Gavin Miller: Adobe Research | Lex Fridman Podcast #23

Lex Fridman 0:00 The following is a conversation with Gavin Miller. He's the head of Adobe Research. Adobe has empowered artists, designers, and creative minds from all professions, working in the digital medium for over 30 years with software such as Photoshop, Illustrator, Premiere, After Effects, InDesign, Audition, software that work with images, video, and audio. Adobe Research is working to define the future evolution of these products in a way that makes the life of creatives easier, automates the tedious tasks, and gives more and more time to operate in the idea space instead of pixel space. This is where the cutting edge, deep learning methods of the past decade can really shine more than perhaps any other application. Gavin is the embodiment of combining tech and creativity. Outside of Adobe Research, he writes poetry and builds robots, both things that are near and dear to my heart as well. This conversation is part of the Artificial Intelligence podcast. If you enjoy it, subscribe on YouTube, iTunes, or simply connect with me on Twitter at Lex Fridman, spelled F-R-I-D. And now, here's my conversation with Gavin Miller. You're head of Adobe Research, leading a lot of innovative efforts in applications of AI, creating images, video, audio, language, but you're also yourself an artist, a poet, a writer, and even a roboticist. So while I promise to everyone listening that I will not spend the entire time we have together reading your poetry, which I love-

Gavin Miller 1:33 Oh. (laughs)

Lex Fridman 1:33 ... I have to sprinkle it in at least a little bit. So some of them are pretty deep and profound, and some are light and silly. Let's start with a few lines from the, uh, the silly variety. You write, uh, in, uh, Je ne regrette rien, a poem that beautifully parodies both, uh, Edith Piaf's Je Ne Regrette Rien and My Way by Frank Sinatra. So it opens with, "And now dessert is near. It's time to pay the final total. I've tried to slim all year, but my diets have been anecdotal." So where does that love for poetry come from for you? And if we dissect your mind-

Gavin Miller 2:17 Mm-hmm.

Lex Fridman 2:17 ... how does it all fit together in the bigger puzzle of Dr. Gavin Miller?

Gavin Miller 2:22 Oh. Well, interesting you chose that one. That was a poem I wrote when I'd been to my doctor and he said, "You really need to lose some weight" and, "Go on a diet."

Lex Fridman 2:30 (laughs)

Gavin Miller 2:31 And whilst the rational part of my brain wanted to do that, uh, the irrational part of my brain was protesting and sort of embraced the opposite idea.

Lex Fridman 2:39 I regret nothing, hence...

Gavin Miller 2:40 Yes, exactly. Taken to an extreme, I thought it would be funny. Obviously, it's a serious topic for so- some people, but, um, uh... I think for me, I've always been interested in writing since I was in high school, as well as doing technology and invention. And sometimes the parallel strands in your life that carry on and, you know, one is more about your private life and one's more about your technological, um, career, and then at sort of happy moments along the way, sometimes the two things touch, one idea informs the other, and, um, we can talk about that, uh, as we go.

Lex Fridman 3:15 Do you think your writing, the art, the poetry contribute indirectly or directly to your research, to your work in Adobe?

Gavin Miller 3:22 Well, sometimes it does if I, say, imagine a future in a science fiction kind of way, and then once it exists on paper I think, "Well, why shouldn't I just build that?" There was an example where when realistic voice synthesis first started in the '90s at Apple where I worked in research, uh, it was done by a friend of mine, um, I sort of sat down and started writing a poem which each line I would enter into the voice synthesizer and see how it sounded-

Lex Fridman 3:50 Hmm.

Gavin Miller 3:51 ... and sort of wrote it for that voice. And at the time, the agents weren't very sophisticated, so they'd sort of add random intonation and I kind of made up the poem to sort of match the tone of the voice, and it sounded slightly sad and depressed, so I, I pretended it was a poem written by an intelligent agent, uh, sort of telling the user to go home and leave them alone, but at the same time, they were lonely and wanted to have company and learn from what the user was saying. And at the time, it was way beyond anything that AI could possibly do, but, you know, since then, it's becoming more within the bounds of possibility. And then at the same time, I had a project at home where I did sort of a smart home. Uh, this was probably '93, '94. And, uh, I had the talking voice who'd remind me when I walked in the door of what things I had to do. I had buttons on my washing machine 'cause I was a bachelor and I'd leave the clothes in there for three days and they'd go moldy, so as I got up in the morning, it would say, "Don't forget the washing," and so on. Um, I made photographic photo albums that used light sensors to know which page you were looking at, would send that over w- wireless radio to the agent who would then play sounds that matched the image you were looking at in the book. So I was kind of in love with this idea of magical realism and whether it was possible to do e- that with technology. So that was a case where the sort of the agent sort of intrigued me from a literary point of view and became a personality. I think more recently, I've also written plays, and when plays you write dialogue, and obviously you write a fixed set of dialogue that follows a linear narrative. But with modern agents, as you design a personality or a capability for conversation, you're sort of thinking of, "I kind of have imaginary dialogue in my head," and then I think, "What would it take not only to have that be real, but for it to really know what it's talking about?"

Lex Fridman 5:44 Mm-hmm.

Gavin Miller 5:44 So it's easy to fall into the uncanny valley with AI where it says something it doesn't really understand, but it sounds good to the person, but you rapidly realize that it's kind of just stimulus response. It doesn't really have real world knowledge about the thing it's describing. And so...... when you get to that point, it really needs to have multiple ways of talking about the same concept so it sounds as though it really understands it. Now, what really understanding means is in the eye-

Lex Fridman 6:11 That's right.

Gavin Miller 6:11 ... of the beholder, right? But if it only has one way of referring to something, it feels like it's a canned response, but if it can reason about it or you can go at it from multiple angles and give a similar kind of response that people would, then it starts to seem more like there's something there that's sentient. Um-

Lex Fridman 6:31 It can say the same thing, multiple things from different perspectives. I mean, with the, the automatic image captioning that I've seen the work that you're doing.

Gavin Miller 6:38 Yes.

Lex Fridman 6:39 There's elements of that, right? Being able to generate different kinds of-

Gavin Miller 6:42 Right.

Lex Fridman 6:42 ... statements about the same picture.

Gavin Miller 6:43 So one, in my team, there's a lot of work on turning a medium from one form to another, whether it's auto-tagging imagery or making up full sentences about what's in the image, um, then changing the sentence, finding another image that matches the new sentence or vice versa. Um, and, uh, in the modern world of GANs, you sort of give it a description and it synthesizes, uh, an asset that matches the description, so. I've sort of gone on a journey. My early days in my career were about 3D computer graphics, the sort of pioneering work sort of before movies had special effects done with 3D graphics, and sort of w- rode that revolution in. That was very much like the Renaissance, where people would model light and color and shape and everything. And now we're kind of in another wave where it's more impressionistic and it's sort of the idea of something can be used to generate an image directly which is, um, sort of the new frontier in, uh, computer image generation using AI algorithms, so.

Lex Fridman 7:43 So the creative process is more in the space of ideas, or becoming more in the space of ideas versus in the raw pixels?

Gavin Miller 7:50 Well, it's interesting. It depends. I think at Adobe we really want to span the entire range from really, really good, what you might call low-level tools, by low level, as close to say analog workflows as possible, so. What we do there is we make up systems that do really realistic oil paint and watercolor simulations, so if you want every bristle to behave as it would in the real world and leave a beautiful analog trail of water and then flow after you've made the brushstroke, you can do that, and that's really important for people who want to create something really expressive or really novel, 'cause they have complete control. And then as certain other tasks become automated, it frees the artist up to focus on the inspiration and less of the perspiration, so, um, thinking about different ideas obviously. Once you've finished the design, there's a lot of work to say, do it for all the different aspect ratio of phones or websites and so on, and that used to take up an awful lot of time for artists. It still does for many, what we call content velocity. And one of the targets of AI is actually to reason about from the first example of what are the likely intent for these other formats, maybe if you change the language to German and the l- words are longer, how do you reflow everything so that it looks nicely artistic in that way? And so the person can focus on the really creative bit in the middle which is, what is the look and style and feel and what's the message and what's the story and the human element? So I think creativity is changing. So that's one way in which we're trying to just make it easier and faster and cheaper to do so that there can be more of it, more demand, 'cause it's less expensive, so everyone wants beautiful artwork for everything from a school website to Hollywood movie. On the other side is some of these things have automatic versions of them. People will possibly change role from being the hands-on artisan to being either the art director or the conceptual artist, and then the computer will be a partner to help create polished examples of the idea that they're exploring.

Lex Fridman 10:00 Let's talk about Adobe products for a second.

Gavin Miller 10:01 Right.

Lex Fridman 10:01 AI and Adobe products.

Gavin Miller 10:03 Yeah.

Lex Fridman 10:03 Uh, just to kn- just so you know where I'm coming from, I'm a huge fan of Photoshop for images.

Gavin Miller 10:09 Mm-hmm.

Lex Fridman 10:09 Premiere for video, Audition for audio. I probably use Photoshop to create the thumbnail for this video.

Gavin Miller 10:16 Mm-hmm.

Lex Fridman 10:16 Premiere to edit the video, Audition to do the audio. That said, everything I do is really manually, and I set up, I use this old school Kinesis keyboard and I have auto hotkey that just, it's really about optimizing the, the flow.

Gavin Miller 10:32 Mm-hmm.

Lex Fridman 10:33 Of, of just making sure there's as few clicks as possible, so just being extremely efficient. It's something you started to speak to.

Gavin Miller 10:39 Right.

Lex Fridman 10:40 Uh, so before we get into the fun, uh, sort of awesome-

Gavin Miller 10:44 Yeah.

Lex Fridman 10:44 ... deep learning things-

Gavin Miller 10:45 Mm-hmm.

Lex Fridman 10:45 ... where does AI, if you could speak a little more to it, AI or just automation in general, uh, do you see in the coming months and years, or in general, prior in 2018, uh, fitting in to making the life, the low-level pixel work, uh, flow easier?

Gavin Miller 11:04 Yeah, that's a great question. So we have a very rich array of algorithms already in Photoshop, just classical procedural algorithms as well as ones based on data. In some cases, they end up with a large number of sliders and degrees of freedom, so one way in which AI can help is just an auto button which comes up with default settings based on the content itself rather than default values for the tool.

Lex Fridman 11:29 Mm-hmm.

Gavin Miller 11:29 At that point you then start tweaking, so that's, that's a very kind of make life easier for people whilst making use of common sense from other example images.

Lex Fridman 11:39 So like smart defaults

Narrator 11:41 Yeah.

Gavin Miller 11:41 Absolutely.

Lex Fridman 11:42 That's great.

Gavin Miller 11:43 Um, another one is something we've spent a lot of work over the last 20 years I've been at Adobe, or 19, thinking about, um, selection for instance, where, um, you know, with, uh, quick select you would look at color boundaries and figure out how to sort of flood fill into regions that you thought were physically connected in the real world.... but that algorithm had no visual common sense about what a cat looks like, or a dog. It would just do it based on rules of thumb which were applied to graph theory. And it was a big improvement over the previous work where you had to sort of almost click every- everything by hand, or if it just did similar colors, it would do little tiny regions that wouldn't be connected. But in the future, using neural nets to actually do a great job with, say, a single click or even in the case of well-known categories like people or animals, no click, where you just say, "Select the object," and it just knows the dominant object is a person in the middle of a photograph. Um, those kinds of things are, um, really valuable if they can be robust enough to give you good quality results. Um, or they-

Lex Fridman 12:49 Are-

Gavin Miller 12:49 ... can be a great start for, like, tweaking it.

Lex Fridman 12:52 So, for example, background removal.

Gavin Miller 12:54 Correct.

Lex Fridman 12:54 Like, uh, one thing I'll, I'll... In, in a thumbnail I'll take a picture of you right now and essentially remove the background behind you.

Gavin Miller 13:01 Correct.

Lex Fridman 13:01 And I want to make that as easy as possible. Now, you don't have flowing hair, uh, like rich-

Gavin Miller 13:07 Sadly, at the moment, yes.

Lex Fridman 13:08 ... rich sort of hair, but-

Gavin Miller 13:09 I had it in the past, it may come again in the future, but for now... (laughs)

Lex Fridman 13:13 (laughs) So that sometimes makes it a little bit more challenging to remove the background. How difficult do you think is that problem for AI for, for basically making the quick selection tool smarter and smarter and smarter?

Gavin Miller 13:25 Well, we have a lot of research on that already. If you want a sort of quick, cheap and cheerful look, I'm pretending I'm in Hawaii, but it's sort of a joke-

Lex Fridman 13:35 Mm-hmm.

Gavin Miller 13:35 ... then you don't need perfect boundaries, and you can do that today with a single click with the algorithms we have. We have other algorithms where with a little bit more guidance on the boundaries, like you might need to touch it up a little bit. We have other algorithms that can pull a nice matte from a accrued selection. So we have combinations of tools that can do all of that. And at our recent MAX Conference, Adobe MAX, we demonstrated how very quickly just by drawing a simple polygon around the object of interest, we could not only do it for a, a single still, but we could pull, uh, a matte ... well, pull at least a selection mask from a moving target like a person dancing in front of a brick wall or something.

Lex Fridman 14:21 So we-

Gavin Miller 14:21 And so it's going from hours to a few seconds for workflows that are really nice. And then you might go in and touch up a little.

Lex Fridman 14:30 So-

Gavin Miller 14:30 Yeah.

Lex Fridman 14:30 ... that's a really interesting question. You mentioned the word "robust".

Gavin Miller 14:33 Mm-hmm.

Lex Fridman 14:34 Uh, you know, there, there's like a journey for an idea, right? And, uh, what you've presented probably at MAX has elements of just sort of it inspires the concept, it can work pretty well in ma- in the majority of cases. But how do you make something that works, well, in majority of cases? How do you make something that works maybe in all cases, or is... becomes a robust tool that can-

Gavin Miller 14:56 Well, there are, there are a couple of things. So that really touches on the difference between academic research and industrial research.

Lex Fridman 15:03 Right.

Gavin Miller 15:03 So in academic research, it's really about who's the person to have the great new idea that shows promise, and we certainly love to be those people too. But we have sort of two forms of publishing. One is academic peer review, which we do a lot of and we have great success there, as much as some universities. Um, but then we also have shipping, which is a different type of review.

Lex Fridman 15:26 (laughs)

Gavin Miller 15:26 And then we get customer review-

Lex Fridman 15:29 Yeah.

Gavin Miller 15:29 ... as well as, you know, product critics. And that might be a case where it's not about being perfect every single time, but perfect enough of the time, plus a mechanism to intervene and recover where you do have mistakes. So we have the luxury of very talented customers. We don't want them to be overly taxed doing it every time, but if they can go in and just take it from 99 to 100 with the, the touch of a, of a mouse or something, then for the professional end, that's something that we definitely want to support as well. And for them, it went from having to do that tedious task all the time to much less often, so I- I think that gives us an out. If it had to be 100% automatic all the time, then that would delay the time at which we could get to market. Um...

Lex Fridman 16:19 So on that thread, maybe you can untangle something. Um, again, I'm sort of just speaking to my own experience. (laughs)

Gavin Miller 16:27 Yeah. No, that's fine.

Lex Fridman 16:29 (laughs) Maybe that is the most useful idea.

Gavin Miller 16:30 Absolutely.

Lex Fridman 16:32 So I think Photoshop, as per- an example, or Premier, uh, has a lot of amazing features that I haven't touched.

Gavin Miller 16:41 Mm-hmm.

Lex Fridman 16:42 And so what's the, in terms of AI helping make my life or the, uh, the, the life of creatives easier, how... This collaboration between human and machine, how, how do you learn to collaborate better? How do you learn the new algorithms? Um, is it something where you have to watch tutorials and you have to watch videos and so on? Or do you, do you ever think of... Do you ever, do you think about the experience itself through exploration, being the teacher?

Gavin Miller 17:11 We absolutely do. So we... I'm glad that you, uh, brought this up. We sort of think about two things. One is helping the person in the moment to do the task that they need to do. But the other is thinking more holistically about their journey learning a tool, and whether it's like... Think of it as Adobe University, where you use the tool long enough, you become an expert. And not necessarily an expert in everything. It's like living in a city, you don't necessarily know every street, but you know the important ones you need to get to. So we have projects in research which actually look at the thousands of hours of tutorials online and try to understand what's being taught in them. And then, we had one publication at CHI where it was looking at, given the last three or four actions you did, what did other people in tutorials do next? So if you want some inspiration for what you might do next, or you just want to watch the tutorial and see, learn from people who are doing similar workflows to you, you can without having to go and search on keywords and everything. So really trying to use the context of your use of the app to make intelligent suggestions, either about choices that you might make...... or in a more assistive way where it could say, "If you did this next, we could show you." And that's basically the frontier that we're exploring now, which is if we really deeply understand the domain in which designers and creative people work, can we combine that with AI and patent-matching of behavior to make intelligent suggestions? Either through, you know, verbal possibilities or just showing the results of if you try this. And that's, that's really the, sort of the ... You know, I was in a meeting today-

Lex Fridman 18:54 (laughs)

Gavin Miller 18:54 ... thinking about these things.

Lex Fridman 18:55 Wow (overlapping)

Gavin Miller 18:55 So it's, it's still a grand challenge. You know, we'd all love an artist over one shoulder and a teacher over the other, right?

Lex Fridman 19:02 Right. Yeah.

Gavin Miller 19:02 And, uh, we hope to get there. And the right thing to do is to give enough at each stage that it's useful in itself, but it builds a foundation for the next level of expectation.

Lex Fridman 19:14 Are you aware of this gigantic medium of YouTube that's creating, uh ... just a bunch of creative people, both artists and teachers of different kinds?

Gavin Miller 19:26 Absolutely. And we, the more we can understand those media types, both visually and in terms of transcripts and words, the more we can bring the wisdom that they embody into the guidance that's embedded in the tool.

Lex Fridman 19:38 That would be brilliant to-

Gavin Miller 19:39 Yeah.

Lex Fridman 19:39 ... to, uh, to remove the barrier from having to, yourself, type in the keyword searching, so on.

Gavin Miller 19:45 Absolutely. And then in the longer term, an interesting discussion is w- does it ultimately not just assist with learning the interface we have, but does it modify the interface to be simpler? Or do you fragment into a variety of tools, each of which has a different level of visibility of the functionality? I like to say that if you add a, if you add a feature to a GUI, you have to have yet more visual complexity confronting the new user. Whereas if you have an assistant with a new skill, if you know they have it, so you know to ask for it, then it's sort of additive without being more intimidating. So, we definitely think about new users and how to onboard them. Um, many actually value the idea of being able to master that complex interface and keyboard shortcuts like you were talking about earlier, because with great familiarity, it becomes a musical instrument for expressing your visual ideas. And, um, other people just want to get something done quickly in the simplest way possible, and that's where a more assistive version of the same technology might be useful. Maybe on a different class of device, which is more in context for capture, say. Whereas somebody who's in a deep post-production workflow maybe want to be on a laptop or a big screen desktop and have more knobs and dials to really express the subtlety of what they want to do.

Lex Fridman 21:12 So, there's so many exciting applications of computer vision and machine learning that Adobe is working on.

Gavin Miller 21:17 Mm-hmm.

Lex Fridman 21:18 Like scene stitching, sky replacement, foreground/background removal, spatial object-based image search, automatic image captioning, like we mentioned, Project Cloak, Project Deep Fill, filling in parts of the images.

Gavin Miller 21:31 Yep.

Lex Fridman 21:32 Project Scribbler, style transfer video, style transform faces and video with, uh, Project Puppetron. Best name ever.

Gavin Miller 21:39 Mm-hmm.

Lex Fridman 21:40 Can you, uh, talk through s- a favorite or some of them, or examples that pop to, uh, in mind? And I'm sure I'll be able to provide links to, um, to all the ones we don't talk about 'cause they ... you know, there's visual elements to all of them (laughs) that, uh, that are exciting.

Gavin Miller 21:59 Why they're interesting for different reasons-

Lex Fridman 22:00 Right.

Gavin Miller 22:00 ... might be a good way to go. So, I think Sky Replace is interesting because, you know, we talked about selection being sort of an atomic operation. It's almost like a, if you think of an assembly language, it's like a single instruction.

Lex Fridman 22:13 Mm-hmm.

Gavin Miller 22:14 Whereas Sky Replace is a compound action where you automatically select the sky, you look for stock content that matches the geometry of the scene. You try to have variety in your choices so that you do coverage of different moods. It then, um, mattes in the sky behind the foreground. But then importantly, it uses the foreground of the other image that you just searched on to recolor the foreground of the image that you're editing. So, if you say, go from a midday sky to an evening sky, it will actually m- add sort of an orange glow to the foreground objects as well. I was a big fan in college of Magritte, and he has a number of paintings where it's surrealism because he'll like do a composite, but the foreground building will be at night and the sky will be during the day. There's one called The Empire of Light, which was on my wall in college. And we're trying not to do surrealism. It can be a choice-

Lex Fridman 23:10 Yeah.

Gavin Miller 23:10 ... but we'd rather have it be natural by default rather than it looking fake and then you have to do a whole bunch of post-production to fix it.

Lex Fridman 23:18 Is that-

Gavin Miller 23:18 So that's a case where we're kind of capturing an entire workflow into a single action and doing it in about a second rather than a minute or two. And when you do that, you can not just do it once, but you can do it for say like ten different backgrounds, and then you're almost back to this inspiration idea of, "I don't know quite what I want, but I'll know it when I see it." And you can just explore the design space as close to final production value as possible. And then when you really pick one, you might go back and slightly tweak the selection mask just to make it perfect and do that kind of polish that professionals like to bring to their work.

Lex Fridman 23:55 So then, then there's this idea of ... so you mentioned the sky, replacing it-

Gavin Miller 23:59 Yeah.

Lex Fridman 23:59 ... to different stock images of the sky.

Gavin Miller 24:01 Yeah.

Lex Fridman 24:01 And generally you have this idea of-

Gavin Miller 24:03 Or it could be on your disc or whatever, but-

Lex Fridman 24:04 Your disc, right.

Gavin Miller 24:05 Yeah. Yeah.

Lex Fridman 24:05 But making even more intelligent choices about ways to search stock images, which is really interesting.

Gavin Miller 24:12 Right.

Lex Fridman 24:12 This kind of spatial-

Gavin Miller 24:13 Absolutely.

Lex Fridman 24:14 ... being able to specify-

Gavin Miller 24:14 You're right. So that was something we called, um, Concept Canvas. So normally when you do a, a s-, uh, say an image search, you would, um, I ... Assuming it's just based on text, you would give the keywords of the things you want to be in the image and it would find the nearest one that had those tags.... for many tasks, you really want, you know, to be able to say, "I want a big person in the middle and a dog to the right and an umbrella above the left," 'cause you want to leave space for the text or whatever-

Lex Fridman 24:41 Mm-hmm.

Gavin Miller 24:42 ... for the... And so, Concept Canvas lets you assign spatial regions to the keywords. And then we've already pre-indexed the images to know where the important concepts are in the picture, so we then go through that index matching to, um, assets. And even though it's just another form of search, because you're doing spatial design or layout, it starts to feel like design.

Lex Fridman 25:06 (laughs)

Gavin Miller 25:06 You sort of feel oddly responsible for the image that comes back as if you invented it a little bit.

Lex Fridman 25:10 Yeah.

Gavin Miller 25:10 So, it, it's, um, it's a good example where giving enough control starts to make people have a sense of ownership over the outcome of the event. And then we also have technologies in Photoshop where you physically can move the dog in post as well.

Lex Fridman 25:25 Mm-hmm.

Gavin Miller 25:25 But, um, but for Concept Canvas, it was just a very fast way to sort of loop through and be able to lay things out.

Lex Fridman 25:33 Uh, in, in terms of being able to remove objects from a scene-

Gavin Miller 25:38 Yeah.

Lex Fridman 25:39 ... and fill in the background-

Gavin Miller 25:40 Right.

Lex Fridman 25:40 ... automatically, I, so that's extremely exciting and that's ... so neural networks are stepping in there?

Gavin Miller 25:48 Mm-hmm.

Lex Fridman 25:48 I just talked this week-

Gavin Miller 25:50 Right.

Lex Fridman 25:50 ... with Ian Goodfellow. So the guy-

Gavin Miller 25:52 Yes.

Lex Fridman 25:52 (laughs)

Gavin Miller 25:52 The GANs for doing that-

Lex Fridman 25:53 He's-

Gavin Miller 25:54 ... is definitely one approach.

Lex Fridman 25:55 Uh, so that, is that, is that a really difficult problem? Is it as difficult as it looks, again, to take it to a robust product level?

Gavin Miller 26:03 Well, there are certain classes of image for which the traditional algorithms, like content-aware fill, work really well, like if you have a naturalistic texture like a gravel path or something. Because it's patch-based, it will make up a very plausible looking intermediate thing and fill in the hole. And then we use some, uh, algorithms to sort of smooth out the lighting so that you don't see any brightness contrasts in that region, or you've gradually ramped from one ... from dark to light if it straddles a boundary. Where it gets complicated is if you have to infer invisible structure behind-

Lex Fridman 26:35 Right.

Gavin Miller 26:36 ... behind the person in front, and that really requires a common sense knowledge of the world to know what... You know, if I see three-quarters of a house, do I have a rough sense of what the rest of the house looks like? If you just fill it in with patches, it can end up sort of doing things that make sense locally, but you look at the global structure and it looks like it's just sort of crumpled or messed up. And so, what GANs and neural nets bring to the table is this common sense learnt from the training set. And, um, the challenge right now is that the generative methods that can make up missing holes using that kind of technology are still only stable at low resolutions. And so, you either need to then go from a low resolution to a high resolution using some other algorithm, or we need to push the state-of-the-art, and it's still in research to get to that point. Um, of course, if you show it something, say it's trained on houses, and then you show it an octopus (laughs) -

Lex Fridman 27:32 (laughs)

Gavin Miller 27:32 ... it's not gonna do a very good job of showing common sense about octopuses. So, again, that ... so you were asking about how you know that it's ready for primetime-

Lex Fridman 27:44 Yeah.

Gavin Miller 27:44 ... you really need a very diverse training set of images. And ultimately, that may be a case where you put it out there with some guide- guard rails where, um, you might do a detector which looks at the image and sort of estimates its own competence of how well a job could this algorithm do.

Lex Fridman 28:05 Right.

Gavin Miller 28:06 So eventually, there may be this idea of what we call an ensemble of experts, where any particular expert is specialized in certain things, and then there's sort of a ... either they vote to say how confident they are about what to do-

Lex Fridman 28:18 Mm-hmm.

Gavin Miller 28:18 ... this is sort of more for- future-looking, or there's some, uh, dispatcher which says, "You're good at houses, you're good at trees." Uh-

Lex Fridman 28:25 (laughs)

Gavin Miller 28:25 ... and, uh-

Lex Fridman 28:26 Sure.

Gavin Miller 28:26 ... uh, so, so I mean, it's ... all this adds up to a lot of work 'cause each of those models would be a whole bunch of work. But I, I think over time, um, you'd gradually fill out the set, and, uh, initially focus on certain workflows and then sort of branch out as you get more capable.

Lex Fridman 28:42 So you mentioned workflows, and, um, have you considered maybe looking f- far into the future, uh, first of all using the, the fact that there is s- a huge amount of people that use Photoshop, for example-

Gavin Miller 28:58 Mm-hmm.

Lex Fridman 28:58 ... and have certain workflows-

Gavin Miller 29:00 Yeah.

Lex Fridman 29:00 ... being able to collect the information by which they ... you know, basically get information about their workflows (laughs) , about what they need, what can ... the ways to help them, w- whether it is houses or octopus that people work on more?

Gavin Miller 29:15 Right.

Lex Fridman 29:15 You know, like basically getting a beat on what kind of, uh, data is needed to be annotated and collected for people to, uh, to build tools that actually work well for people.

Gavin Miller 29:26 Right, absolutely. And this is a big topic in the whole world of AI, is what data can you gather and why?

Lex Fridman 29:32 Right.

Gavin Miller 29:33 At one level, the way to think about it is we not only want to train our customers in how to use our products, but we want them to teach us what's important-

Lex Fridman 29:42 Exactly.

Gavin Miller 29:42 ... and what's useful. At the same time, we want to respect their privacy.

Lex Fridman 29:46 Right.

Gavin Miller 29:46 And obviously we wouldn't do things without their explicit permission. And I think the modern spirit of the age around this is you have to demonstrate to somebody how they're benefiting from sharing their data with, with, with the tool. Either it's helping in the short term to understand their intent so you can make better recommendations, or if they're friendly to your cause or your tool, or they want to help you evolve quickly 'cause they depend on you for their livelihood, uh, they may be willing to share some of their, um, workflows or, um, choices with, with the dataset to be then trained. There are technologies for looking at learning without necessarily storing all the information permanently so that you can sort of learn on the fly but not, um, keep a record of what somebody did. So, we're definitely exploring all of those possibilities.

Lex Fridman 30:39 And I think Adobe exists in a space where Photoshop-Like, if, if I look at the data I've created and own, you know, I'm mo- mo- less comfortable sharing data with social networks than I am with Adobe because there's a g- a, just exactly as you said, uh, there's an obvious, uh, benefit for sharing the data that I use to create in Photoshop because it's helping improve the workflow in the future.

Gavin Miller 31:06 Right.

Lex Fridman 31:06 As opposed to it's not clear what the benefit is in social networks.

Gavin Miller 31:10 It's nice of you to say that. I mean, I think there are some professional workflows where people might be very protective of what they're doing, such as if I was preparing evidence for a legal case, I wouldn't want (laughs) any, any of that, you know, phoning home to help train the algorithm or anything. Um, there may be other cases where people are, say, having a trial version or they're doing some... I'm not saying we're doing this today, but there's a future scenario where somebody has a more permissive relationship with Adobe where they explicitly say, "I'm fine. I'm only doing hobby projects or things which are non-confidential, um, and in exchange for some benefit, um, tangible or otherwise, I'm willing to share very fine-grain data." So, another possible scenario is to capture relatively crude high-level things from more people and then more detailed knowledge from people who are willing to participate. We do that today with explicit customer studies where, you know, we go and visit somebody and ask them to try the tool and we human observe what they're doing. Um, in the future, to be able to do that enough to be able to train an algorithm, we'd need a more systematic process, but we'd have to do it very consciously because one of the things people treasure about Adobe is a sense of trust and we don't want to endanger that through overly aggressive data collection. So, uh, we have a chief privacy officer and it's definitely front and center of thinking about AI rather than an afterthought.

Lex Fridman 32:37 Well, when you start that program, sign me up. I'm-

Gavin Miller 32:40 Okay. Happy to. (laughs)

Lex Fridman 32:41 (laughs) Is, is there other projects that you wanted to mention that, um, that I didn't perhaps that pop into mind?

Gavin Miller 32:48 Well, you covered a number. I think you mentioned Project Puppetron.

Lex Fridman 32:52 Puppetron.

Gavin Miller 32:52 I think that one is interesting because it's, um... You might think of Adobe as only thinking in 2D, um, and that's a good example where we're actually thinking more three-dimensionally about how to assign features to faces so that we can... You know, if you take... So, what Puppetron does, it takes, um, either a still or a video of a person talking and then it can take a painting of somebody else and then apply the style of the painting to the person who's talking in the video. And, um, it... Unlike a sort of screen door post filter effect that you sometimes see online, um, it really looks as though it's sort of a- somehow attached or reflecting the motion of the face. And so that's a case where even to do a 2D workflow like stylization you really need to infer more about the 3D structure of the world, and I think as 3D computer vision algorithms get better, initially they'll focus on particular domains like faces where you have a lot of prior knowledge about structure and you can maybe have a parameterized template that you fit to the image.

Lex Fridman 33:58 Mm-hmm.

Gavin Miller 33:58 But over time this should be possible for more general content. And it might even be invisible to the user that you're doing 3D reconstruction, but... Under the hood, but it might then let you, um, do edits much more reliably or correctly than you would otherwise.

Lex Fridman 34:16 And, y- you know, face is a very important application, right? So-

Gavin Miller 34:21 Absolutely.

Lex Fridman 34:21 ... um, making things work-

Gavin Miller 34:22 And a very sensitive one. If you do something uncanny, it's very disturbing.

Lex Fridman 34:26 That's right.

Gavin Miller 34:27 Yeah.

Lex Fridman 34:27 (laughs) Because-

Gavin Miller 34:28 Yeah.

Lex Fridman 34:28 ... you have to get it, you have to get it right. So, um, in the space of augmented reality and virtual reality-

Gavin Miller 34:36 Yes.

Lex Fridman 34:37 ... uh, what do you think is the role of AR and VR in, in the content we consume as people, as consumers, and the content we create as creators today?

Gavin Miller 34:45 Now, that's a great question. We, we think about this a lot too. So, I think VR and AR serve slightly different purposes. So, VR can really transport you to an entire immersive world, um, no matter what your personal situation is. To that extent, it's a bit like a really, really widescreen television where it sort of snaps you out of your context and puts you in a new one, and I think it's still evolving in terms of the hardware. I actually worked on VR in the '90s trying to solve the latency and sort of nausea problem, which we did, but it was very expensive and a bit early. Um, there's a new wave of that now, I think, uh, and increasingly those devices are becoming all-in-one rather than something that's tethered to a box. I think, uh, the market seems to be bifurcating into things for consumers and things for professional use cases, like for architects and people designing where your product is a building and you really want to experience it better than looking at a scale model or a drawing, I think, um, e- or even than a video. So, I think for that where you need a sense of scale and spatial relationships, it's great.

Lex Fridman 35:52 Mm-hmm.

Gavin Miller 35:52 I think AR holds the promise of sort of taking digital assets off the screen and putting them in context in the real world on the table in front of you, on the wall behind you, and that has the corresponding need that the assets need to adapt to the physical context in which they're being placed. I mean, it's... It's a bit like having a live theater troupe come, come to your house and put on Hamlet. Uh-

Lex Fridman 36:18 (laughs) .

Gavin Miller 36:18 ... my mother had a friend who used to do this at stately homes in England, uh, for the National Trust, and they would adapt the scenes and even... They'd walk the audience through the rooms to see the action based on the country house they found themselves in for two days, and I think AR will have the same issue that, you know, if you have a tiny table in a big living room or something, it'll try to figure out what can you change and what's fixed, um, and there's a little bit of a tension between fidelity where...... if you captured, say, Nureyev doing a fantastic ballet, you'd want it to be sort of exactly reproduced and maybe all you could do is scale it down. Whereas somebody telling you a story might be walking around the room doing some gestures, and that could adapt to the room in which they were telling the story, so-

Lex Fridman 37:07 And do you think fidelity is that important in that space? Or is it more about the storytelling?

Gavin Miller 37:13 I think it may depend on the characteristic of the media. If it's a famous celebrity, then it may be that you want to catch every nuance-

Lex Fridman 37:20 (laughs)

Gavin Miller 37:20 ... and they don't want to be reanimated by some algorithm. It could be that if it's really, you know, a lovable frog telling you a story (laughs) , um, and it's about a princess and a frog, then it doesn't matter if the frog moves in a different way. Um, I think a lot of the ideas that have sort of grown up in the game world will now come into the broader commercial sphere once they're needing adaptive characters in AR.

Lex Fridman 37:46 Are you thinking of engineering tools that allow creators to create in the augmented world? Basically making a Photoshop for the augmented world? (laughs)

Gavin Miller 37:57 Well, we have shown a few demos of sort of taking a Photoshop layer stack and then expanding it into 3D, that's actually been shown publicly as one example in AR. Where we're particularly excited at the moment is in 3D. 3D design is still a very challenging space, and we believe that it's a worthwhile experiment to try to figure out if AR or immersive makes 3D design more spontaneous.

Lex Fridman 38:23 Can you give me an example of 3D design, just-

Gavin Miller 38:26 Well-

Lex Fridman 38:26 ... like applications we can talk about?

Gavin Miller 38:27 ... literally, well, a simple one would be laying out objects, right? So, on a conventional screen you'd sort of have a plan view and a side view and a perspective view, and you'd sort of be dragging it around with a mouse and if you're not careful it would go through the wall and all that.

Lex Fridman 38:39 Right.

Gavin Miller 38:39 Whereas if you were really laying out objects in an, say, in a VR headset, you could literally move your head to see a different viewpoint. They'd be in stereo so you'd have a sense of depth 'cause you're already wearing the depth glasses, right, uh, in there. So it would be a, those sort of big, gross motor, move things around kind of skills seem much more spontaneous, just like they are in the real world. Um, the, the frontier for us, I think, is whether that same medium can be used to do fine-grain design tasks, like very accurate, you know, constraints on, say, a CAD model or something. That may be m- better done on a desktop, but it may just be a matter of inventing the right UI. So, we're hopeful that, um, because there will be this potential explosion of demand for 3D assets driven by AR and more real-time animation on conventional screens, that that will, those tools will also help with, uh, or those devices will help with designing the content as well.

Lex Fridman 39:41 You've mentioned quite a few interesting sort of new ideas. Do you, uh... And at the same time, there's old-timers like me that are stuck in their old ways and, uh, uh, are-

Gavin Miller 39:50 Well, I think I'm the old-timer (laughs) .

Lex Fridman 39:51 Okay. All right. All right. But the, uh, opposed all change at all costs kind of-

Gavin Miller 39:55 Yes.

Lex Fridman 39:56 ... is there, um... When you're thinking about creating new interfaces, do you feel the burden of just this giant user base that loves the current product? And so anything new you do that, uh, any new idea comes at a cost that you'll be resisted?

Gavin Miller 40:15 Well, I think if you have to trade off control for convenience, then our existing user base would definitely be offended by that. I think if there are some things where you have more convenience and just as much control, that may be more welcome. We do think about not breaking well-known metaphors for things.

Lex Fridman 40:35 Mm-hmm.

Gavin Miller 40:36 Um, so things should sort of make sense. Um, Photoshop has never been a static target, it's always been evolving and growing, and to some extent there's been a lot of brilliant thought along the way of how it works today, so we don't want to just throw all that out. If there's a fundamental breakthrough, like a single click is good enough to select an object rather than having to do lots of strokes, that actually fits in quite nicely to the existing tool set, either as an optional mode or as a starting point. I think where we're looking at radical simplicity, where you could encapsulate an entire workflow with a much simpler UI, then sometimes that's easier to do in the context of either a different device, like a mobile device where the affordances are naturally different, or in a tool that's targeted at a, a different workflow where it's about spontaneity and velocity rather than precision. And we have projects like Rush which can let you do professional-quality video editing for a certain class of media output that is targeted very differently, uh, in terms of users and, uh, the experience. And ideally, people would go, "If I'm feeling like doing Premiere, big project, I'm doing a, you know, four-part t- television series, that's definitely a Premiere thing. But if I want to do something to show my recent vacation, maybe I'll just use Rush because I can do it in the half an hour I have free at home rather than the four hours I'd need to do it at work." And for the, for the use cases which we can do well, it really is much faster to get the same output, but the more professional tools obviously have a much richer toolkit and more flexibility in what they can do.

Lex Fridman 42:21 And then at the same time, with the flexibility and control, I, I like this idea of smart d- defaults of, uh, using AI to, to coach you, uh, to, like what Google has a I'm Feeling Lucky button.

Gavin Miller 42:34 Right.

Lex Fridman 42:34 Uh, where one button kind of gives you a pretty good set of settings and then you almost, that's almost an educational tool-

Gavin Miller 42:42 Absolutely. Yeah.

Lex Fridman 42:43 ... to show... Uh, because sometimes when you have this, all this control, you're not sure about the, um-... correlation between the different bars that control different elements-

Gavin Miller 42:54 Mm-hmm.

Lex Fridman 42:54 ... of the image and so on. And som- sometimes there's a degree of, uh, you don't know what the optimal is. (laughs)

Gavin Miller 43:02 And then some things are sort of on demand like help, right? Where-

Lex Fridman 43:05 Help, yeah.

Gavin Miller 43:06 ... I'm stuck, I need to know what to look for, I'm not quite sure what it's called.

Lex Fridman 43:10 Right.

Gavin Miller 43:10 And something that was proactively making helpful suggestions or-

Lex Fridman 43:14 Yeah.

Gavin Miller 43:15 ... you know, make a s- you know, you could imagine a make a suggestion button, um, where you'd use all of that knowledge of workflows and everything to maybe suggest something that, to go and learn about, or just, just to try or show the answer.

Lex Fridman 43:26 Yeah.

Gavin Miller 43:26 Um, and maybe it's not one intelligent default, but it's like a variety of defaults, and then you go, "Oh, I like that one."

Lex Fridman 43:33 Yeah.

Gavin Miller 43:34 Yeah.

Lex Fridman 43:34 Yeah.

Gavin Miller 43:34 Yeah.

Lex Fridman 43:35 Several options.

Gavin Miller 43:36 Yeah.

Lex Fridman 43:37 So back to poetry. (laughs)

Gavin Miller 43:39 Ah. Yes. (laughs)

Lex Fridman 43:40 We're gonna, we're gonna interleave. So first few lines of a recent poem of yours, uh, before I ask the next question.

Gavin Miller 43:48 Yep.

Lex Fridman 43:49 This is about, uh, the smartphone. "Today I left my phone at home and went down to the sea. The sand was soft, the ocean glass, but I was still just me." So this is a poem about you leaving your phone behind and, uh, feeling quite liberated because of it.

Gavin Miller 44:08 Mm-hmm.

Lex Fridman 44:09 So this is kind of a difficult topic, and, uh, let's see if we can talk about it-

Gavin Miller 44:13 (laughs)

Lex Fridman 44:14 ... and fi- figure it out. But so with the help of AI, more and more, we can create sort of versions of ourselves, versions of reality that are in some ways more beautiful than actual reality.

Gavin Miller 44:26 Mm-hmm.

Lex Fridman 44:27 Uh, you know, and, uh, some of the creative effort there is, is part of doing this, creating this illusion. So, of course, this is inevitable, but how do you think we should adjust as human beings to live in this digital world that's partly artificial? Like, that's better than, uh, the, the world that we lived in 100 years ago-

Gavin Miller 44:51 (sighs)

Lex Fridman 44:51 ... when you didn't have Instagram and, uh, Facebook versions of ourselves and the online versions?

Gavin Miller 44:56 Oh, this is sort of showing off better versions of ourselves.

Lex Fridman 44:59 Well, using the tooling of, of modifying the images or even with artificial intelligence ideas of deepfakes and creating s- adjusted or fake versions of ourselves and reality.

Gavin Miller 45:14 I, I think it's an interesting question. You are sort of historical bent on this, so...

Lex Fridman 45:18 (laughs) So-

Gavin Miller 45:19 I, I, I actually wonder if 18th century aristocrats who commissioned famous painters to paint portraits of them-

Lex Fridman 45:25 (laughs)

Gavin Miller 45:25 ... had portraits that were slightly nicer than they-

Lex Fridman 45:28 Touché.

Gavin Miller 45:28 ... actually looked in practice.

Lex Fridman 45:29 Well played, sir.

Gavin Miller 45:30 Uh, so-

Lex Fridman 45:31 (laughs)

Gavin Miller 45:31 ... uh (laughs) human desire to put your best foot forward has always been true. I, I think it's interesting. You, you sort of framed it in two ways. One is, if we can imagine alternate realities and visualize them, d- is that a good or bad thing? In the old days, you do it with storytelling and words and poetry, uh, which still resides sometimes on-

Lex Fridman 45:51 (laughs)

Gavin Miller 45:51 ... websites. But, uh, you know, we've become a very visual culture in particular. In the 19th century, we were very much a text-based culture. Um, people would read long tracts. Political speeches were very long. Nowadays, everything's very kind of quick and visual and snappy. I think it depends on how harmless your intent. It's, a lot of it's about intent. So if you, um, have a somewhat flattering photo that you pick out of the photos that you have in your inbox to say, "This is what I look like." It's, it's probably fine. If someone's gonna judge you by how you look, then they'll decide soon enough when they meet you whether the reality... you know?

Lex Fridman 46:39 (laughs) Yeah, right.

Gavin Miller 46:39 Um, I think where it can be harmful is if people hold themselves up to an impossible standard which they then feel bad about themselves for not meeting. I think that's definitely a, uh, can be a, an issue. And, uh, but I think the ability to imagine and visualize an alternate reality, which sometimes of w- sometimes of which you then go off and build later, um, can be a wonderful thing, too. People can imagine architectural styles which they then, you know, have a startup, make a fortune, and then build a house that looks like their favorite video game. Is that a terrible thing? I think, uh, I used to worry about exploration actually, that part of the joy of going to the moon when I was a tiny child, I remember it in grainy black and white, was to know what it would look like when you got there. And I think now we have such good graphics for knowing, for visualizing the experience before it happens, that I slightly worry that it may take the edge off actually wanting to go. You know what I mean?

Lex Fridman 47:42 (laughs) Yeah.

Gavin Miller 47:42 Because we've seen it on TV. We kind of, "Oh..." You know, by the time we finally get to Mars, we'll go, "Oh, yeah, so this is Mars. It, this is what it looks like." But then, you know, the outer exploration, I mean, I think, um, Pluto was a fantastic recent discovery where nobody had any idea what it looked like, and it was just breathtakingly varied and beautiful. So I think expanding the ability of the human toolkit to imagine and communicate, on balance, is a good thing. I think there are abuses. We definitely take them seriously and try to discourage them. Um, I think there's a parallel side where the public needs to know what's possible through events like this, right?

Lex Fridman 48:23 Mm-hmm.

Gavin Miller 48:24 So that, um, you don't believe everything you read in print anymore. And it may, over time, become true of images as well. Um, or you need multiple sets of evidence to really believe something, rather than a single media asset. So I think it's a constantly evolving thing. It's been true forever. There's a famous story about Anne of Cleves and Henry VIII, where luckily for Anne, they didn't get married, right? Um, so... Or they got married and broke up and it-

Lex Fridman 48:53 How... What's the story? What's the, uh, how did the-

Gavin Miller 48:54 Oh, so Holbein went and painted a picture, and then Henry VIII wasn't pleased and, you know-

Lex Fridman 48:59 Oh, yeah, the ...

Gavin Miller 49:00 History doesn't record whether Anne was pleased, but I think she was pleased not to, uh, be married more than a day or something.

Lex Fridman 49:05 Yeah. (laughs)

Gavin Miller 49:06 So, I mean, this has gone on for a long time, but, um, I think it's just a part of the magnification of human capability.

Lex Fridman 49:14 You've kind of built up a, an amazing research environment here, research culture, research lab. And you've written that the secret to a thriving research lab is interns. Can you un- unpack that a little bit?

Gavin Miller 49:26 Oh, absolutely. So, a couple of reasons. As you see looking at my personal history, there are certain ideas you bond with at a certain stage of your career, and you tend to keep revisiting them through time. Um, if you're lucky, you pick one that doesn't just get solved in the next five years, and then you're sort of out of luck. So I think a constant influx of new people brings new ideas with it. From the point of view of industrial research, because a big part of what we do is really taking those ideas to the point where they can ship as very robust features, you end up investing a lot in a particular idea. And if you're not careful, people can get too conservative in what they choose to do next, knowing that the product teams will want it. And interns let you explore the more fanciful or unproven ideas in a relatively lightweight way, um, ideally leading to new publications for the intern and for the researcher. And it gives you then a portfolio from which to draw which idea am I going to then try to take all the way through to being robust in the next year or two to ship. So it sort of becomes part of the funnel. It's also a great way for us to identify future full-time researchers. Many of our greatest researchers were former interns. It builds a bridge to university departments so we can get to know and build an enduring relationship with the professors, and we often do academic gift funds too as well, as an acknowledgement to the value the interns add and their own collaborations. So it's sort of a virtuous cycle. And then, the long-term legacy of a great research lab hopefully will be not only the people who stay, but the ones who move through and then go off and carry that same model to other companies. And so, we believe strongly in industrial research and how it can complement academia, and we hope that this model will continue to propagate and be invested in by other companies. Which makes it harder for us to recruit, of course, but, you know, uh, that's a sign of success, and a, a rising tide lifts all ships in that sense.

Lex Fridman 51:32 And where, where is the idea born with, with the interns? Is there brainstorming? Is there discussions about, you know, like what d- Adobe ...

Gavin Miller 51:42 Where do the ideas come from?

Lex Fridman 51:43 Well, yeah.

Gavin Miller 51:44 Yeah.

Lex Fridman 51:44 As, I, as I'm s- asking the question, I realize how dumb it is, but I'm hoping-

Gavin Miller 51:48 No, it's not a dumb question at all.

Lex Fridman 51:48 ... you have a better answer, better answer than a butte- than ... (laughs)

Gavin Miller 51:50 It's actually a question I ask at the beginning of every summer.

Lex Fridman 51:53 Okay. (laughs)

Gavin Miller 51:54 Um, so what will happen is we'll send out a call for interns. They'll, um, we'll have a number of resumes come in. People will contact the candidates, talk to them about their interests. They'll usually try to find some- somebody who has a reasonably good match to what they're already doing or just has a really interesting domain that they've been pursuing in their PhD and we think we'd love to do one of those projects too. And then the intern stays in touch with the, uh, mentor, as we call them, and then they come and in the fir- uh, at the end of two weeks, they have to decide. So, they'll often have a general sense by the time they arrive, and we'll have internal discussions about what are all the general ideas that we're wanting to pursue to see whether two people have the same idea and maybe they should talk and all that. But then once the intern actually arrives, sometimes the idea goes linearly and sometimes it takes a giant left turn and we go, "That sounded good, but when we thought about it, there's this other project," or, "It's already been done and we found this paper, uh, we were scooped-"

Lex Fridman 52:55 Yeah.

Gavin Miller 52:55 "... but we have this other great idea." So it's pretty, pretty flexible at the beginning. Um, one of the questions for research labs is, who's deciding what to do? And then, who's to blame if it goes wrong, who gets the credit if it goes right? And so in Adobe, we push the needle very much towards freedom of choice of projects by the researchers and the interns, but then we reward people based on impact, so if the projects ultimately end up impacting the products and having papers and so on. And so, um, the alternative model, just to be clear, is that you have one lab director who thinks he's a genius and tells everybody what to do, takes all the credit if it goes well, blames everybody else if it goes badly.

Lex Fridman 53:39 Yeah.

Gavin Miller 53:39 So we don't want that model. Um, and this, this helps new ideas percolate up. The art of running such a lab is that there are strategic priorities for the company, and there are areas where we do want to invest in pressing problems. And so, it's a little bit of a trickle down and filter up, meets in the middle. And so you, you don't tell people, "You have to do X," but you say, "X would be particularly appreciated this year."

Lex Fridman 54:05 Mm-hmm.

Gavin Miller 54:05 And then people reinterpret X through the filter of things they want to do and they're interested in. And it, miraculously, it usually comes together very well. Um, one thing that really helps is Adobe has a really broad portfolio of products. So, if we have a good idea, um, there's usually a product team that is intrigued or interested. Uh, so it means we don't have to qualify things too much ahead of time. Once in a while, the product teams sponsor a extra intern 'cause they have a particular problem that they really care about, in which case it's a little bit more, "We really need one of these." And then we sort of say, "Great, I get an extra intern." The in- we find an intern who thinks that's a great problem. But that's not the typical model. That's sort of the icing on the cake as far as the budget's concerned. And, um, and all of the above end up being important. Uh, it's really hard to predict at the beginning of the summer which of the ... We all have high hopes of all of the intern projects, but ultimately some of them pay off and some of them...... sort of a very nice paper but don't turn into a feature. Others turn out not to be as novel as we thought, but they'd be a great, uh, feature but not a paper. And then others, we make a little bit of progress and we realize how much we don't know. And maybe we revisit that problem several years in a row until it... finally, we have a breakthrough, and then it becomes more on track to impact a product.

Lex Fridman 55:26 Jumping back to, uh, big over- overall view-

Gavin Miller 55:30 Mm-hmm.

Lex Fridman 55:30 ... of Adobe Research, what are you looking forward to in 2019 and beyond? What is, uh, you mentioned there's a giant suite of products.

Gavin Miller 55:39 Yes.

Lex Fridman 55:39 A giant suite of ideas, new interns, new ye- uh, you re- uh, uh, a, a large team of researchers. Where do you think, um, uh, that what the... what do you think the future holds?

Gavin Miller 55:52 In terms of the technological breakthroughs?

Lex Fridman 55:54 Technological breakthroughs, especially ones that, um, will make it into product, will, will get to impact the world.

Gavin Miller 56:01 So I think the creative or the analytics assistants that we talked about where they're constantly trying to figure out what you're trying to do and how can they be helpful and make useful suggestions is a really hot topic. And it's very unpredictable as to when it'll be ready, but I'm really looking forward to seeing how much progress we make against that. I think, um, the... some of the core technologies, like generative adversarial networks, are immensely promising. And seeing how quickly those become practical for mainstream use cases at high resolution with really good quality is also exciting. And they also have this sort of strange way of even the things they do oddly are odd in an interesting way.

Lex Fridman 56:43 Mm-hmm.

Gavin Miller 56:43 So it could look like dreaming or something.

Lex Fridman 56:46 (laughs)

Gavin Miller 56:46 So that's fascinating. I think internally, we have a Sensei platform, which is a way in which we're pooling our neural net and other intelligence models into a central platform, which can then be leveraged by multiple product teams at once. So we're in the middle of transitioning from a, you know, once you have a good idea, you pick a product team to work with and they... you sort of hand design it for that use case, to a more sort of Henry Ford stand it up in a standard way which can be accessed in a standard way. Which should mean that the time between a good idea and impacting our products will be greatly shortened. And when one product has a good idea, many of o- many of the other products can just leverage it too. So it's sort of an economy of scale. So that's more about the how than the what, but that combination of this sort of renaissance in AI, there's a comparable one in graphics with real-time ray tracing and other really exciting emerging technologies. And when these all come together, you're sort of basically be dancing with light, right? Where you'll have real-time shadows, reflections and, um, as if it's a real world in front of you, but then with all these magical properties brought by AI where it sort of anticipates or modifies itself in ways that make sense based on how it understands the creative task you're trying to do.

Lex Fridman 58:06 That's a, that's a really exciting, uh, f- future for creative, for, for myself too-

Gavin Miller 58:11 Mm-hmm.

Lex Fridman 58:11 ... as a creator. So, so first of all, I work-

Gavin Miller 58:13 Yeah.

Lex Fridman 58:13 ... in autonomous vehicles. I'm a roboticist. I love robots. And I think, uh, you have a fascination with snakes, both natural and artificial snake robots.

Gavin Miller 58:21 Mm-hmm. Absolutely.

Lex Fridman 58:21 I share your fascination. I mean, their movement is beautiful, adaptable-

Gavin Miller 58:26 Mm-hmm.

Lex Fridman 58:26 The adaptability is, is fascinating. There are, I looked it up, 2,900 species of snakes in the world.

Gavin Miller 58:33 Wow.

Lex Fridman 58:34 375 venomous. Some are tiny, some are huge. Uh, saw that there's one that's 25 feet in some cases. So what's the, the most interesting thing that you connect with in terms of snakes, both natural and artificial? Why... what, what was the connection with-

Gavin Miller 58:52 What, why did-

Lex Fridman 58:52 ... robotics AI and this particular form of a robot?

Gavin Miller 58:56 Well, it actually came out of my work in the '80s on computer animation, where I started doing things like cloth simulation and other kind of soft body simulation. And you'd sort of drop it and it would bounce, and then it would just sort of stop moving. And I thought, "Well, what if you animate the spring lengths and simulate muscles?" And the simplest object I could do that for was an earthworm. So I actually did a paper in 1988, um, called The Motion Dynamics of Snakes and Worms. And I read the physiology literature on both how snakes and worms move, and then did some of the early computer animation examples of that and, uh...

Lex Fridman 59:32 So your interest in robotics-

Gavin Miller 59:33 Came out of-

Lex Fridman 59:34 ... started with graphics?

Gavin Miller 59:34 Came, came out of simulation and graphics.

Lex Fridman 59:38 That's cool.

Gavin Miller 59:38 When I moved from Alias to Apple, we actually did a movie called Her Majesty's Secret Serpent, which is about a secret agent snake that parachutes in and captures a film canister the- from a satellite, which tells you how old-fashioned we were thinking back then. Sort of classic 19 sort of 50s or '60s Bond movie kind of thing. Um, and at the same time I'd always made radio controlled ships when I was a child, and, uh, from scratch. And I thought, "Well, how hard can it be to build a real one?" And so then started a, what turned out to be like a 15-year obsession with, uh, trying to build better snake robots. And the first one that I built just sort of slithered sideways, but didn't actually go forward. Then I added wheels. And building things in real life makes you honest about the friction. The, the thing that appeals to me is I, I love creating the illusion of life, which is what drewed me to an- drove me to animation. And if you have a robot with enough degrees of coordinated freedom that move in a kind of biological way, then it starts to cross the uncanny valley and to see me like a creature rather than a thing. And I certainly got that with the early snakes by S3. I had it able to sidewind as well as go directly forward. My wife to be suggested that it would be the ring bearer at our wedding, so it actually went down the aisle carrying the rings, and uh (laughs) got in the local paper for that, which was really fun. And this was all done as a hobby. And then I, um... at the time the com- onboard compute was incredibly limited. It was sort of-

Lex Fridman 1:01:08 Yeah, so you should explain that.

Gavin Miller 1:01:09 Yeah.

Lex Fridman 1:01:09 The, these snakes though, the whole idea is that you would... you're trying to run it sta- um, autonomously, so-

Gavin Miller 1:01:15 Autonomously.

Lex Fridman 1:01:16 ... all the computers onboard.

Gavin Miller 1:01:16 Onboard power, onboard bra- yeah, and so...The very first one I actually built the controller from discrete logic, uh, 'cause I used to do LSI, you know, circuits and things when I was a teenager. And then, um, the second and third one, the 8-bit microprocessors were available with, like, a whole 256 bytes of RAM, um, which you could just about squeeze in. So they were radio controlled rather than autonomous, and really were more about the physic- physicality and coordinated motion. Um, I've occasionally taken a sidestep into, "If only I could make it cheaply enough, it would make a great toy," which has been a lesson in how clockwork is its own, uh, magical realm that you venture into and learn things about backlash and other things you don't take into account as a computer scientist, which is why what seemed like a good idea doesn't work. So it was quite humbling. And then more recently, I've been building S9, which is a much better engineered version of S3 where the motors wore out and it doesn't work anymore and you can't buy replacements, which is sad given that it was such a meaningful one. Uh, S5 was about twice as long and, um, looked much more biologically, uh, inspired. I, unlike the typical roboticist, I taper my snakes. Uh, there are good mechanical reasons to do that, but it also makes them look more biological. Although it means every segment's unique rather than, um, a re- repetition, which is why most engineers don't do it. Uh, it actually saves weight and leverage and everything. And that one is currently on display at the International Spy Museum in Washington, DC.

Lex Fridman 1:02:52 (laughs)

Gavin Miller 1:02:53 Not that it's done any spying. It was on YouTube and it got its own conspiracy theory where people thought that it wasn't real 'cause I work at Adobe. It must be fake graphics.

Lex Fridman 1:03:02 (laughs)

Gavin Miller 1:03:03 And people would write to me, "Tell me it's real." You know, they say, "The background doesn't move." And it's like, "It's, it's on a tripod, you know?" Um, so that one... But you can see the real thing, so it really is true. And then the latest one is the first one where I could put a Raspberry Pi, which leads to all sorts of terrible jokes about pythons and things. But, um...

Lex Fridman 1:03:22 (laughs) Yeah.

Gavin Miller 1:03:23 Yeah. But this one can have onboard compute. And then where my hobby work and my work work are converging is you can now add vision accelerator chips which can evaluate neural nets and do object recognition and everything. So both for the snakes and more recently for the spider that I've been working on, um, having, you know, desktop-level compute is now opening up a whole world of true autonomy with onboard compute, onboard batteries, and, um, still having that sort of biomimetic quality that people a- that appeals to children in particular. They're really drawn to them. And adults think they look creepy, but children actually think they look charming. Um, and, uh, uh, I gave a series of lectures at Girls Who Code to encourage people to take an interest in technology. And at the moment, I'd say they're still more expensive than the value that they add, which is why they're a great hobby for me, but they're not really a great product. Um, it makes me think about doing that very early thing I did at Alias with changing the muscle rest lengths. If I could do that with a real artificial muscle material, then the next snake ideally would use that rather than motors and gearboxes and everything. It would be lighter, much stronger, and more continuous and smooth. So, um, it's s- I like to say being in research is a license to be curious, and I have the same feeling with my hobby. It forced me to read biology and be curious about things that otherwise would have just been, you know, a National Geographic special. Suddenly I'm thinking, "How does that snake move? Can I copy it?" I look at the trails that sidewinding snakes leave in sand and see if my snake robots would do the same thing. Um-

Lex Fridman 1:05:09 So, uh, out of something inanimate, I like the way you put it, try to bring life into it and beauty.

Gavin Miller 1:05:13 Absolutely.

Lex Fridman 1:05:14 And-

Gavin Miller 1:05:14 And then ultimately give it a personality, which is where the intelligent agent research will, uh, converge with the vision and voice synthesis to give it a sense of having, not necessarily human-level intelligence. I think the Turing test is such a high bar, it's-

Lex Fridman 1:05:30 Mm-hmm.

Gavin Miller 1:05:30 ... a little bit self-defeating. But having one that you can have a meaningful conversation with, uh, especially if you have a reasonably good sense of what you can say. Um, so not trying to have it so, uh, a stranger could walk up and have one. But so as a, as a pet owner or robot pet owner, you could know what it thinks about and what it can reason about, and...

Lex Fridman 1:05:53 Or sometimes just a meaningful interaction. If, if you have a, the kind of interaction you have with a dog, sometimes you might have a conversation, but it's usually one way.

Gavin Miller 1:06:02 Absolutely. Right.

Lex Fridman 1:06:02 And, uh, nevertheless it feels like a meaningful, uh, connection.

Gavin Miller 1:06:06 And one of the things that I'm trying to do in the sample audio that we'll play you is beginning to get towards the point where the, the reasoning system can explain why it knows something or why it thinks something. And that, again, creates the sense that it really does know what it's talking about. But also, for debugging. (laughs)

Lex Fridman 1:06:26 (laughs)

Gavin Miller 1:06:26 As you get, uh, more and more elaborate behavior, it's like, "Why did you decide to do that? Why, what, you know, how do you know that?" I think it's, the robot's really are my muse for helping me think about the future of AI and what to invent next.

Lex Fridman 1:06:42 So even a- at Adobe, that's mostly operating in digital world.

Gavin Miller 1:06:47 Correct.

Lex Fridman 1:06:47 Do you ever th- Do you see a future where Adobe even expands into the more physical world perhaps? So bringing life not into animations, but bringing life into physical o- objects with whether it's... Well, do you, you, uh-

Gavin Miller 1:07:04 I have to say at the moment, it's a twinkle in my eye. I think the more likely thing is that we will bring virtual objects into the physical world through augmented reality.

Lex Fridman 1:07:14 Augmented reality.

Gavin Miller 1:07:15 And many of the ideas that might take five years to build a robot to do, you can do in a few weeks with digital assets. So I think when really intelligent robots finally become commonplace, they won't be that surprising because we'll have been living with those personalities for, in the virtual sphere for a long time. And then they'll just say, "Oh, it's, you know, Siri with legs," or Alexa, "Alexa on hoofs," or something. Um, so I can see that world coming. Um, and for now, it's still an adventure and we don't know quite what the experience will be like, and it's really exciting to sort of see all of these different strands of my career converge.

Lex Fridman 1:07:57 Yeah, in interesting ways.

Gavin Miller 1:07:59 Yeah.

Lex Fridman 1:07:59 And it's, it is definitely a fun adventure.

Gavin Miller 1:08:01 Yeah.

Lex Fridman 1:08:01 So, uh, l- let me end with, uh, my favorite poem, uh, the last few lines of my favorite poem of yours-

Gavin Miller 1:08:08 Go ahead.

Lex Fridman 1:08:09 ... that ponders mortality, and in some sense immortality b- you know, as our ideas live through the ideas of others-

Gavin Miller 1:08:15 Yes.

Lex Fridman 1:08:16 ... through the work of others. It ends with, "Do not weep or mourn. It was enough the little atomies permitted just a single dance. Scatter them as deep as your eyes can see. I'm content they'll have another chance. Sweeping more centered parts along to join a jostling, lifting throng as others danced in me." Beautiful poem, beautiful way to end it. And Gavin, thank you so much for talking today. And thank you for inspiring and empowering millions of people s- like myself for creating amazing stuff.

Gavin Miller 1:08:49 Well, thank you. It was a great conversation.

Lex Fridman 1:08:57 Yeah.