into that later, I'm sure. 00:04:23.700 --> 00:04:28.040 HUIZINGA: Yeah, yeah. So just quickly back to  the DNA thing. Was that another collaboration?   00:04:28.040 --> 00:04:31.800 I had Karin Strauss on the podcast a  while ago, and she talked about that. 00:04:31.800 --> 00:04:35.680 SMITH: Oh, absolutely. Yeah, this was with  Karin, and we had great collaborators,   00:04:35.680 --> 00:04:40.360 also at the University of Washington in  the Molecular Information Systems Lab,   00:04:40.360 --> 00:04:45.880 or MISL, who did a lot of work with us on the  practicalities of working with DNA once it's   00:04:45.880 --> 00:04:50.783 synthesized and how would you do things like  retrieve information from a big pool of DNA. 00:04:50.783 --> 00:04:52.920 HUIZINGA: Right. Right. They could …  people could go back to that podcast   00:04:52.920 --> 00:04:58.160 because she does unpack that quite a bit. Well,  Aniruddh, you describe yourself as a “trained   00:04:58.160 --> 00:05:02.680 mechanician who hangs out with chemists,”  hence your friendship with Jake here,   00:05:02.680 --> 00:05:06.280 but for your day job, you're a professor  and you have your own lab that conducts   00:05:06.280 --> 00:05:11.400 interdisciplinary research at the intersection,  as you say, of mechanics and material science.   00:05:11.400 --> 00:05:15.821 So what made you want to move to that  neighborhood, and what goes on there? 00:05:15.821 --> 00:05:20.280 ANIRUDDH VASHISTH: Yeah. Well, again, thank you so  much for having me here. I’m super excited about   00:05:20.280 --> 00:05:25.880 this. Yeah, just a little bit of background  about me. So I started off with my undergrad   00:05:25.880 --> 00:05:32.160 in civil and mechanics from IIT BHU, did a PhD  in mechanics at Penn State, and moved to Texas… 00:05:32.160 --> 00:05:35.400 HUIZINGA: Go back … go back  to, what’s the first one? 00:05:35.400 --> 00:05:39.333 VASHISTH: It’s Indian Institute of  Technology, in India, so that’s … 00:05:39.340 --> 00:05:45.520 … IIT. I did my undergrad there and  then straight away came to the US to do my PhD   00:05:45.520 --> 00:05:51.160 in mechanics at Penn State and then ended  up going to Texas, to Texas A&M University,   00:05:51.160 --> 00:05:54.880 and postdoc-ed in a chemical engineering  lab, and that's how I became, like,   00:05:54.880 --> 00:06:03.040 super familiar and fond of chemical engineers  and chemists! [LAUGHTER] And we moved to Seattle,   00:06:03.040 --> 00:06:08.440 when I got the job at University of Washington  in 2021, with my wife and my daughter. And what   00:06:08.440 --> 00:06:15.040 we do in our lab is we make and break things  now! [LAUGHS] We try to see, like, you know,   00:06:15.040 --> 00:06:18.800 when we are making and breaking these things,  we try to see them from an experimental and   00:06:18.800 --> 00:06:24.280 a simulation point of view and try to gain  some understanding of the mechanics of these   00:06:24.280 --> 00:06:29.720 different types of materials. Especially,  we are very interested in polymers. I   00:06:29.720 --> 00:06:36.280 always joke with my students and my class that  go about one day without touching a polymer,   00:06:36.800 --> 00:06:42.120 and I'm always surprised by the smiles or  the smirks that I get! But in general, like,   00:06:42.120 --> 00:06:47.760 we have been super, super excited and interested  about sustainable polymers, making sustainable   00:06:47.760 --> 00:06:53.720 composites. Particularly, we are very excited  and interested in vitrimer polymers. So let me   00:06:53.720 --> 00:06:57.984 just take, like, a step back. I'll probably  wear my professor hat straight away here. 00:06:57.984 --> 00:07:01.040 HUIZINGA: Yeah. Let's do! Let's go. [LAUGHTER] 00:07:01.040 --> 00:07:04.200 VASHISTH: And I'll tell you, just, like,  taking a step back, what are the different   00:07:04.200 --> 00:07:10.640 types of polymers. So in general, you can think  of polymers as thermosets or thermoplastics. So   00:07:10.640 --> 00:07:16.040 to Jake's point, let's just go to the molecular  scale there, and you can think of polymers as   00:07:16.040 --> 00:07:22.280 bunch of these pasta noodles which can slide over  each other, right. Or these bunch of pasta noodles   00:07:22.280 --> 00:07:27.800 which are packed together. So thermoset,  as the name suggests, it's a set network.   00:07:27.800 --> 00:07:33.720 The pasta noodles are kind of, like, set in their  place. Thermoplastics is when these pasta noodles   00:07:33.720 --> 00:07:39.800 can slide over each other. So you've probably  put too much sauce in there! [LAUGHTER] Yeah,   00:07:39.800 --> 00:07:45.000 so a good analogy there would be a lot of the  adhesives that we use are thermosets because   00:07:45.000 --> 00:07:50.280 they set after a while. Thermoplastic …  we use plastics for 3D printing a lot,   00:07:50.280 --> 00:07:54.320 so those are thermoplastics. So they're solid.  You can heat them up, you can make them flow,   00:07:54.320 --> 00:08:00.080 print something, and they solidify. Vitrimers are  very exciting because, just like thermoplastics,   00:08:00.080 --> 00:08:05.040 they have this flowability associated to  them but more at a molecular scale. Like,   00:08:05.040 --> 00:08:10.240 if you think of a single pasta noodle, it can  unclick and re-click back again. So it's like,   00:08:10.240 --> 00:08:15.180 you know, it's made up of these small LEGO blocks  that can unclick and re-click back again ... 00:08:15.180 --> 00:08:16.244 HUIZINGA: LEGO pasta … 00:08:16.244 --> 00:08:16.910 VASHISTH: LEGO pasta … HUIZINGA: I like that! [LAUGHS] 00:08:16.910 --> 00:08:21.760 VASHISTH: Exactly. So this unclicking and  re-clicking can make them re-processable,   00:08:21.760 --> 00:08:26.080 reusable, recyclable. Gives them,  like, much longer life because you   00:08:26.080 --> 00:08:31.760 can heal them. And then vitrimers basically  become the vampires of the polymer universe! 00:08:31.760 --> 00:08:34.560 HUIZINGA: Meaning they don't die? 00:08:34.560 --> 00:08:36.110 VASHISTH: Well … 00:08:36.110 --> 00:08:37.110 HUIZINGA: Or ... 00:08:37.110 --> 00:08:39.428 VASHISTH: They have like  much longer life! [LAUGHTER] 00:08:39.428 --> 00:08:43.280 SMITH: They sleep every now and  then to regenerate! Yes … [LAUGHS] 00:08:43.280 --> 00:08:46.720 HUIZINGA: Aniruddh, sticking with you for a  minute, before we get into the collaboration,   00:08:46.720 --> 00:08:52.000 let's do a quick level set on what we might call  “The Secret Life of Circuit Boards.” For this,   00:08:52.000 --> 00:08:58.000 I'd like you to channel David Attenborough  and narrate this PCB documentary. Where do   00:08:58.000 --> 00:09:02.960 we find printed circuit boards in their  natural habitat? How many species are   00:09:02.960 --> 00:09:07.480 there? What do they do during the day? How long  do they live? And what happens when they die? 00:09:07.480 --> 00:09:11.310 VASHISTH: OK, so do I have to speak like David … ? 00:09:11.310 --> 00:09:15.640 HUIZINGA: Yes, I’d appreciate it if you’d  try. [LAUGHTER] … No. Just be your voice. 00:09:15.640 --> 00:09:21.800 VASHISTH: Yeah. Yeah. So PCBs are, if you think  about it, they are everywhere. PCBs are in these   00:09:21.800 --> 00:09:28.200 laptops that we have in front of us. Probably  there are PCBs in these mics. Automobiles.   00:09:28.200 --> 00:09:33.800 Medical devices. So PCBs are, they're just,  like, everywhere. And depending upon, like,   00:09:33.800 --> 00:09:39.160 what is their end applications, they have  a composite part of it, where you have,   00:09:39.160 --> 00:09:43.160 like, some sort of a stiff inclusion in a  polymeric matrix, which is holding this part   00:09:43.160 --> 00:09:47.800 together and has bunch of electronics on top  of it. And depending on the end application,   00:09:47.800 --> 00:09:51.640 it might come in different flavors: something  that can sustain much higher temperatures;   00:09:51.640 --> 00:09:59.600 something which is flexible. Things of that sort.  And they live as long as we use the material for,   00:09:59.600 --> 00:10:03.480 like, you know, as long as we are using  these laptops or as long as we end up   00:10:03.480 --> 00:10:09.691 using our cars. And unfortunately, there is  a lot of e-waste which is created at the end. 00:10:09.710 --> 00:10:13.160 There’s been a lot of effort  in recycling and reusing these materials,   00:10:13.160 --> 00:10:14.920 but I'm confident we can do more. 00:10:14.920 --> 00:10:15.440 HUIZINGA: Right. 00:10:15.440 --> 00:10:19.972 VASHISTH: I think there's like  close to 50 million metric tons of … 00:10:19.972 --> 00:10:20.977 HUIZINGA: Wow! 00:10:20.977 --> 00:10:25.677 VASHISTH: … of e-waste which is generated—more  than that actually—every year, so … 00:10:25.677 --> 00:10:26.304 HUIZINGA: OK. 00:10:26.304 --> 00:10:28.620 VASHISTH: … a lot of scope for us to work there. 00:10:28.620 --> 00:10:33.480 HUIZINGA: Um, so right now, are they sort of  uniform? The printed circuit board? I know   00:10:33.480 --> 00:10:37.600 we're going to talk about vitrimer-based  ones, but I mean, other than that,   00:10:37.600 --> 00:10:43.480 are there already multiple materials used for  these PCBs? Jake, you can even address that. 00:10:43.480 --> 00:10:48.680 SMITH: Yeah. Of course. So there are, like, kind  of, graded ranks of circuit board materials … 00:10:48.680 --> 00:10:49.669 HUIZINGA: OK. 00:10:49.669 --> 00:10:52.800 SMITH: … that as Aniruddh said, you know,  might be for specialty applications where   00:10:52.800 --> 00:10:57.670 you need higher-temperature tolerance than normal  or you need lower noise out of your circuit board. 00:10:57.670 --> 00:10:58.170 HUIZINGA: Gotcha. 00:10:58.170 --> 00:11:01.320 SMITH: But, kind of, the bog-standard circuit  board, the green one that you think about if   00:11:01.320 --> 00:11:07.400 you've ever seen a circuit board, this is like  anti-flammability coating on a material called   00:11:07.400 --> 00:11:13.400 FR-4. So FR-4—which is an industrial name for  a class of polymers that are flame-retardant,   00:11:13.400 --> 00:11:19.799 thus FR, and 4 gives you the general  class—this is the circuit board material … 00:11:19.799 --> 00:11:20.532 HUIZINGA: OK … 00:11:20.532 --> 00:11:22.800 SMITH: … that, you know, we  really targeted with this effort. 00:11:22.800 --> 00:11:28.000 HUIZINGA: Interesting. So, Jake, let's zoom out  for a minute and talk about the big picture and   00:11:28.000 --> 00:11:32.720 why this is interesting to Microsoft  Research. I keep hearing two phrases:   00:11:32.720 --> 00:11:37.960 sustainable electronics and a circular  economy. So talk about how the one feeds   00:11:37.960 --> 00:11:42.520 into the other and what an ultimate  success story would look like here. 00:11:42.520 --> 00:11:47.520 SMITH: Yeah, absolutely. So I'll start with the  latter. When we set out to start the Microsoft   00:11:47.520 --> 00:11:54.320 Climate Research Initiative, we started with this  vision of a circular economy that would do things   00:11:54.320 --> 00:12:01.040 that avoid what we, you know, can avoid using.  But there are many cases where you can't avoid   00:12:01.040 --> 00:12:06.000 using something that is nonrenewable. And there,  what we really want to do is we want to recapture   00:12:06.000 --> 00:12:11.040 what we can't avoid. And this project, you know,  falls in the latter. There's a lot of things that   00:12:11.040 --> 00:12:18.320 fall in the latter case. So, you know, we were  looking at this at a very carbon dioxide-centric   00:12:18.320 --> 00:12:22.920 viewpoint where CO2 is ultimately the thing that  we're thinking about in the circle, although you   00:12:22.920 --> 00:12:27.840 can draw a circular economy diagram with a lot of  things in the circle. But from the CO2 viewpoint,   00:12:27.840 --> 00:12:32.880 you know, what led us to this project with  Aniruddh is we thought, we need to capture CO2,   00:12:32.880 --> 00:12:36.560 but once you capture CO2, you know, what do  you do with it? [LAUGHTER] You can pump some   00:12:36.560 --> 00:12:39.920 of it back into the ground, but this is,  you know, an economically non-productive   00:12:39.920 --> 00:12:43.060 activity. And so it's something we have  to do. It's not something we want to do. 00:12:43.060 --> 00:12:43.549 HUIZINGA: Right. 00:12:43.549 --> 00:12:49.400 SMITH: And so what could we want to do with the  CO2 that we've captured? And the thought was we   00:12:49.400 --> 00:12:54.480 do something economically viable with it. We, you  know, upcycle the CO2 into something interesting,   00:12:54.480 --> 00:12:59.360 and what we really want, and what we  still really want, is to be able to take   00:12:59.360 --> 00:13:03.612 that CO2, convert it down into a useful chemical  feedstock, and there are great laboratories … 00:13:03.612 --> 00:13:04.612 HUIZINGA: Oh, interesting … 00:13:04.612 --> 00:13:08.320 SMITH: … doing work on this, and then we could,  you know, look at our plastic design problem and   00:13:08.320 --> 00:13:13.760 say, hey, we have all this FR-4 in the world.  How could we replace the FR-4—the, you know,   00:13:13.760 --> 00:13:18.360 explicit atoms that are in the FR-4—with atoms  that have come from CO2 that we pulled out   00:13:18.360 --> 00:13:24.040 of the air? And so this is, you know, the  circular economy portion. We come down to,   00:13:24.040 --> 00:13:28.080 you know, the specific problem here.  Aniruddh talked a lot about e-waste. 00:13:28.080 --> 00:13:28.829 HUIZINGA: Yeah. 00:13:28.829 --> 00:13:30.920 SMITH: And I have great colleagues  who also collaborated with us on   00:13:30.920 --> 00:13:34.960 this project—Bichlien Nguyen, Kali  Frost—who have been doing work with   00:13:34.960 --> 00:13:39.240 our product teams here at Microsoft on,  you know, what can we do to reduce the   00:13:39.240 --> 00:13:42.220 amount of e-waste that they put out  towards Microsoft's climate goals? 00:13:42.220 --> 00:13:42.820 HUIZINGA: Right. 00:13:42.820 --> 00:13:48.200 SMITH: And Microsoft, as a producer of  consumer electronics and a consumer of,   00:13:48.200 --> 00:13:53.320 you know, industrial electronics, has a  big e-waste problem itself that we need to,   00:13:53.320 --> 00:13:59.320 you know, actually take research steps in  order to ultimately address, and so what we   00:13:59.320 --> 00:14:04.160 thought was, you know, we have this end-of-life  electronic. We can do things like desolder the   00:14:04.160 --> 00:14:07.480 components. We can recapture those ICs,  which have a lot of embedded carbon in   00:14:07.480 --> 00:14:14.280 them in the silicon that's actually there. We  can take and we can etch out the copper that   00:14:14.280 --> 00:14:19.840 has been put over this to form the traces, and  we can precipitate out that electrochemically   00:14:19.840 --> 00:14:23.280 to recapture the copper, but at the end of the  day, we’re left with this big chunk of plastic,   00:14:23.280 --> 00:14:28.200 and it's got some glass inside of it, too, for  completeness sake, and the thought was, you know,   00:14:28.200 --> 00:14:34.027 how do we do this? You can't recapture this with  FR-4. FR-4, to go back to the spaghetti thing, … 00:14:34.027 --> 00:14:35.080 HUIZINGA: Right … [LAUGHS] 00:14:35.080 --> 00:14:38.360 SMITH: … spaghetti is glued to itself. It  doesn't come apart. It rips apart if you   00:14:38.360 --> 00:14:43.040 try and take it apart. And so we wanted  to say, you know, what could we do and,   00:14:43.040 --> 00:14:47.120 you know, what could we do with Aniruddh and  his lab in order to get at this problem and   00:14:47.120 --> 00:14:53.110 to get us at a FR-4 replacement that we could  actually reach this complete circularity with. 00:14:53.110 --> 00:14:58.520 HUIZINGA: Interesting! Well, Jake, that is an  absolutely perfect segue into “how I met your   00:14:58.520 --> 00:15:04.360 mother,” which is, you know, how you all started  working together. Who thought of who first,   00:15:04.360 --> 00:15:09.240 and so on. I'm always interested to  hear both sides of the meet-up. So,   00:15:09.240 --> 00:15:13.040 Aniruddh, why don't you take the baton  from Jake right there and talk about,   00:15:13.040 --> 00:15:17.840 from your perspective, how you saw this  coming together, who approached who,   00:15:17.840 --> 00:15:22.070 what happened—and then Jake can  confirm or deny the story! [LAUGHTER] 00:15:22.070 --> 00:15:27.480 VASHISTH: Yeah, yeah. So it actually started  off, I have a fantastic colleague and a very   00:15:27.480 --> 00:15:33.080 good friend in CS department, Professor Vikram  Iyer, and he actually introduced me to Bichlien   00:15:33.080 --> 00:15:38.480 Nguyen from Microsoft, and we got a coffee  together and we were talking about vitrimers,   00:15:38.480 --> 00:15:43.920 like the work that we do in our lab, and I  had this one schematic—I forget if it was on   00:15:43.920 --> 00:15:49.400 my phone or I was carrying around one paper  in my pocket—and I showed them. I was like,   00:15:49.400 --> 00:15:55.280 you know, if we can actually do a bunch of  simulations, guide an ML model, we can create,   00:15:55.280 --> 00:16:01.120 for lack of a better word, like a ChatGPT-type of  model where instead of telling like, “This is the   00:16:01.120 --> 00:16:06.440 chemistry; tell me what the properties are,” we  can go from the other side. You can ask the model,   00:16:06.440 --> 00:16:10.880 “Hey, I want a vitrimer chemistry  which is recyclable, re-processable,   00:16:10.880 --> 00:16:15.520 that I can make airplanes out of or I can make  glasses out of. Tell me what that chemistry   00:16:15.520 --> 00:16:22.280 would look like.” And I think, you know, Bichlien  was excited about this idea, and she connected me   00:16:22.280 --> 00:16:29.003 with Jake, and I think I've been enjoying this  collaboration for the last couple of years, ... 00:16:29.003 --> 00:16:29.746 HUIZINGA: Right … 00:16:29.746 --> 00:16:30.520 VASHISTH: … working on that. 00:16:30.520 --> 00:16:37.200 HUIZINGA: Was there a paper that started the  talk, or was it just this napkin drawing? [LAUGHS] 00:16:37.200 --> 00:16:40.000 VASHISTH: I think, to give myself  a little bit of credit there,   00:16:40.000 --> 00:16:42.734 I think there was a paper  with a nice drawing on it. 00:16:42.734 --> 00:16:44.180 HUIZINGA: Right? VASHISTH: Yeah. There was a white paper. Yeah. 00:16:44.180 --> 00:16:47.800 HUIZINGA: That's good. Well, Jake,  what's your side of this story? 00:16:47.800 --> 00:16:51.390 SMITH: Ah, this is awesome! We got the  first half that I didn't know, so ... 00:16:51.390 --> 00:16:52.000 HUIZINGA: Oh—filling in gaps! 00:16:52.000 --> 00:16:57.280 SMITH: This was the Bichlien-mediated half!  [LAUGHTER] I was sharing an office with Bichlien,   00:16:57.280 --> 00:16:59.960 who apparently came up from  this meeting, and, you know,   00:16:59.960 --> 00:17:07.520 I saw the mythical paper! She put this on my desk.  And I'll plug another MCRI project that we were   00:17:07.520 --> 00:17:14.000 working on there where—or at the time—where  we were attempting to do reverse design,   00:17:14.000 --> 00:17:19.360 or inverse design, of metal organic frameworks,  which are these really interesting molecules   00:17:19.360 --> 00:17:23.427 that have the possibility to actually  serve as carbon capture absorbents, … 00:17:23.427 --> 00:17:23.927 HUIZINGA: Oh, wow. 00:17:23.927 --> 00:17:29.120 SMITH: … but the approach there was to use machine  learning to help us, you know, sample this giant   00:17:29.120 --> 00:17:32.000 space of metal organic frameworks and  find ones that had the property that we   00:17:32.000 --> 00:17:36.800 cared about. I mean, you draw this diagram  that's much like Aniruddh just described,   00:17:36.800 --> 00:17:40.680 where you've got this model that you train  and out the other side comes what you want,   00:17:40.680 --> 00:17:44.160 and so this paper came down on my desk, and I  looked at it and I said, “Hey, that's what we're   00:17:44.160 --> 00:17:49.840 doing!” [LAUGHTER] And it, kind of, you know,  went from there. We had a chat. We determined,   00:17:49.840 --> 00:17:54.017 hey, we’re both interested in, you know, this  general approach to getting to novel materials. 00:17:54.017 --> 00:17:54.521 HUIZINGA: Right. 00:17:54.521 --> 00:17:57.600 SMITH: And then, you know, we've already  talked about the synergy between our   00:17:57.600 --> 00:18:02.160 interests and Microsoft's interests and the,  you know, great work or the great particular   00:18:02.160 --> 00:18:06.940 applications that are possible with the  type of polymer work that Aniruddh does. 00:18:06.940 --> 00:18:13.360 HUIZINGA: Yeah. So the University of Washington  and Microsoft meet again. [LAUGHTER] Well, Jake,   00:18:13.360 --> 00:18:18.080 let's do another zoom out question because  I know there's more than just the Microsoft   00:18:18.080 --> 00:18:23.480 Climate Research Initiative. This project is a  perfect example of another broader initiative   00:18:23.480 --> 00:18:29.120 within Microsoft which has the potential to  quote “accelerate and enhance current research,”   00:18:29.120 --> 00:18:33.360 and that's AI for Science. So talk  about the vision behind AI for Science,   00:18:33.360 --> 00:18:38.300 and then if you have any success stories—maybe  including this one—tell us how it's working out. 00:18:38.300 --> 00:18:44.000 SMITH: Yeah, absolutely. We are—and by we, I mean  myself and my immediate colleagues—are certainly   00:18:44.000 --> 00:18:48.480 not the only ones interested in applying AI  to scientific discovery at Microsoft. And   00:18:48.480 --> 00:18:53.880 it turned out, a year or two after we started  this collaboration, a bigger organization named   00:18:53.880 --> 00:19:00.320 AI for Science arose, and we became part of it.  And it's, you know, generally a group of people   00:19:00.320 --> 00:19:04.520 who—along with our kind of sister organization  in research called Health Futures, who work more   00:19:04.520 --> 00:19:11.960 on the biology side—are interested in how AI  can help us do science in (a) a faster way,   00:19:11.960 --> 00:19:17.720 but (b) maybe a smarter, better-use-of-resources  way, or the ultimate goal, or the ultimate dream,   00:19:17.720 --> 00:19:22.240 is (c) a way that we just can't think of  doing right now. A way that, you know,   00:19:22.240 --> 00:19:27.320 it just is fundamentally incompatible with the  way that research has historically been done in,   00:19:27.320 --> 00:19:32.480 you know, small groups of grad students directed  by a professor who are themselves, you know,   00:19:32.480 --> 00:19:37.200 the actual engine behind the work that happens.  And so, the AI for Science vision, you know,   00:19:37.200 --> 00:19:41.760 it's got a couple of parts that really map very  well onto this project. The first part is we   00:19:41.760 --> 00:19:47.480 want to be able to simulate bigger systems. We  want to be able to run simulations for longer,   00:19:47.480 --> 00:19:53.840 and we want to be able to do simulations at  higher accuracy. When we get into the details of,   00:19:53.840 --> 00:19:57.040 you know, the particulars of the vitrimer  project, you'll see that one of the fundamental   00:19:57.040 --> 00:20:03.000 blocks here is the ability to run simulations,  and Aniruddh’s excellent grad student Yiwen,   00:20:03.000 --> 00:20:10.440 you know, spent a ton of time trying to identify  the appropriate simulation parameters in order to   00:20:10.440 --> 00:20:15.200 capture the behavior that we care about here. And  so, the first AI for Science vision says we don't   00:20:15.200 --> 00:20:19.280 need Yiwen to do that, you know, we're going to  have a drop-in solution or we're going to have,   00:20:19.280 --> 00:20:23.440 you know, a set of drop-in solutions that  can, you know, take this work away from you   00:20:23.440 --> 00:20:27.520 and make it much easier for you to go straight  to running the simulations that you care about. 00:20:27.520 --> 00:20:32.720 HUIZINGA: Yeah. A couple questions. Not  on the list here, but you prompted them.   00:20:32.720 --> 00:20:38.280 No pun intended. Are these specialized models  with the kinds of information … I mean, if I   00:20:38.280 --> 00:20:44.280 go to ChatGPT and ask it to do what you guys are  doing, I'm not going to get the same return am I? 00:20:44.280 --> 00:20:45.000 SMITH: Absolutely. 00:20:45.000 --> 00:20:45.880 HUIZINGA: Am I? 00:20:45.880 --> 00:20:51.160 SMITH: Oh, no, no, no, no! [LAUGHTER] I was  saying you were absolutely correct. [LAUGHS] You   00:20:51.160 --> 00:20:57.000 can ask ChatGPT, and it will tell you all sorts of  things that are very interesting. It can tell you,   00:20:57.000 --> 00:20:59.880 probably, a vitrimer. It could give you  Aniruddh’s spiel about the spaghetti,   00:20:59.880 --> 00:21:05.200 I'm sure, if you prompted it in the correct  way. But what it can't tell you is, you know,   00:21:05.200 --> 00:21:08.840 “Hey, I have this particular vitrimer composition,   00:21:08.840 --> 00:21:13.240 and I would like to know at what temperature  it's going to melt when I heat it up.” 00:21:13.240 --> 00:21:17.280 HUIZINGA: Right. OK, so I have one  more question. You talk about the   00:21:17.280 --> 00:21:20.847 simulations. Those take a lot of  compute. Am I right? Am I right? 00:21:20.847 --> 00:21:21.980 SMITH: You're absolutely right. 00:21:21.980 --> 00:21:23.070 VASHISTH: Yeah. 00:21:23.070 --> 00:21:26.760 HUIZINGA: So is that something that Microsoft  brings to the party in terms of … I mean,   00:21:26.760 --> 00:21:32.340 does the University of Washington have the same  access to that compute, or what's the deal? 00:21:32.340 --> 00:21:37.520 VASHISTH: I think especially on the scale,  we were super happy and excited that we   00:21:37.520 --> 00:21:41.360 were collaborating with Microsoft. I  think one of these simulations took,   00:21:41.360 --> 00:21:46.160 like, close to a couple of weeks, and  we ended up doing, I would say, like,   00:21:46.160 --> 00:21:52.440 close to more than 30,000 simulations. So that's  a lot of compute time if you think about it. 00:21:52.440 --> 00:21:54.080 HUIZINGA: To put that in perspective,   00:21:54.080 --> 00:21:58.987 how long would it take a human  to do those simulations? [LAUGHS] 00:21:58.987 --> 00:22:03.480 SMITH: [LAUGHS] Oh, man, to try and  actually, like, go do all this in the lab … 00:22:03.480 --> 00:22:09.000 SMITH: First, you got to make these 30,000, like,  starting materials. This in itself … let's say  00:22:09.000 --> 00:22:13.640 you could buy those. Then to actually run the  experiments, how long does it take to do one … 00:22:13.640 --> 00:22:14.800 HUIZINGA: And how much money? 00:22:14.800 --> 00:22:17.669 VASHISTH: That's … that's like you're  talking about like one PhD student there. 00:22:17.669 --> 00:22:18.169 HUIZINGA: Right? 00:22:18.169 --> 00:22:20.640 VASHISTH: That’s like, you know,  it takes like a couple of years   00:22:20.640 --> 00:22:26.334 just to synthesize something properly  and then characterize it, and it's … 00:22:26.350 --> 00:22:30.227 Yeah, no, I think the virtual world does have some pluses to it. 00:22:30.227 --> 00:22:34.600 HUIZINGA: So this is a really  good argument for AI for Science,   00:22:34.600 --> 00:22:38.800 meaning the things that it can do,  artificial intelligence can do,   00:22:38.800 --> 00:22:42.280 at a scale that's much smaller than  what it would take a human to do. 00:22:42.280 --> 00:22:46.520 SMITH: Yeah, absolutely. And I'll plug  the other big benefit now, which is, hey,   00:22:46.520 --> 00:22:50.880 we can run simulations. This is fantastic.  But the other thing that I think all of   00:22:50.880 --> 00:22:55.052 us really hope AI can do is it can help  us determine which simulations to run … 00:22:55.052 --> 00:22:56.840 HUIZINGA: Ooh … SMITH: … so we need less compute overall, 00:22:56.840 --> 00:22:59.780 we need less experiments if we have to  go do the experiments, and this is … 00:22:59.780 --> 00:23:01.080 HUIZINGA: So it’s the winnowing process. 00:23:01.080 --> 00:23:01.720 SMITH: Exactly. 00:23:01.720 --> 00:23:03.680 HUIZINGA: OK. That's actually really interesting. 00:23:03.680 --> 00:23:05.280 SMITH: And this is, like, the second,   00:23:05.280 --> 00:23:09.100 or maybe even the largest, vector  for acceleration that we could see. 00:23:09.100 --> 00:23:15.760 HUIZINGA: Cool. Well, every show I ask, what could  possibly go wrong if you got everything right?   00:23:15.760 --> 00:23:21.120 And, Aniruddh, I want to call this the “Defense  Against the Dark Arts” question for you. You're   00:23:21.120 --> 00:23:27.040 using generative AI to propose what you call  novel chemistries, which can sound really cool   00:23:27.040 --> 00:23:33.040 or really scary, depending on how you look at it.  But you can't just take advice from a chatbot and   00:23:33.040 --> 00:23:37.800 apply it directly to aerospace. You have to  kind of go through some processes before. So   00:23:37.800 --> 00:23:42.800 what role do people, particularly experts in  other disciplines, play here, and what other   00:23:42.800 --> 00:23:47.840 things do you need to be mindful of to ensure  the outputs you get from this research are valid? 00:23:47.840 --> 00:23:53.200 VASHISTH: Yeah, yeah. That's a fantastic question.  And I'll actually piggyback on what Jake just said   00:23:53.200 --> 00:24:00.520 here, about Yiwen Zheng, who's like a fantastic  graduate student that we have in our lab. He   00:24:00.520 --> 00:24:05.800 figured out how to run these simulations at the  first point. It was like six months of … like,   00:24:05.800 --> 00:24:11.360 really long ordeal. How to make sure that in  the virtual world, we are synthesizing these   00:24:11.360 --> 00:24:17.800 polymers correctly and we are testing them  correctly. So that human touch is essential,   00:24:17.800 --> 00:24:24.080 I feel like, at every step of this research,  not just like doing virtual characterization   00:24:24.080 --> 00:24:29.120 or virtual synthesis of these materials,  training the models, but eventually,   00:24:29.120 --> 00:24:33.680 when you train the models also and the  model tells you that, well, these are, like,   00:24:33.680 --> 00:24:39.640 the 10 best polymers that would work out, there  you need people like Jake who are like chemists,   00:24:39.640 --> 00:24:42.440 you know. They come in [LAUGHTER] and  they're like, hey, you know what? Like,   00:24:42.440 --> 00:24:46.920 out of these 10 chemistries, this one you  can actually synthesize. It's a one-step   00:24:46.920 --> 00:24:53.560 reaction or things of that sort. So we have  a chemist in our lab also, Dr. Agni Biswal,   00:24:53.560 --> 00:24:58.640 who’s a postdoc. So we actually show him all  these chemistries, apart from Jake and Bichlien.   00:24:58.640 --> 00:25:02.960 We show the chemistries to all the chemists  and say, like, OK, what do you think about   00:25:02.960 --> 00:25:08.273 this? How do these look like? Are they totally  insane, or can we actually make them? [LAUGHTER] 00:25:08.273 --> 00:25:11.460 SMITH: Yeah, we still need that, like, human  evaluation step at the end, at this point. 00:25:11.460 --> 00:25:12.440 HUIZINGA: Yeah … VASHISTH: Exactly. 00:25:12.440 --> 00:25:16.240 HUIZINGA: Ask a chemist! Well, and  I would imagine it would be further   00:25:16.240 --> 00:25:20.240 than just “this would be the best one”  or something like “you better not do   00:25:20.240 --> 00:25:26.640 that one.” Are there ever like crazy  responses or replies from the model? 00:25:26.640 --> 00:25:31.480 SMITH: [LAUGHS] It's fascinating. Models are very  good—and particularly we'll talk about models that   00:25:31.480 --> 00:25:36.800 generate small organic structures—at generating  things that look reasonable. They follow all the   00:25:36.800 --> 00:25:41.880 rules. But there's this next step beyond that. And  you see this when you talk to people who've worked   00:25:41.880 --> 00:25:46.120 in med chem for, you know, 30 years of their life.  Well, they’ll look at a structure and they'll,   00:25:46.120 --> 00:25:51.720 like, get this gut feeling like, you know, a storm  is coming in and their knee hurts, and they really   00:25:51.720 --> 00:25:55.440 don't like that molecule. [LAUGHTER] And if you  push them a little bit, you know, sometimes they   00:25:55.440 --> 00:25:59.240 can figure out why. They'll be like, oh, I worked  on, you know, a molecule that looked like that 20   00:25:59.240 --> 00:26:02.840 years ago, and it, you know, turned out to  have this toxicity, and so I don't want to   00:26:02.840 --> 00:26:07.257 touch that again. But oftentimes, people can't  even tell you. They’ve just got this instinct … 00:26:07.257 --> 00:26:10.400 … that they've built up, and trying to, you know,   00:26:10.400 --> 00:26:15.720 capture that intuition is a really interesting  next frontier for this sort of research. 00:26:15.720 --> 00:26:20.320 HUIZINGA: Wow. You know, you guys are just  making my brain fry because it's like so many   00:26:20.320 --> 00:26:23.920 other questions I want to ask, but we're actually  getting there to some of them, and I'm hoping   00:26:23.920 --> 00:26:29.200 we'll address those questions with the other  things I have. So, Jake, I want to come … Well,   00:26:29.200 --> 00:26:35.040 first of all, Aniruddh, have you finished  your defense against the dark arts? [LAUGHS] 00:26:35.040 --> 00:26:39.360 VASHISTH: I think I can point out one more  thing very quickly there, and as Jake said,   00:26:39.360 --> 00:26:43.000 like, we are learning a lot, particularly  about these materials, like, the vitrimer   00:26:43.000 --> 00:26:47.960 materials. These are new chemistries, and we  are still learning about, like, the mechanical,   00:26:47.960 --> 00:26:54.560 thermorheological properties; how to handle  these materials. So I think there's a lot that   00:26:54.560 --> 00:27:00.680 we don't know right now. So it's like a bunch  of, like, unknown unknowns that are there. So … 00:27:00.680 --> 00:27:05.320 HUIZINGA: Well, and that's research,  right? The unknown unknowns. Jake,   00:27:05.320 --> 00:27:11.080 I want to come back to the vision of the climate  research initiative for a minute. One goal is to   00:27:11.080 --> 00:27:17.720 develop technologies that reduce the raw tonnage  of e-waste, obviously. But if we're honest,   00:27:17.720 --> 00:27:22.520 advances in technology have almost encouraged  us to throw stuff away. It's like before it   00:27:22.520 --> 00:27:27.880 even wears out. And I think we talked earlier  about, you know, this will last as long as my   00:27:27.880 --> 00:27:32.840 car lasts or whatever, but I don't like my  car in five years. I want a different one,   00:27:32.840 --> 00:27:38.000 right? So I wonder if you've given any thought  to what things, in addition to the work on   00:27:38.000 --> 00:27:44.840 reusable and recyclable components, we might do  to reverse engineer the larger throwaway culture? 00:27:44.840 --> 00:27:47.440 SMITH: This was interesting. I feel like this gets   00:27:47.440 --> 00:27:53.835 into real questions about social  psychology and our own behaviors … 00:27:53.835 --> 00:27:58.560 … with individual things. Why do I have  this can of carbonated water here when I could   00:27:58.560 --> 00:28:05.181 have a glass of carbonated water? But I want  to, kind of, completely sidestep that because … 00:28:05.181 --> 00:28:06.960 HUIZINGA: Yeah … Well, we know  why! Because it's convenient,   00:28:06.960 --> 00:28:09.520 and you can take it in your car and not spill. 00:28:09.520 --> 00:28:14.480 SMITH: Agreed. Yes. All right. [LAUGHTER] I also  have this cup, and it could not spill, as well. 00:28:14.480 --> 00:28:15.780 HUIZINGA: True! Recyclable—reusable. 00:28:15.780 --> 00:28:19.640 SMITH: Ahhh … no, no … this is like  a—it's an ingrained consumer behavior   00:28:19.640 --> 00:28:25.000 that I've developed that might … I’ll slip  into “Jake's Personal Perspectives” here,   00:28:25.000 --> 00:28:32.720 which is that it should not be on the individual  consumer behavior changes to ultimately drive a   00:28:32.720 --> 00:28:38.000 shift towards reusable and recyclable  things. And so one of the fundamental,   00:28:38.000 --> 00:28:42.200 like, hypotheses that we had with the,  you know, design of the projects we put   00:28:42.200 --> 00:28:48.360 together with the MCRI was that if we put  appropriate economic incentives in place,   00:28:48.360 --> 00:28:53.800 then we can naturally guide behavior at a much  bigger scale than the individual consumer. And   00:28:53.800 --> 00:28:58.520 maybe we'll see that trickle down to the consumer.  Or maybe this means that the actual actors,   00:28:58.520 --> 00:29:02.855 the large-scale actors, then have the  economic incentive to follow it themselves. 00:29:02.855 --> 00:29:03.355 HUIZINGA: Right. 00:29:03.355 --> 00:29:08.040 SMITH: And so with the e-waste question in  particular, we talked a lot about FR-4 and,   00:29:08.040 --> 00:29:09.920 you know, it's the part of the circuit board   00:29:09.920 --> 00:29:12.353 that you're left over with at the end  that there's just nothing to do with … 00:29:12.353 --> 00:29:13.069 HUIZINGA: Right. 00:29:13.069 --> 00:29:17.640 SMITH: … and so you toss into landfill, you  burn it, you do something like this. But,   00:29:17.640 --> 00:29:22.200 you know, with a project like this, where our  goal was to take that material and now make   00:29:22.200 --> 00:29:27.660 it reusable, we can add this actual  economic value to the waste there. 00:29:27.660 --> 00:29:33.760 HUIZINGA: Yeah. I realized even as I asked that  question, that I had the answer embedded in the   00:29:33.760 --> 00:29:38.387 question because, in part, how we design  technologies drives how people use things. 00:29:38.387 --> 00:29:38.990 SMITH: Yeah, absolutely. VASHISTH: Yeah. 00:29:38.990 --> 00:29:41.400 HUIZINGA: And usually, the drivers are convenience   00:29:41.400 --> 00:29:50.320 and economics. So if upstream of consumer  … consumption? [LAUGHTER] Upstream of that,   00:29:50.320 --> 00:29:58.360 the design drives environmental health and so  on, that's actually … that's up to you guys! So   00:29:58.360 --> 00:30:04.800 let's get out of this booth and get back to  work! [LAUGHTER] Well, Jake, to that point,   00:30:04.800 --> 00:30:09.520 talk about the economics. We talk about a  circular economy. And I know that recycling   00:30:09.520 --> 00:30:15.280 is expensive. Can you talk a little bit about how  that could be impacted by work that you guys do? 00:30:15.280 --> 00:30:20.506 SMITH: Recycling absolutely is expensive  relative to landfilling or a similar alternative. 00:30:20.513 --> 00:30:24.280 One of the things that makes us  target e-waste is that there are things   00:30:24.280 --> 00:30:29.600 of value in e-waste that are, like, innately  valuable. When you go recollect that copper or   00:30:29.600 --> 00:30:32.800 the gold that you've put into this, when  you recollect the integrated circuits,   00:30:32.800 --> 00:30:38.000 you know, they had value, and so a lot of  the economic drive is already there to get   00:30:38.000 --> 00:30:42.200 you to the point where you have these circuit  boards. And then, you know, the question was,   00:30:42.200 --> 00:30:46.720 how do we get that next bit of economic  value so that you've taken steps this far,   00:30:46.720 --> 00:30:51.160 you have this pile of circuit boards, so  you've already been incentivized to get to   00:30:51.160 --> 00:30:57.040 here and it will be easy to make this—even if  it's not a completely economically productive   00:30:57.040 --> 00:31:02.760 material—versus synthesizing a circuit board  from virgin plastic, but it's offset enough.   00:31:02.760 --> 00:31:08.503 We've taken enough of that penalty for reuse  out that it can be justifiable to go do. 00:31:08.503 --> 00:31:11.560 HUIZINGA: Right. OK. So talk—again,  off script a little bit—but talk a   00:31:11.560 --> 00:31:15.700 little bit about how vitrimers  help take it to the last mile. 00:31:15.700 --> 00:31:21.000 VASHISTH: Yeah, I think the inherent  property of the polymer to kind of unclick   00:31:21.000 --> 00:31:25.720 and re-click back again, the heal-ability of  the polymer, that's something that, kind of,   00:31:25.720 --> 00:31:32.320 drives this reusability and re-processability of  the material. I'll just, like, point out, like,   00:31:32.320 --> 00:31:37.920 you know, particularly to the PCB case, where we  recently published a collaborative paper where we   00:31:37.920 --> 00:31:43.920 showed that we can actually make PCB boards using  vitrimers. We can unassemble everything. We can   00:31:43.920 --> 00:31:49.080 take out the electronics, and even the composite,  the glass fiber and the polymer composite,   00:31:49.080 --> 00:31:54.521 we can actually separate that, as well, which  is, in my mind, like, a pretty big success. 00:31:54.521 --> 00:31:55.021 HUIZINGA: Yeah. 00:31:55.021 --> 00:31:58.080 VASHISTH: And then we can actually put  everything back together and remake   00:31:58.080 --> 00:32:01.910 a PCB board, and, you know,  keep on doing that. So … 00:32:01.910 --> 00:32:04.200 HUIZINGA: OK, so you had talked to me before about   00:32:04.200 --> 00:32:08.660 “Ring Around the Rosie” and the  hands and the feet. Can you … ? 00:32:08.660 --> 00:32:09.780 SMITH: [LAUGHS] His favorite analogy! 00:32:09.780 --> 00:32:13.240 HUIZINGA: Do that one just for  our audience because it's good. 00:32:13.240 --> 00:32:18.160 VASHISTH: OK. So I'll talk a little bit  about thermoset/thermoplastic again,   00:32:18.160 --> 00:32:21.700 and then I'll just give you a  much broader perspective there. 00:32:22.100 --> 00:32:27.040 VASHISTH: So the FR-4 PCBs that are made, they  are usually made with thermosetting polymers.   00:32:27.040 --> 00:32:31.800 So if you think about thermosetting polymers,  just think of kids playing “Ring of Roses,”   00:32:31.800 --> 00:32:36.960 right? Like their hands are fixed and their  feet are fixed. Once the network is formed,   00:32:36.960 --> 00:32:42.160 there's no way you can actually destroy that  network. The nice thing about vitrimers is   00:32:42.160 --> 00:32:46.480 that when you provide an external stimulus,  like, just think about these kids playing   00:32:46.480 --> 00:32:52.000 “Ring of Roses” again. Their feet can move and  their handshakes can change, but the number of   00:32:52.000 --> 00:32:56.857 handshakes remain the same. So the polymer is kind  of, like, unclicking and re-clicking back again. 00:32:56.870 --> 00:33:00.800 VASHISTH: And if you can cleverly use  this mechanism, you can actually recycle,   00:33:00.800 --> 00:33:06.240 reprocess the polymer itself. But what we showed,  particularly for the PCB paper, was that you can   00:33:06.240 --> 00:33:12.546 actually separate all the other constituents  that are associated with this composite, yeah. 00:33:12.546 --> 00:33:16.600 HUIZINGA: OK. That's … I love that. Well,  sticking with you for a second, Aniruddh,   00:33:17.360 --> 00:33:23.960 talking about mechanical reality—not just chemical  reality, but mechanical reality—even the best   00:33:23.960 --> 00:33:30.720 composites wear out, from wear and tear. Talk  about the goal of this work on novel polymers   00:33:30.720 --> 00:33:36.980 from an engineering perspective. How do you  think about designing for reality in this way? 00:33:36.980 --> 00:33:43.200 VASHISTH: Yeah, yeah. That's a fantastic  question. So we were really motivated by what   00:33:43.200 --> 00:33:49.320 type of mechanical or thermal loadings materials  see in day-to-day life. You know, I sit in my car,   00:33:49.320 --> 00:33:54.360 I drive it, it drives over the road, there is  some fatigue loadings, there's dynamic loading,   00:33:54.360 --> 00:33:59.160 and that dynamic loading actually leads  to some mechanical flaws in the material,   00:33:59.160 --> 00:34:05.760 which damages it. And the thought was always  that, can we restrict that flaw, or can we go   00:34:05.760 --> 00:34:10.880 a step further? Can we actually reverse that  damage in these composites? And that's where,   00:34:10.880 --> 00:34:13.920 you know, that unclicking/re-clicking  behavior of vitrimer becomes, like,   00:34:13.920 --> 00:34:19.200 really powerful. So actually, the first work  that we did on these type of materials was that   00:34:19.200 --> 00:34:24.000 we took a vitrimer composite and we applied  fatigue loading on it, cyclic loading on it,   00:34:24.000 --> 00:34:29.720 mechanical loading. And then we saw that when  there was enough damage accumulated in the system,   00:34:29.720 --> 00:34:34.400 we healed the system. And then we did this again.  And we were able to do it again and again until   00:34:34.400 --> 00:34:41.200 I was like, I've spent too much money on this  test frame! [LAUGHS] But it was really exciting   00:34:41.200 --> 00:34:45.680 because for a particular loading case that  we were looking at, traditional composites   00:34:45.680 --> 00:34:52.840 were able to sustain that for 10,000 cycles, but  for vitrimers, if we did periodic healing in the   00:34:52.840 --> 00:34:57.640 material, we were able to go up to a million  cycles. So I think that's really powerful. 00:34:57.640 --> 00:34:58.200 HUIZINGA: Orders of magnitude. 00:34:58.200 --> 00:34:59.320 VASHISTH: Yeah, exactly. 00:34:59.320 --> 00:35:05.160 HUIZINGA: Wow. Jake, I want to broaden the  conversation right now, beyond just you and   00:35:05.160 --> 00:35:11.160 Aniruddh, and talk about the larger teams you  need to assemble to ensure success of projects   00:35:11.160 --> 00:35:16.400 like this. Do you have any stories you could  share about how you go about building a team? You   00:35:16.400 --> 00:35:20.680 kind of alluded to it at the beginning. There's  sort of a pickup basketball metaphor there. Hey,   00:35:20.680 --> 00:35:27.640 he's doing that. We're doing this. But you have  some intentionality about people you bring in. So   00:35:28.240 --> 00:35:31.740 what strengths do each institution  bring, and how do you build a team? 00:35:31.740 --> 00:35:35.920 SMITH: Yeah, absolutely. We've tried a bunch  of these collaborations, and we've definitely   00:35:35.920 --> 00:35:40.000 got some learnings about which ones work better  than others. This has been a super productive   00:35:40.000 --> 00:35:45.080 one. I think it's because it has that right mix of  skills and the right mix of things that each side   00:35:45.080 --> 00:35:51.040 are bringing. So what we want from a Microsoft  side for a successful collaboration is we want a   00:35:51.040 --> 00:35:56.880 collaborator who is really a domain expert in,  you know, something that we don't necessarily   00:35:56.880 --> 00:36:03.760 understand but who can tell us, in great detail,  these are the actual design criteria; these are,   00:36:03.760 --> 00:36:08.840 you know, where I run into trouble with my  traditional research; this is the area that,   00:36:08.840 --> 00:36:13.640 you know, I'd like to do faster, but I don't  necessarily know how. And this was the critical   00:36:13.640 --> 00:36:20.440 part, I think, you know, from the get-go. They  need to, themselves, be an extremely, you know,   00:36:20.440 --> 00:36:24.720 capable subject matter expert. Otherwise,  we're just kind of chatting. We don't have   00:36:24.720 --> 00:36:29.960 anyone that really knows what the problem truly  is and you make no progress or you … worse,   00:36:29.960 --> 00:36:34.504 you spend a whole lot of resources to  make “progress”—I'm doing air quotes ... 00:36:34.504 --> 00:36:36.200 HUIZINGA: Yeah. I love air quotes on a podcast! 00:36:36.200 --> 00:36:40.880 SMITH: [LAUGHS]—that is actually just completely  tangential to what the field needs or what the   00:36:40.880 --> 00:36:46.960 actual device needs. So this was, you know, the  fundamental ingredient. And then on top of that,   00:36:46.960 --> 00:36:51.570 we need to find a problem that's of  joint interest where, in particular, … 00:36:51.589 --> 00:36:55.800 SMITH: … computation can help. You talked about  the amount of computation that we have at our   00:36:55.800 --> 00:37:00.160 disposal as researchers at Microsoft, which is  a tremendous strength. And so we want to be able   00:37:00.160 --> 00:37:05.800 to leverage that. And so for a collaboration like  this, where running a large number of simulations   00:37:05.800 --> 00:37:10.720 was a fundamental ingredient to doing it, this  was, you know, a really good fit, that we could   00:37:10.720 --> 00:37:16.743 come in and we could enable them to have more  data to train the models that we build together. 00:37:16.743 --> 00:37:19.480 HUIZINGA: Mm-hm. Well, as researchers,  are you each kind of always scanning   00:37:19.480 --> 00:37:24.720 the horizon for who else is doing things  in your field that—or tangential to your   00:37:24.720 --> 00:37:28.920 field but necessary? How does that  work for recruiting, I would say? 00:37:28.920 --> 00:37:33.480 VASHISTH: Yeah, that's a good question. I think  … I mean, that's kind of like the job, right.   00:37:33.480 --> 00:37:39.760 For the machine learning work we did, we saw a  lot of inspiration from biology, where people   00:37:39.760 --> 00:37:44.720 have been designing biomolecules. The challenges  are different for us. Like we are designing much   00:37:44.720 --> 00:37:50.360 larger chains, but we saw some inspiration from  there. So always, like, looking out for, like,   00:37:50.360 --> 00:37:55.120 who is doing what is super helpful, and it leads  to, like, really nice collaborations, as well.   00:37:55.640 --> 00:38:00.600 We've had, like, really fruitful collaborations  with the professor Sid Kumar at TU Delft,   00:38:00.600 --> 00:38:05.600 and we always get his wisdom on some of these  things, as well. But yeah, recruiting students   00:38:05.600 --> 00:38:11.070 also becomes, like, very interesting and how,  like, people who can help us achieve our idea … 00:38:11.070 --> 00:38:16.280 HUIZINGA: Yeah. Jake, what's your take on it  from the other seat? I mean, do you look actively   00:38:16.280 --> 00:38:24.313 at universities around the world—and even in  your backyard—to … like U Dub … ? [LAUGHTER] 00:38:24.313 --> 00:38:28.520 SMITH: My perspective on, like, how collaborations  come in to be is they're really serendipitous. You   00:38:28.520 --> 00:38:34.040 know, we talked about how this one came in to be,  and it was because we all happen to know Vikram,   00:38:34.040 --> 00:38:37.840 and Vikram happened to connect Bichlien  with Aniruddh, and it kind of rolled from   00:38:37.840 --> 00:38:43.080 there. But you can have serendipitous, you know,  meetings at a conference, where you happen to,   00:38:43.080 --> 00:38:48.600 you know, sit next to someone at a talk  and you both share the same perspective on,   00:38:48.600 --> 00:38:52.760 you know, how a research problem should  be tackled, and something could come   00:38:52.760 --> 00:38:57.440 out of that. Or in some cases, you go  actually shopping for a collaborator. 00:38:57.440 --> 00:38:58.697 HUIZINGA: Right. [LAUGHTER] 00:38:58.697 --> 00:39:00.560 SMITH: You know, you need to talk to 10   00:39:00.560 --> 00:39:06.920 people to find the one that has that same research  perspective as you. I’ll second Aniruddh’s,   00:39:06.920 --> 00:39:10.960 you know, observation that you get a  very different perspective if you go   00:39:10.960 --> 00:39:15.800 find someone who, they may have the same, like,  perspective on how research should be tackled,   00:39:15.800 --> 00:39:21.760 but they have a different perspective on what the  ultimate output of that research would be. But,   00:39:21.760 --> 00:39:26.360 you know, they can often point you in areas  where your research could be helpful that you   00:39:26.360 --> 00:39:30.740 can't necessarily see because you lack the domain  knowledge or you lack that particular angle on it. 00:39:30.740 --> 00:39:34.520 HUIZINGA: Which is another interesting thing  in my mind is, you know, the role that papers,   00:39:34.520 --> 00:39:40.760 published papers, play—that’s a lot of p’s  in a sentence [LAUGHTER] … alliteration—that   00:39:41.480 --> 00:39:45.400 you would be reading or hearing  about either in a lightning talk   00:39:45.400 --> 00:39:49.040 or a presentation at a conference.  Does that broaden your perspective,   00:39:49.040 --> 00:39:54.600 as well? And how do you … like, do you  call people up? “I read your paper… ”? 00:39:54.600 --> 00:39:59.960 SMITH: [LAUGHS] I have cold-emailed people.  You know, this works sometimes! Sometimes this   00:39:59.960 --> 00:40:05.560 is just the introduction that you need. But  the interesting thing in my mind is how much   00:40:06.240 --> 00:40:11.480 the computer science conferences and things  like ChemRxiv and arXiv have really replaced,   00:40:11.480 --> 00:40:18.520 for me, the traditional chemistry literature  or the traditional publishing literature where   00:40:18.520 --> 00:40:21.960 you can have a conversation with this person  while they're still actively doing the work   00:40:21.960 --> 00:40:25.160 because they put their initial draft  up there and it still needs revision,   00:40:25.160 --> 00:40:30.920 and there's opportunities even earlier on in  the research process than we've had in the past. 00:40:30.920 --> 00:40:35.680 HUIZINGA: Huh. And to your earlier point,  I'm envisioning an Amazon shopping cart for   00:40:35.680 --> 00:40:42.720 research collaborators. [LAUGHTER] “Oh,  he looks good. Into my cart.” Aniruddh,   00:40:42.720 --> 00:40:47.960 I always like to know where a project is on  the spectrum from what I call lab to life,   00:40:47.960 --> 00:40:52.400 and I know there are different development stages  when it comes to technology finding its way into   00:40:52.400 --> 00:40:57.920 production and then into broader use. So to use  another analogy I like, pretend this is a relay   00:40:57.920 --> 00:41:03.420 race and research is the first leg. Who else  has to run, and who brings it across the line? 00:41:03.420 --> 00:41:07.800 VASHISTH: Yeah, yeah. So I think  the initial work that we have done,   00:41:07.800 --> 00:41:12.120 I think it's been super fruitful, and  to Jake's point, like, converging to,   00:41:12.120 --> 00:41:16.280 like, a nice output. It took a bunch  of chemists, mechanical engineers,   00:41:16.280 --> 00:41:22.600 simulation folks, machine learning scientists  to get where we are. And, as Jake mentioned,   00:41:22.600 --> 00:41:29.640 we've actually put some of our publications on  arXiv, and it's getting traction now. So we've   00:41:29.640 --> 00:41:37.000 had some excitement from startups and companies  which make polymers asking us, “Oh, can you   00:41:37.000 --> 00:41:43.320 actually … can we get a slice of this framework  that you're developing for designing vitrimers?”   00:41:43.320 --> 00:41:49.440 Which is very promising. So we have done very  fundamental work, but now, like, what's called   00:41:49.440 --> 00:41:54.865 “the valley of death” in research, [LAUGHTER] like  taking it from lab to like production scale, … 00:41:54.880 --> 00:42:00.720 VASHISTH: … it's usually a very tightly  knit collaboration between industry, labs,   00:42:00.720 --> 00:42:05.320 and sometimes national labs, too. So we're  excited that, actually, a couple of national   00:42:05.320 --> 00:42:10.640 labs have been interested in the work that we  have been doing, so super optimistic about it. 00:42:10.640 --> 00:42:14.800 HUIZINGA: So would you say that the  vitrimer-based printed circuit board   00:42:14.800 --> 00:42:20.180 is a proof of concept right now? Or have  you made prototypes? Where is that now? 00:42:20.180 --> 00:42:24.040 SMITH: Yeah, absolutely. We've  mentioned our other collaborator,   00:42:24.040 --> 00:42:27.120 Vikram Iyer, a couple of times.  And in collaboration with his lab,   00:42:27.640 --> 00:42:33.600 we did actually make a prototype circuit board. We  showed that it works as you expect. We showed that   00:42:33.600 --> 00:42:37.030 it can be disassembled. It can be put back  together, and it still works as expected … 00:42:37.030 --> 00:42:37.630 HUIZINGA: The “break  stuff/make stuff back” thing … 00:42:37.630 --> 00:42:38.660 VASHISTH: Yeah, exactly. 00:42:38.660 --> 00:42:44.480 SMITH: But, you know, I think to the spirit of the  question, it's still individual kind of one-off   00:42:44.480 --> 00:42:48.880 experiments being run in a lab, and Aniruddh  is right. There's a long way to go from, like,   00:42:49.440 --> 00:42:56.333 Technology Readiness Level 3, where we're doing it  ourselves on bench scale, up to, you know, the 7,   00:42:56.333 --> 00:43:01.699 8, 9, where it's actually commercially viable and  someone has been able to reproduce this at scale. 00:43:01.699 --> 00:43:03.280 HUIZINGA: Right. … So that's  when you bring investors in   00:43:03.280 --> 00:43:06.320 or labs that can make stuff in and scale. 00:43:06.320 --> 00:43:08.680 VASHISTH: Yeah. Yeah, I think once you’re, like,   00:43:08.680 --> 00:43:12.460 close to 7, I think that's where you're  pretty much ready for the big show. 00:43:12.460 --> 00:43:13.760 HUIZINGA: So where are you now? 2? 3? 00:43:13.760 --> 00:43:15.560 VASHISTH: I would say, like, 2 or 3 … 00:43:15.560 --> 00:43:17.335 SMITH: 2, 3, somewhere in that range. 00:43:17.500 --> 00:43:21.240 SMITH: The scales, kind of, differ depending  on which agencies you see put it out. 00:43:21.240 --> 00:43:27.240 HUIZINGA: So, Jake, before we close, I want  to talk briefly about other applications   00:43:27.240 --> 00:43:31.480 of recyclable vitrimer-based polymers,  in light of their importance to the   00:43:31.480 --> 00:43:36.120 climate research initiative and AI for  Science. So what other industries have   00:43:36.120 --> 00:43:41.600 polymer components that have nowhere  to go after they die but the landfill,   00:43:41.600 --> 00:43:45.920 and will this research transfer  across to those industries? 00:43:45.920 --> 00:43:51.360 SMITH: An excellent question. So my personal view  on this is that there's a couple of classes of   00:43:51.360 --> 00:43:55.720 polymers. There's these very high-value  application uses of polymers where we're   00:43:55.720 --> 00:43:59.480 talking about the printed circuit boards;  we're talking about aerospace composite;   00:43:59.480 --> 00:44:03.581 we're talking about the panels on your car;  we're talking about things like wind turbines … 00:44:03.600 --> 00:44:08.760 SMITH: … where there's a long life cycle. You  have this device that's going to be in use for   00:44:08.760 --> 00:44:14.080 five years, 50 years, and at the end of that, the  polymer itself is still probably pretty good. You   00:44:14.080 --> 00:44:17.960 could still use it and regenerate it. And so  Aniruddh’s lab has done great work showing that   00:44:17.960 --> 00:44:24.320 you can take things like the side panel of a plane  and actually disassemble this thing, heal it, keep   00:44:24.320 --> 00:44:29.760 it in use longer, and use it at the end of its  lifetime. There's this other class of polymers,   00:44:29.760 --> 00:44:33.960 which I think are the ones that most people  think about—your Coke bottle—and vitrimers   00:44:33.960 --> 00:44:39.480 seem like a much harder sell there. I think this  is more the domain of, you know, biodegradable   00:44:39.480 --> 00:44:45.560 polymers in the long run to really tackle the  issues there. But I'm very excited in this,   00:44:45.560 --> 00:44:50.360 you know, high-value polymer, this long-lifetime  polymer, this, like, permanent install polymer,   00:44:50.360 --> 00:44:53.900 however you want to think about it,  for work like this to have an impact. 00:44:53.900 --> 00:44:56.680 HUIZINGA: Yeah. From your lab’s perspective,   00:44:56.680 --> 00:45:00.620 Aniruddh, where do you see other  applications with great promise? 00:45:00.620 --> 00:45:06.240 VASHISTH: Yeah. So as Jake said, places where  we need high-performance polymers is where we   00:45:06.240 --> 00:45:12.866 can go. So PCBs is one, aerospace and automotive  industry is one, and maybe medical industry is, … 00:45:12.866 --> 00:45:13.426 HUIZINGA: Oh, interesting… 00:45:13.426 --> 00:45:17.440 VASHISTH: … yeah, is another one where we  can actually … if you can make prosthetics   00:45:17.440 --> 00:45:21.640 out of vitrimers … prosthetics actually  lose a little bit of their stiffness,   00:45:21.640 --> 00:45:26.200 you know, as you use them, and that's because  of localized damage. It's the fatigue cycle,   00:45:26.200 --> 00:45:32.200 right. So what if you can actually heal your  prosthetics and reuse them? So, yeah, I feel like,   00:45:32.200 --> 00:45:36.600 you know, there's so many different applications,  so many different routes that we can go down. 00:45:36.600 --> 00:45:41.840 HUIZINGA: Yeah. Well, I like to end our  Collaborators shows with a little vision casting,   00:45:41.840 --> 00:45:47.720 and I feel like this whole podcast is that. I  should also say, you know, back in the ’50s,   00:45:47.720 --> 00:45:55.880 there was the big push to make plastics! Your  word is vitrimers! So let's do a little vision   00:45:55.880 --> 00:46:01.800 casting for vitrimer-based polymers. Assuming  your research is wildly successful and becomes   00:46:01.800 --> 00:46:07.240 a truly game-changing technology, what does  the future look like—I mean, specified future,   00:46:07.240 --> 00:46:12.800 not general future—and how has your work  disrupted this field and made the world   00:46:12.800 --> 00:46:16.660 a better place? I'll let you each have  the last word. Who'd like to go first? 00:46:16.660 --> 00:46:19.760 VASHISTH: Sure, I can go first. I'll try to make   00:46:19.760 --> 00:46:22.411 sure that I break it up into  computation and experiments … 00:46:22.426 --> 00:46:26.920 VASHISTH: … so that once I go back, like,  my lab does not, like, pounce on me.   00:46:27.760 --> 00:46:33.200 [LAUGHS] Yeah, so I think from the computation  point of view, we always thought that if somebody   00:46:33.200 --> 00:46:37.600 gave us, like, a hundred different chemistries,  we can actually bottle it down to, like,   00:46:37.600 --> 00:46:41.600 we can do a bunch of simulations; tell you, like,  10 of these actually work. What we've been able   00:46:41.600 --> 00:46:46.600 to do specifically for vitrimers is that we're  able to look at the problem from the other side,   00:46:46.600 --> 00:46:50.880 and we are able to say that if you  tell me a particular application,   00:46:50.880 --> 00:46:55.200 this particular chemistry would work best  for you. In essence, what we were thinking   00:46:55.200 --> 00:47:00.600 of is that if aliens abducted all the chemists  from the world, can we actually come up with a   00:47:00.600 --> 00:47:05.560 framework? [LAUGHS] So I think it'll be difficult  to get there because as I said earlier that,   00:47:05.560 --> 00:47:10.600 you know, you need that human touch. But I think  we are happy that that we are getting there. And   00:47:10.600 --> 00:47:16.200 I think what remains to be seen now is, like, you  know, now that we have this type of a framework,   00:47:16.200 --> 00:47:20.480 like what are the next challenges? Like, we are  going from the lab to the large scale; like,   00:47:20.480 --> 00:47:24.800 what challenges are associated there? And  I think similarly for the experimental side   00:47:24.800 --> 00:47:30.640 of things also, we know a lot—we have developed  frameworks—but there's a lot of work that still   00:47:30.640 --> 00:47:36.420 needs to be done in understanding and translating  these technologies to real-life applications. 00:47:36.420 --> 00:47:39.400 HUIZINGA: I like that you're kind of hedging  your bets there, saying, I'm not going to   00:47:39.400 --> 00:47:43.480 paint a picture of the perfect world because my  lab is going to be responsible for delivering   00:47:43.480 --> 00:47:52.920 it. [LAUGHTER] Jake, assuming you haven't been  abducted by aliens, what's your take on this? 00:47:52.920 --> 00:47:58.360 SMITH: I view, kind of, the goal of this  work and the ideal impact of this work as   00:47:58.360 --> 00:48:03.200 an acceleration of getting us to these  polymers being deployed in all these   00:48:03.200 --> 00:48:06.695 other applications that we've talked  about, and we can go broader than this. 00:48:06.713 --> 00:48:10.800 SMITH: I think that there's a lot of work,  both within the MCRI, within Microsoft,   00:48:10.800 --> 00:48:16.560 and outside of Microsoft in the bigger field,  focused on acceleration towards a specific   00:48:16.560 --> 00:48:22.800 goal. And if all of this work is successful,  in 10 years, maybe our materials design process   00:48:22.800 --> 00:48:28.240 looks completely different, where we've gone  from this kind of brute-force screening that   00:48:28.240 --> 00:48:33.040 Aniruddh has talked about to an approach where  you start with the properties that you care about;   00:48:33.040 --> 00:48:38.320 they're defined by the application that you have  in mind. You want to make your vitrimer PCB,   00:48:38.320 --> 00:48:42.680 it needs to have, you know, a specific temperature  where it becomes gummy; it needs to have a   00:48:42.680 --> 00:48:51.440 specific resistance to burning; it needs to be  able to effectively serve as the dielectric for   00:48:51.440 --> 00:48:58.000 your bigger circuits. And we use this, like, “need  space” to define the material that we would like,   00:48:58.000 --> 00:49:04.440 and we can use machine learning, artificial  intelligence, in order to get us to the structure   00:49:04.440 --> 00:49:09.720 that we need to make in order to actually achieve  this design space. And so, this was, you know,   00:49:09.720 --> 00:49:14.760 our big bet within AI for Science. This is the  big bet of this project. And with this project,   00:49:14.760 --> 00:49:18.800 you know, we take one step towards showing  that you can do this in one case. And the   00:49:18.800 --> 00:49:25.584 future casting would be we can do this in every  materials design case that you can think about. 00:49:25.584 --> 00:49:30.520 HUIZINGA: Hmmm. You know, I'm thinking of  lanes—track analogy again—but, you know,   00:49:30.520 --> 00:49:34.520 you've got mechanical engineering, you've got  chemistry, and you've got artificial intelligence,   00:49:34.520 --> 00:49:39.480 and each of those sciences is advancing,  and they're using each other to, sort of,   00:49:39.480 --> 00:49:45.160 help advance in various ways, so this is an  exciting, exciting project and collaboration. 00:49:45.160 --> 00:49:45.960 [MUSIC] 00:49:45.960 --> 00:49:50.240 Jake, Aniruddh, thanks for joining us  today on Collaborators. This has been   00:49:50.240 --> 00:49:55.680 really fun for me. [LAUGHTER] So thanks for  coming in and sharing your stories today. 00:49:55.680 --> 00:49:56.640 VASHISTH: Thank you so much. 00:49:56.640 --> 00:50:13.160 SMITH: Yeah. Of course. Thank you. 00:50:13.160 --> 00:50:13.910 [MUSIC FADES]