RoboBoom
Alex Conconi· March 9, 2023· 1:01:59

EP2 Geordie Rose: AI, LLMs, Humanoid Robots, Consciousness - The Full Convo with Sanctuary AI CEO

Geordie Rose, CEO of Sanctuary AI and founder of D-Wave and Kindred AI, discusses language models, minds and consciousness, quantum computing, and building general-purpose humanoid robots with Alex Conconi.

In this talk

  1. 3:54

    What language models can and cannot tell us

    Rose describes ChatGPT as a statistical system trained to predict tokens, whose simple interface helped people notice how capable AI had become. He argues that fluent language can resemble thought without proving that a system understands the world or has a mind. He frames a mind more broadly as a system that takes in perceptions and produces actions.

  2. 12:25

    Goals, work, and consciousness

    Rose defines intelligence as the computational ability to pursue goals, so a system’s performance depends on which goals it is given. Sanctuary’s approach is to measure progress across portfolios of work tasks, which Rose sees as overlapping with human goals. He says current language models are not conscious, while the nature of first-person experience remains unresolved and may challenge scientific assumptions.

  3. 35:21

    Humanoid robots and the work they can do

    Sanctuary measures robot progress by how well systems complete work goals, improving performance through incremental changes. Rose says human-like hands matter because they carry out tasks such as sewing, typing, and sorting laundry. He distinguishes humanoid control from driving: the robot’s hands alone have 34 degrees of freedom, creating a much larger control problem.

  4. 50:33

    Teleoperation as a path to autonomy

    Human operators control Sanctuary robots through a defined interface that transfers sensory input and movement commands. Rose says this lets the company deliver useful work before autonomous control is ready and collect examples for training software. The commercial plan is to deploy robots for paid labor, then lower costs and increase autonomy over time.

  5. 57:38

    Labor shortages and deployment

    Rose says Sanctuary looks for tasks workers dislike and targets those first, arguing that automation can ease bottlenecks as well as staffing gaps. He cites 17 million unfilled jobs across G7 countries and estimates that robots serving those roles could generate $150,000 in annual recurring revenue apiece. He presents demographic change and advances in AI, robotics, and manufacturing as reasons automated labor may become increasingly necessary.

Briefs summarise talks and interviews; timestamps open the video at that moment. Claims in a talk are the speaker's own.

More talks