WEBVTT 00:01.290 --> 00:04.019 Really excited to introduce our next speaker 00:04.019 --> 00:05.325 or set of speakers. 00:10.290 --> 00:14.820 Jana Saadi is-- actually, you began your journey here at MIT 00:14.820 --> 00:16.050 like nine years ago. 00:16.050 --> 00:19.680 It's hard to believe, but she began her journey here 00:19.680 --> 00:22.560 as an undergrad in mechanical engineering. 00:22.560 --> 00:25.835 And she found herself in Maria Yang's class. 00:25.835 --> 00:27.960 I don't know how many of you have taken that class. 00:27.960 --> 00:30.150 Probably half of you. 00:30.150 --> 00:34.020 It's about toy design, and she took that. 00:34.020 --> 00:36.360 Seems to be a transformative class for you 00:36.360 --> 00:38.370 because now she's gone full circle, 00:38.370 --> 00:41.640 and she is in Maria's lab called Ideation Lab, 00:41.640 --> 00:46.470 where she's working on AI processes or AI tools 00:46.470 --> 00:49.630 for early design stage processes. 00:49.630 --> 00:54.630 So she's joined with Maria, of course, and Maria, 00:54.630 --> 00:55.860 I have to read your titles. 00:55.860 --> 00:57.070 There are so many. 00:57.070 --> 01:00.210 So I don't know how you do it, but Maria Yang 01:00.210 --> 01:02.260 is a Professor here at MIT. 01:02.260 --> 01:06.060 She is a professor of mechanical engineering, associate dean 01:06.060 --> 01:08.550 of engineering, and she's associated 01:08.550 --> 01:11.430 director of the Morningside Academy for Design. 01:11.430 --> 01:15.150 And the two of them are going to talk about generative design 01:15.150 --> 01:15.790 tools. 01:15.790 --> 01:17.985 So please welcome both of them. 01:17.985 --> 01:21.240 [APPLAUSE] 01:24.040 --> 01:24.954 Yes. 01:24.954 --> 01:26.690 You got everything? 01:26.690 --> 01:27.430 We get two mics. 01:27.430 --> 01:27.670 Hi. 01:27.670 --> 01:28.170 Hi. 01:28.170 --> 01:31.950 So I'm very excited to introduce Jana Saadi. 01:31.950 --> 01:35.020 She's an amazing designer and amazing researcher. 01:35.020 --> 01:37.412 And she thinks about how people design 01:37.412 --> 01:39.370 so that we can understand how they design so we 01:39.370 --> 01:40.537 can help them design better. 01:40.537 --> 01:41.960 That's really what she does. 01:41.960 --> 01:43.630 And I will also do a quick poll. 01:43.630 --> 01:46.760 How many people in this room design things at all? 01:46.760 --> 01:47.260 Yes. 01:47.260 --> 01:47.830 Lots. 01:47.830 --> 01:49.300 I knew there was a ringer question. 01:49.300 --> 01:52.150 But how many of you use computational tools like CAD, 01:52.150 --> 01:53.810 or Rhino, or any of those things? 01:53.810 --> 01:54.860 So you-- so good. 01:54.860 --> 01:56.470 So I'm just trying to sort of prime 01:56.470 --> 01:58.180 that a little bit before Donna does her speech. 01:58.180 --> 01:58.720 So OK. 01:58.720 --> 02:00.010 All right. 02:00.010 --> 02:00.820 Thank you. 02:00.820 --> 02:04.000 So I'll be talking on my work with generative design tools 02:04.000 --> 02:06.910 that I did during my year as a fellow in the Morningside 02:06.910 --> 02:07.870 Academy. 02:07.870 --> 02:11.590 And as [? Faba ?] had said, my journey in design here at MIT 02:11.590 --> 02:14.170 really started with a lot of hands-on classes 02:14.170 --> 02:17.440 that introduced me to this traditional human-centered 02:17.440 --> 02:19.750 design process, where it's very hands-on. 02:19.750 --> 02:20.632 You're sketching. 02:20.632 --> 02:21.340 You're iterating. 02:21.340 --> 02:22.840 You're making physical prototypes. 02:22.840 --> 02:27.100 You're testing with users, and it's very product-driven. 02:27.100 --> 02:30.070 You're making either a physical product or service, 02:30.070 --> 02:32.853 and it's led by the human designer. 02:32.853 --> 02:34.270 And I personally very much enjoyed 02:34.270 --> 02:37.030 being that designer that was leading the process 02:37.030 --> 02:39.460 and really making these products that could 02:39.460 --> 02:41.200 be very impactful for people. 02:41.200 --> 02:43.300 And it wasn't until Maria actually poked me 02:43.300 --> 02:46.300 towards this direction of computational design tools 02:46.300 --> 02:48.340 that I started looking more into this direction 02:48.340 --> 02:49.870 of generative design. 02:49.870 --> 02:52.570 And these tools are being introduced more often, 02:52.570 --> 02:55.150 and it's more about this collaboration between the human 02:55.150 --> 02:58.010 and the AI in the design process. 02:58.010 --> 02:59.800 So I started off first by looking 02:59.800 --> 03:02.900 at products that were created using generative design tools. 03:02.900 --> 03:04.540 So you'll see an example of them here. 03:04.540 --> 03:07.960 On the top left, you have the airplane partition 03:07.960 --> 03:10.240 made for the Airbus airplane, which was 03:10.240 --> 03:12.250 optimized to be lighter weight. 03:12.250 --> 03:14.807 And it's significantly lighter than its original counterpart, 03:14.807 --> 03:16.390 which you'd imagine for an airplane is 03:16.390 --> 03:19.660 amazing in terms of the amount of fuel that it's able to save. 03:19.660 --> 03:22.690 Next to that is the Volkswagen tire 03:22.690 --> 03:25.090 hub, which was optimized for strength, of course, 03:25.090 --> 03:27.920 because it's a part of a car. 03:27.920 --> 03:29.680 But it also took into account aesthetics. 03:29.680 --> 03:32.650 They were really trying to make it this nice, symmetric kind 03:32.650 --> 03:34.450 of star-shaped design. 03:34.450 --> 03:38.560 Below that is the new balance shoe insert, which is actually 03:38.560 --> 03:40.480 specific to your own foot map. 03:40.480 --> 03:42.250 So they take the pressure of your foot, 03:42.250 --> 03:44.933 and they'll have a denser lattice structure in places 03:44.933 --> 03:46.100 where you need more support. 03:46.100 --> 03:48.880 So really opened this door for this more customizable 03:48.880 --> 03:50.740 individual generative design. 03:50.740 --> 03:53.770 And then finally, next to that is the Voxman Music building, 03:53.770 --> 03:55.750 which was optimized for acoustics. 03:55.750 --> 03:57.670 So it took into account lighting, sound, 03:57.670 --> 04:00.160 and all those things for their music hall. 04:00.160 --> 04:02.200 So you can see really hear the examples 04:02.200 --> 04:04.270 of the potential of generative design 04:04.270 --> 04:07.690 to create these higher-performing designs that 04:07.690 --> 04:11.170 have this very specific or very new visual aesthetics. 04:11.170 --> 04:14.170 It generates them very quickly, and it generates a lot of them, 04:14.170 --> 04:17.350 much faster than what a designer is able to do on their own. 04:17.350 --> 04:19.149 And they're very complex. 04:19.149 --> 04:21.640 So it's often something that's more complex 04:21.640 --> 04:24.340 than an individual designer is able to do. 04:24.340 --> 04:26.500 These tools are starting to be adopted more 04:26.500 --> 04:28.150 in research and industry, but it's not 04:28.150 --> 04:30.310 yet super widely adopted. 04:30.310 --> 04:33.368 And what I had started to realize while looking into this 04:33.368 --> 04:35.410 is a lot of the research is about the development 04:35.410 --> 04:37.330 of these tools and trying to make them better. 04:37.330 --> 04:40.597 But we don't quite understand how they're used. 04:40.597 --> 04:43.180 And so that's something that I was interested in understanding 04:43.180 --> 04:45.130 more in my own research in trying 04:45.130 --> 04:47.950 to see how the use of these generative design tools 04:47.950 --> 04:51.160 can affect the design process, the designers' behaviors, 04:51.160 --> 04:53.470 and their design outcomes. 04:53.470 --> 04:55.450 And like I said, they're becoming 04:55.450 --> 04:57.560 to be used more in the future of design. 04:57.560 --> 05:00.220 And I think it's super critical to understand the impacts 05:00.220 --> 05:02.230 that they have in the design as they're 05:02.230 --> 05:04.370 becoming more commonly used. 05:04.370 --> 05:08.740 So I did this by conducting interviews of designers that 05:08.740 --> 05:11.290 use these generative design tools in their own work 05:11.290 --> 05:13.600 to try to understand their process 05:13.600 --> 05:15.130 and using the commercially-available 05:15.130 --> 05:17.333 generative design tools. 05:17.333 --> 05:19.750 So I'm actually just going to walk through a quick example 05:19.750 --> 05:20.750 of this process. 05:20.750 --> 05:23.080 So here there's an example of a G engine 05:23.080 --> 05:27.513 bracket done with Fusion 360's generative design tools. 05:27.513 --> 05:28.930 And it's actually just the example 05:28.930 --> 05:30.388 that they have a built-in tutorial. 05:30.388 --> 05:33.590 So you can actually go home and try this out for yourself. 05:33.590 --> 05:36.040 So you start off with a CAD model that 05:36.040 --> 05:40.560 typically will introduce some-- 05:40.560 --> 05:42.060 I don't know if it's playing or not. 05:49.660 --> 05:50.160 All right. 05:52.740 --> 05:54.930 So I don't know if the video is playing, but anyway. 05:54.930 --> 05:57.820 You'll start off with this CAD model, which you'll see here. 05:57.820 --> 06:00.450 And it's typically to be able to give some geometry 06:00.450 --> 06:02.590 considerations or constraints. 06:02.590 --> 06:04.710 So over here you'll, for example, 06:04.710 --> 06:06.330 start selecting some of the geometries 06:06.330 --> 06:09.610 you need to keep, often for assembly purposes. 06:09.610 --> 06:13.660 So these bolts that you have on the edges here. 06:13.660 --> 06:16.080 So these geometries so you can actually put in the bolts. 06:16.080 --> 06:17.790 And then of the other geometries where 06:17.790 --> 06:20.820 you want to make sure that they're 06:20.820 --> 06:21.930 like the keep-out zones. 06:21.930 --> 06:23.070 So there are places where you actually 06:23.070 --> 06:25.153 need to put the bolts through or things like that. 06:25.153 --> 06:27.160 So all of these inputs to the tool. 06:27.160 --> 06:29.310 You can also put in things like their forces 06:29.310 --> 06:31.393 that you're going to have to be able to withstand, 06:31.393 --> 06:34.450 the manufacturing methods, your safety factor. 06:34.450 --> 06:39.480 So all of these inputs into this tool, and often these inputs 06:39.480 --> 06:41.190 are decided by the designer. 06:41.190 --> 06:43.320 And a lot of their intuition and expertise 06:43.320 --> 06:44.950 goes into this process. 06:44.950 --> 06:47.038 So for example, how do you know which bolt to use? 06:47.038 --> 06:47.580 I don't know. 06:47.580 --> 06:49.960 In the past, we've used the half-inch bolt. It's worked. 06:49.960 --> 06:52.502 It hasn't failed me, so I'm just going to keep on using that. 06:52.502 --> 06:54.240 So really the designer's intuition 06:54.240 --> 06:56.880 in defining the objectives, parameters, and constraints 06:56.880 --> 06:59.100 goes into this process of-- 06:59.100 --> 07:01.650 into the generative design tool. 07:01.650 --> 07:03.810 And from that, the tool will generate 07:03.810 --> 07:05.730 a whole bunch of designs. 07:05.730 --> 07:07.650 And there are quite a lot, and the designer 07:07.650 --> 07:09.030 will go through these results. 07:09.030 --> 07:12.190 And they'll iterate on the parameters and constraints. 07:12.190 --> 07:14.190 So often you'll have these results, 07:14.190 --> 07:16.530 and it's almost never that the first time you 07:16.530 --> 07:18.570 use the tool you'll get the perfect design. 07:18.570 --> 07:20.430 You'll always end up with something where-- 07:20.430 --> 07:22.767 I mean, you can see over here my first iteration. 07:22.767 --> 07:24.600 I have some really chunky designs out there. 07:24.600 --> 07:26.552 Like, what was going on over there? 07:26.552 --> 07:28.260 And I realized, oh, I forgot some forces. 07:28.260 --> 07:31.260 So you go back, and you iterate on your constraints 07:31.260 --> 07:32.490 and parameters. 07:32.490 --> 07:35.430 And you kind of keep going through that loop 07:35.430 --> 07:38.860 until you're finally happy with your set of results. 07:38.860 --> 07:41.760 And then finally, you select from the results 07:41.760 --> 07:43.380 that come out once you're done. 07:43.380 --> 07:46.750 And again, this selection can be quite a task on the designer. 07:46.750 --> 07:48.930 They're really balancing all of the objectives, 07:48.930 --> 07:50.950 both qualitative and quantitative. 07:50.950 --> 07:53.760 So what you're able to actually tell the tool and maybe things 07:53.760 --> 07:54.940 that you're not. 07:54.940 --> 07:58.145 So for example, we're optimizing for weight over here. 07:58.145 --> 07:59.520 So we want a lightweight bracket, 07:59.520 --> 08:01.380 but I might actually select one that's 08:01.380 --> 08:03.940 slightly heavier because I know it's easier to manufacture 08:03.940 --> 08:06.190 or something like that, or maybe it just looks cooler, 08:06.190 --> 08:07.935 and I just really liked how it looked. 08:07.935 --> 08:10.560 So there's that selection again, where the designer's intuition 08:10.560 --> 08:11.770 comes into play. 08:11.770 --> 08:13.920 And then there's a level of refinement 08:13.920 --> 08:16.530 that can happen afterwards, where it either could just 08:16.530 --> 08:18.750 be like simple refinement, like kind of smoothing out 08:18.750 --> 08:21.240 some surfaces, or it could be something 08:21.240 --> 08:25.553 that's more intense, like a much larger redesign. 08:25.553 --> 08:26.970 So this is the process that I just 08:26.970 --> 08:28.280 described in diagram format. 08:28.280 --> 08:30.780 And I think what was cool is that through all the interviews 08:30.780 --> 08:33.419 that we've done, despite the different backgrounds 08:33.419 --> 08:35.970 of the interviewees and the types of tools that they use, 08:35.970 --> 08:38.039 this high-level process is something 08:38.039 --> 08:39.309 they all went through. 08:39.309 --> 08:42.580 And as you can see here, it's not a simple linear process. 08:42.580 --> 08:45.030 It's not just use the tool, magic button, 08:45.030 --> 08:47.040 your perfect dream design comes out. 08:47.040 --> 08:49.200 There's a lot of collaboration and dialogue 08:49.200 --> 08:52.080 between the designer and the generative design 08:52.080 --> 08:53.880 tool that has to happen. 08:53.880 --> 08:55.380 And I think that's what I was trying 08:55.380 --> 08:57.570 to do with my own research, is really bringing out 08:57.570 --> 09:01.050 this process and understanding more systematically 09:01.050 --> 09:03.060 how the designer interacts with the tool 09:03.060 --> 09:06.150 and what the implications of using these generative design 09:06.150 --> 09:10.450 tools can have on the design process. 09:10.450 --> 09:13.210 So I'll leave you with three main implications. 09:13.210 --> 09:16.242 The first is that it's a very different way of thinking. 09:16.242 --> 09:18.450 As you saw in the beginning, I wasn't really thinking 09:18.450 --> 09:20.568 about how I wanted the bracket to look. 09:20.568 --> 09:22.110 I was thinking about the constraints. 09:22.110 --> 09:24.610 I was thinking about what are the geometries I need to have, 09:24.610 --> 09:25.590 what are my forces. 09:25.590 --> 09:28.080 So it becomes this very constraint-driven design, 09:28.080 --> 09:30.390 and designers become designers of constraints 09:30.390 --> 09:34.330 rather than designers of actual physical shape of the objects. 09:34.330 --> 09:37.800 So as one of the interviews we had mentioned, kind of called 09:37.800 --> 09:40.500 this the ability to have an abstraction of your goal, 09:40.500 --> 09:42.300 to really understand the constraints 09:42.300 --> 09:45.750 and not care about the shape anymore. 09:45.750 --> 09:48.720 Secondly, it's still very much a human-driven process, 09:48.720 --> 09:52.620 where the generative design is a tool that the designer can use. 09:52.620 --> 09:55.360 And they can use it in very creative ways. 09:55.360 --> 09:57.690 So some of our interviewees kind of mentioned 09:57.690 --> 10:00.450 how they would use the generative design just 10:00.450 --> 10:03.630 to give a set of inspiration and then maybe do a bit more 10:03.630 --> 10:05.910 traditional design afterwards. 10:05.910 --> 10:09.450 And it just really opens the door for the type of designs 10:09.450 --> 10:11.400 and creative designs that you didn't really 10:11.400 --> 10:14.070 think to make on your own because it gives you 10:14.070 --> 10:16.950 so many results in the end that you can kind of map 10:16.950 --> 10:19.500 this whole space of cool designs and then 10:19.500 --> 10:22.530 use it more creatively in your own work. 10:22.530 --> 10:25.080 And then finally, it also really offers this learning 10:25.080 --> 10:26.170 experience. 10:26.170 --> 10:29.040 So here is actually my first attempt at making a bookend. 10:29.040 --> 10:31.110 And as you can see, it was a very poor attempt. 10:31.110 --> 10:32.520 I think it will not quite work. 10:32.520 --> 10:35.040 It'll just fall over, and it won't really hold any books. 10:35.040 --> 10:37.170 And almost immediately when I saw that design, 10:37.170 --> 10:39.315 I realized, oh, we're the center of mass 10:39.315 --> 10:41.190 is a constraint I have to pay attention 10:41.190 --> 10:42.690 to make sure it doesn't fall. 10:42.690 --> 10:44.440 So as soon as you start doing this, 10:44.440 --> 10:46.260 you realize, OK, if I were designing 10:46.260 --> 10:48.420 this bookend in a more traditional way 10:48.420 --> 10:51.120 and sketching it out by hand, I would have intuitively made 10:51.120 --> 10:52.810 sure that it didn't fall over. 10:52.810 --> 10:54.780 But by using this generative design tool, 10:54.780 --> 10:56.880 I was able to pull out that constraint 10:56.880 --> 10:59.910 and understand more thoroughly what the constraints are 10:59.910 --> 11:03.810 in the design problem, how they relate to the design 11:03.810 --> 11:06.480 geometries in the end, and what the potential solution might 11:06.480 --> 11:07.030 look like. 11:07.030 --> 11:08.850 So once you go through all that, you 11:08.850 --> 11:10.290 can get a much better bookend. 11:10.290 --> 11:11.317 Someone online did that. 11:11.317 --> 11:12.900 I haven't even reached this stage yet, 11:12.900 --> 11:14.650 but you could see that it takes some time. 11:14.650 --> 11:16.320 It takes some learning and just trial 11:16.320 --> 11:21.450 and error iteration that helps you build this understanding. 11:21.450 --> 11:24.330 But I'm quite personally excited about the potential 11:24.330 --> 11:27.130 of using these generative design tools in design, 11:27.130 --> 11:28.830 especially in early-stage ideation 11:28.830 --> 11:31.650 to get a better understanding of the design problem, what 11:31.650 --> 11:33.420 the constraints are, how they influence 11:33.420 --> 11:36.540 the shape and the geometry, what the potential solutions might 11:36.540 --> 11:39.870 look like, and how you might use that in more traditional design 11:39.870 --> 11:40.860 spaces. 11:40.860 --> 11:43.680 But I do think that our design education will 11:43.680 --> 11:47.580 have to change in order to incorporate these design tools. 11:47.580 --> 11:50.478 Like I said, it requires a very different way of thinking. 11:50.478 --> 11:52.020 And so this is something that we have 11:52.020 --> 11:57.730 to be trained about more in our design education curriculum. 11:57.730 --> 11:59.543 And I think we also have the responsibility 11:59.543 --> 12:00.960 to make sure that we're using them 12:00.960 --> 12:02.850 to design very sustainably. 12:02.850 --> 12:05.970 Like you saw, a lot of the designs could be very complex, 12:05.970 --> 12:08.100 and they can really open to higher performance 12:08.100 --> 12:11.820 as such as lighter weight and very strong products. 12:11.820 --> 12:14.340 So using these tools to our advantage 12:14.340 --> 12:17.490 to make sure that we're really pushing the bounds for how 12:17.490 --> 12:20.250 we can design very sustainably. 12:20.250 --> 12:22.170 And then finally, making sure that we still 12:22.170 --> 12:24.060 do all of this while designing ethically 12:24.060 --> 12:25.440 and justly with everything. 12:25.440 --> 12:28.020 So making sure that we're aware of the biases 12:28.020 --> 12:30.480 that the tools bring, the biases that we bring 12:30.480 --> 12:33.090 as human designers, where the tools fall short, 12:33.090 --> 12:36.240 and how as human designers we can check them, 12:36.240 --> 12:38.310 and just making sure that this collaboration 12:38.310 --> 12:42.600 between human designer and AI is done in a very just way where 12:42.600 --> 12:45.870 we're still having a very human-centered process as we 12:45.870 --> 12:48.570 move more forward into this future of design 12:48.570 --> 12:51.870 of using these types of generative design tools. 12:51.870 --> 12:55.050 And I hope through this you saw the exciting opportunities 12:55.050 --> 12:58.230 that generative design has in pushing our design 12:58.230 --> 12:59.550 potentials forward. 12:59.550 --> 13:01.860 And I'm happy to take any questions 13:01.860 --> 13:04.073 after a discussion with Maria. 13:04.073 --> 13:07.384 [APPLAUSE] 13:07.384 --> 13:08.330 That's really nice. 13:11.180 --> 13:13.270 So thank you, Jana. 13:13.270 --> 13:15.080 I'm very excited about that work, 13:15.080 --> 13:16.640 and I've done such a wonderful job. 13:16.640 --> 13:17.690 And I think it's really important 13:17.690 --> 13:19.050 right now because we're at this time where 13:19.050 --> 13:21.508 you have all these sorts of tools and all things happening. 13:21.508 --> 13:23.630 And there's this maybe very romantic notion 13:23.630 --> 13:24.680 that you're just going to push a button, 13:24.680 --> 13:26.330 and then magically you're going to get this perfect design. 13:26.330 --> 13:29.060 But what your work really shows in a very systematic way 13:29.060 --> 13:31.993 is there's a lot of workarounds and things 13:31.993 --> 13:34.160 that people do that still are fundamentally designed 13:34.160 --> 13:36.210 but maybe redefined in what we do in the future. 13:36.210 --> 13:36.710 Right? 13:36.710 --> 13:39.330 So are we all constraint designers? 13:39.330 --> 13:40.010 I don't know. 13:40.010 --> 13:43.490 So I have my little list of questions for you, 13:43.490 --> 13:45.230 if you would like to, if you don't mind. 13:45.230 --> 13:45.840 Yes. 13:45.840 --> 13:47.442 So there is a new piece of software 13:47.442 --> 13:48.650 that you might have heard of. 13:48.650 --> 13:49.442 Maybe you have too. 13:49.442 --> 13:51.020 It's called ChatGPT. 13:51.020 --> 13:51.530 Yes. 13:51.530 --> 13:55.760 So what do you think these sorts of new software of approaches 13:55.760 --> 13:59.105 of AI that is so pervasive and uses 13:59.105 --> 14:00.230 deep learning, large angle. 14:00.230 --> 14:04.380 How does that how does this kind of work fit into that world? 14:04.380 --> 14:04.880 Yes. 14:04.880 --> 14:06.980 Well, so I actually started this project 14:06.980 --> 14:08.840 before ChatGPT became famous. 14:08.840 --> 14:12.410 So we were the cool kids before it became cool. 14:12.410 --> 14:15.320 And I think ChatGPT, it's kind of-- 14:15.320 --> 14:18.200 as you can kind of see the process that I described. 14:18.200 --> 14:20.660 As I see people using ChatGPT, I kind of 14:20.660 --> 14:22.950 see this same parallel process happening. 14:22.950 --> 14:25.430 So I don't know if anyone here has used ChatGPT, 14:25.430 --> 14:27.170 but it very much becomes something 14:27.170 --> 14:30.590 where what you give it is very determining of what 14:30.590 --> 14:31.590 comes out of it. 14:31.590 --> 14:33.330 So again, you're kind of designing 14:33.330 --> 14:34.580 for constraints in that sense. 14:34.580 --> 14:37.910 Like, how are you constraining the tool in the ways 14:37.910 --> 14:41.750 you're giving your input to the ChatGPT and the types 14:41.750 --> 14:43.880 of outputs that you can bring. 14:43.880 --> 14:47.720 I think in terms of design ChatGPT, 14:47.720 --> 14:50.220 right now it's very much based on text. 14:50.220 --> 14:54.030 So I think in terms of the hardcore product design 14:54.030 --> 14:54.530 that we do. 14:54.530 --> 14:57.450 It won't really give you a physical design outcome, 14:57.450 --> 14:59.450 which is something that the tools that I studied 14:59.450 --> 15:01.490 would actually be able to do. 15:01.490 --> 15:04.500 But it is good at some things-- like, for example, 15:04.500 --> 15:06.830 writing software. 15:06.830 --> 15:08.690 Again, it's something that-- 15:08.690 --> 15:10.250 so if you're trying to design an app, 15:10.250 --> 15:12.530 you very much could do that in ChatGPT. 15:12.530 --> 15:14.240 But it's something you still have 15:14.240 --> 15:16.460 to make sure that you're checking and making 15:16.460 --> 15:17.750 sure it does-- 15:17.750 --> 15:20.010 it can give you the wrong thing. 15:20.010 --> 15:24.470 So if you just take its word as perfect code, 15:24.470 --> 15:28.160 then you could very much end in making 15:28.160 --> 15:31.778 an app or some type of software that's not quite what 15:31.778 --> 15:32.570 your intention was. 15:32.570 --> 15:36.240 So making sure, again, that we're checking our work 15:36.240 --> 15:38.450 and while using all of these tools. 15:38.450 --> 15:40.740 I think it has that potential. 15:40.740 --> 15:42.860 And I think the fact that a lot of people 15:42.860 --> 15:46.010 are loving how ChatGPT works is kind of an indication where 15:46.010 --> 15:48.260 if you looked at the other tools that maybe do things 15:48.260 --> 15:51.260 like ChatGPT but in different atmospheres, 15:51.260 --> 15:53.810 you'll see that they're very exciting. 15:53.810 --> 15:56.658 And you can definitely use them in other spaces. 15:56.658 --> 15:57.200 That's great. 15:57.200 --> 15:58.970 Thank you. 15:58.970 --> 16:02.430 I have another question for you around design, 16:02.430 --> 16:03.980 not just sort of the ChatGPT world 16:03.980 --> 16:06.560 but more when we think of design. 16:06.560 --> 16:10.130 Often, we think of design being desirable, beautiful, 16:10.130 --> 16:11.930 and visually appealing. 16:11.930 --> 16:14.600 And you kind of referenced that in the wheel hub design, 16:14.600 --> 16:17.263 and also there's lots of researchers at MIT 16:17.263 --> 16:17.930 that think that. 16:17.930 --> 16:20.420 Caitlin Mueller's work comes to mind. 16:20.420 --> 16:23.032 Could you say something about how 16:23.032 --> 16:25.490 the role of visual design and qualitative aspects of design 16:25.490 --> 16:27.260 fit into these sorts of tools? 16:27.260 --> 16:28.200 Yeah, absolutely. 16:28.200 --> 16:31.610 So I studied mainly commercial tools, which I would say 16:31.610 --> 16:35.300 don't have direct input for controlling visual design 16:35.300 --> 16:36.410 or aesthetics. 16:36.410 --> 16:38.480 So from the people that we interviewed, 16:38.480 --> 16:41.300 they actually all mentioned an attempt 16:41.300 --> 16:42.950 to control the aesthetics or trying 16:42.950 --> 16:47.450 to really select and refine based off of desired aesthetic. 16:47.450 --> 16:50.030 But they all did it through their own unique workarounds. 16:50.030 --> 16:52.790 So maybe someone was like, I wanted this beam 16:52.790 --> 16:55.640 to be thicker, so I made my forces heavier so 16:55.640 --> 16:57.380 that it can be a bit thicker. 16:57.380 --> 17:00.950 So trying to be creative in how they control the outputs, 17:00.950 --> 17:02.810 which comes through an understanding 17:02.810 --> 17:06.270 of the tool itself as well and the design problem. 17:06.270 --> 17:08.270 So knowing that this force is what's 17:08.270 --> 17:10.339 going to make this beam thicker. 17:10.339 --> 17:13.040 So again, kind of that iteration that 17:13.040 --> 17:16.250 really helps with the learning of what the constraints are 17:16.250 --> 17:20.150 and how they affect the design outcomes. 17:20.150 --> 17:23.390 It's still very much something that designers themselves 17:23.390 --> 17:24.260 are interested in. 17:24.260 --> 17:26.750 So these tools are very quantitative. 17:26.750 --> 17:28.500 They're very much focused on the numbers, 17:28.500 --> 17:30.375 but the qualitative aspect is very important. 17:30.375 --> 17:32.000 And that's where the designer comes in. 17:32.000 --> 17:34.460 So the designer is really trying to control the parameters 17:34.460 --> 17:35.390 and constraints. 17:35.390 --> 17:37.340 They're refining the product afterward 17:37.340 --> 17:40.440 to really bring in that more qualitative aspect. 17:40.440 --> 17:42.080 And so they're the ones balancing 17:42.080 --> 17:44.000 the quantitative and qualitative to bring 17:44.000 --> 17:46.370 this beautiful aesthetic that they were 17:46.370 --> 17:47.870 trying to go for in the design. 17:47.870 --> 17:49.230 That's good. 17:49.230 --> 17:51.480 So actually, how are we doing on time? 17:51.480 --> 17:51.980 Two minutes. 17:51.980 --> 17:52.160 OK. 17:52.160 --> 17:53.720 I have one more question for you. 17:53.720 --> 17:57.470 So we're all people who are students or educators, 17:57.470 --> 17:59.510 and we think a lot about design in that respect. 17:59.510 --> 18:03.590 How do you think these tools affect the way designers learn 18:03.590 --> 18:04.200 to design? 18:04.200 --> 18:04.700 Right? 18:04.700 --> 18:08.570 If you can just push the button, and then how do you-- 18:08.570 --> 18:09.440 Yeah. 18:09.440 --> 18:11.420 Like I said, I think it's a whole different way 18:11.420 --> 18:12.500 of thinking. 18:12.500 --> 18:16.550 And that's maybe part of why it's not so super widely used 18:16.550 --> 18:19.250 is because a lot of us aren't trained in using 18:19.250 --> 18:20.660 these generative design tools. 18:20.660 --> 18:23.388 I personally never use it in any of my core curriculum at MIT, 18:23.388 --> 18:25.055 despite having been here for nine years. 18:25.055 --> 18:25.980 [CHUCKLES] 18:25.980 --> 18:28.220 So there are some courses, but it's not really 18:28.220 --> 18:30.570 taught in the core curriculum. 18:30.570 --> 18:35.180 And so being able to expose yourself to all of that work 18:35.180 --> 18:37.400 and build that understanding of, oh, I 18:37.400 --> 18:39.998 should be thinking about things in a different way-- 18:39.998 --> 18:41.540 and I think it's important, actually, 18:41.540 --> 18:44.570 and critical for students to be doing that in the classroom, 18:44.570 --> 18:46.280 where it's OK to fail. 18:46.280 --> 18:48.410 And it's better to do that there than it is to do 18:48.410 --> 18:50.778 that in the real world, where there's actually, 18:50.778 --> 18:52.820 a lot of companies who are trying to incorporate. 18:52.820 --> 18:54.890 Especially after ChatGPT, everyone's coming in. 18:54.890 --> 18:57.260 Like, top management's coming down to the developers 18:57.260 --> 18:59.510 and saying, hey, someone incorporate AI into our work. 18:59.510 --> 19:01.802 And everyone's like, well, I don't know how to do that. 19:01.802 --> 19:03.650 So making sure that we're actually 19:03.650 --> 19:06.440 learning how to do that in proper ways 19:06.440 --> 19:08.240 and not just taking the tool. 19:08.240 --> 19:10.100 Again, not taking it for granted, 19:10.100 --> 19:12.680 knowing how to use it to check its work, and making sure 19:12.680 --> 19:14.720 we're designing responsibly. 19:14.720 --> 19:15.555 Thank you. 19:15.555 --> 19:16.055 OK. 19:16.055 --> 19:17.060 [APPLAUSE] 19:17.060 --> 19:18.365 Thank you. 19:18.365 --> 19:20.090 It's fantastic. 19:20.090 --> 19:22.030 Thank you.