Gavin Miller: Adobe Research

Episode Summary

Executive Summary: Gavin Miller, head of Adobe Research, discusses how AI can reduce tedious creative work and expand human creativity across image, video, audio, and 3D tools. He connects research, poetry, and robotics to argue for human-machine collaboration, smart defaults, robust productization, privacy-aware data use, and future interfaces that preserve control while making creative workflows faster and more intuitive.

Main Topics: AI as a creative accelerator (Priority: 5/5): Adobe’s goal is to automate repetitive, low-level tasks so creators can spend more time in ideation, composition, and storytelling rather than pixel-level labor. From research to product robustness (Priority: 5/5): Miller contrasts academic novelty with industrial shipping, emphasizing that Adobe must build features that work reliably in real workflows, with fallback tools and human override. Intelligent tools for selection, fill, and editing (Priority: 5/5): The conversation dives into AI-assisted selection, background removal, content-aware fill, sky replacement, and object manipulation as examples of tasks AI can dramatically accelerate. Learning from user workflows and tutorials (Priority: 4/5): Adobe aims to infer intent from usage context, tutorials, and workflow patterns to provide proactive guidance, onboarding, and assistant-like suggestions. Privacy, trust, and data collection (Priority: 5/5): Miller stresses that Adobe must earn explicit permission, demonstrate user benefit, and avoid compromising trust when collecting data to improve AI systems. AR/VR, 3D design, and future interfaces (Priority: 4/5): He argues that immersive and spatial tools may reshape content creation, especially for 3D layout, augmented reality assets, and more natural interaction models. Poetry, robotics, and biomimicry as creative research (Priority: 3/5): Miller explains how poetry and snake-robot hobby projects inform his imagination, helping him think about personality, motion, autonomy, and the future of intelligent agents.

Key Arguments: AI should relieve creatives of repetitive production work so they can focus on higher-value conceptual decisions. Effective creative AI must be robust enough for professional use, but partial automation plus human correction is often the fastest path to value. Smart defaults are a powerful use of AI because many users need good starting points before fine-tuning settings. Selection, masking, and content-aware editing are ideal AI applications because they consume huge amounts of time and benefit from semantic understanding. Creative tools should learn from real workflows, tutorials, and prior actions to become better teachers and assistants. Adobe must treat privacy as foundational; users should see clear value before sharing data, and some workflows require strict confidentiality. Generative and 3D-aware methods will increasingly enable more natural image synthesis, editing, and immersive design experiences. The future of creativity is not replacing artists, but shifting them toward art direction, concept development, and higher-level authorship. Robotics, poetry, and visual simulation all inform Miller’s view that technology can create the illusion of life and personality. Interns are essential to research culture because they bring fresh ideas, keep labs exploratory, and build a pipeline of future researchers.

Data Points: Years at Adobe: 19-20 years - Miller notes he has been at Adobe for roughly two decades while discussing industrial research and productization. Selection workflow speedup: hours to a few seconds - He describes AI-assisted masking/selection of moving subjects as reducing work from hours to seconds. Adobe product span: Photoshop, Illustrator, Premiere, After Effects, InDesign, Audition - Listed as examples of Adobe’s creative software ecosystem at the start of the conversation. Snake species: 2,900 species - Miller cites this number while discussing his fascination with snakes and snake robots. Venomous snakes: 375 species - Used to illustrate the diversity of natural snakes. Early robot memory: 256 bytes of RAM - He mentions using 8-bit microprocessors with extremely limited memory in early snake robot prototypes. First snake robot year: 1988 - He references a paper on the motion dynamics of snakes and worms from 1988. Project timing: 1993-1994 - He says his smart-home and photo-album experiments were built around this period. Conference: Adobe Max - Used as the venue where Adobe demonstrated rapid polygon-based selection masks for moving subjects.

Pivotal Quotes: "We really want to span the entire range from really, really good low-level tools ... to automate the tedious tasks, and give more and more time to operate in the idea space instead of pixel space." — Gavin Miller: On Adobe’s philosophy for combining manual control with AI automation in creative software. "If you add a feature to a GUI, you have to have yet more visual complexity confronting the new user." — Gavin Miller: On why assistants and smart capabilities may be preferable to simply adding more interface controls. "Being in real Research is a license to be curious." — Gavin Miller: On the role of exploratory research, interns, and personal curiosity in driving long-term innovation.

Implications: Creative software is moving toward AI-powered assistance, semantic editing, and multimodal workflows. For users, this means faster production with less manual toil; for Adobe and the industry, it means balancing capability, control, trust, and privacy as tools become more predictive and assistant-like.