Rodney Brooks: Robotics | Lex Fridman Podcast #217

Rodney Brooks' Background

• Led MIT's Computer Science and Artificial Intelligence Laboratory (CSAIL).
• Co-founded iRobot, which deployed over 30 million home-cleaning robots (Roomba).
• Co-founded Rethink Robotics, creating collaborative arms Baxter and Sawyer with human-safe force control.
• Co-founded Robust.ai with the mission to teach robots common sense.
• Currently writing a 480-page book titled Not Even Wrong.

Robotics History and Philosophy

• Built his first "computer" at age 8–9 using an ice cube tray, nails, and voltage to learn addition via copper bridges.
• Believes machines can "think" because humans are biological machines that think.
• Argues that current definitions of computation (Turing machines) are metaphors based on human limitations (limited memory), not universal laws of physics.
• Proposes a 2x2 matrix of four disciplines emerging 1945–1965: Neuroscience, Artificial Life, Artificial Intelligence, and Abiogenesis.
• States that "computation" has become a dominant metaphor that may not accurately describe physical processes like quantum mechanics or dark matter.
• Notes that 600 million years of evolution prioritized perception and mobility, making them biologically hard and engineeringly difficult, while high-level reasoning (like chess) is evolutionarily recent and easier to solve artificially.

Perception and Intelligence

• Supports Moravec's Paradox: Reasoning is easy, but perception and mobility are hard.
• Argues that deep learning solves the "labeling" problem (classification) but not the "grounding" problem (understanding objects in context).
• Cites a 1946 report where a human with an abacus beat an early electronic calculator, suggesting human calculation speed was overestimated relative to machine potential.
• Believes AlphaZero and AlphaGo are impressive engineering feats but remain within "closed" information games, unlike the open-ended world humans inhabit.
• Claims that human intelligence relies heavily on interaction with others and shared tools, which single neural networks cannot replicate.
• Suggests that human perception involves constructing reality (e.g., color constancy) rather than simply receiving photons, a nuance AI currently lacks.

Autonomous Driving and Infrastructure

• Acknowledges being "surprised" by Tesla's vision-based Autopilot and its ability to catch up to Mobileye without radar.
• Predicts that widespread fully autonomous driving will require infrastructure changes (e.g., dedicated lanes, modified road markings) similar to fully autonomous trains.
• Notes that fully autonomous trains exist in Japan with platform barriers, but US adoption is delayed due to infrastructure and regulatory hurdles.
• Criticizes the lack of infrastructure adaptation for autonomous vehicles in the US, which hinders adoption compared to closed environments.
• Highlights safety issues where autonomous vehicles halt unnecessarily (e.g., at crosswalks) due to over-sensitivity, disrupting human traffic flow.
• Argues that the death rate from human driving (~35,000/year in the US) makes the public irrational in rejecting autonomous vehicles that might kill even fewer people.
• States that Waymo has operated driverless cars in Chandler, Arizona, though rescue vehicles are still on standby for edge cases.

Company Experiences (iRobot, Rethink Robotics)

• iRobot: Successfully hit a $200 price point for the Roomba by sourcing low-cost chips from Taiwan, whereas competitors like Electrolux sold for $2,000.
• Deployed robots to Fukushima Daiichi in 2011 due to prior experience with tele-operated robots in Iraq and Afghanistan.
• Rethink Robotics: Initially aimed to create $3,000 collaborative arms but failed to meet the price target after engineers insisted on metal gearboxes over plastic ones, raising costs to $25,000+ and losing the target market.
• Lost $150 million in capital at Rethink Robotics; company failed due to acquisition hurdles (CFIUS blocking Chinese investment) and aggressive legal tactics by a competing buyer.
• Believes the "next trillion-dollar robotics company" will likely emerge from solving the demographic inversion (caring for an aging population), though current technology is not yet capable.

AI Predictions and Future

• Maintains a personal blog of predictions made every January 1st, planned to be reviewed until 2050.
• Predicted dedicated self-driving lanes on Highway 101 by a specific year, which did not materialize due to infrastructure and regulatory delays.
• Does not believe fully autonomous taxis will be viable in San Francisco (open city streets) for 10+ years, but expects deployment in gated communities and campuses first.
• Expresses skepticism that the Turing Test is a valid measure of intelligence, viewing it as a game of tricking humans rather than testing genuine understanding.
• Argues that current conversational AI lacks continuity, memory, and intention, preventing humans from forming genuine "friendship" bonds.
• Believes romantic human-AI love is possible but requires a convergence where humans also modify themselves with technology (e.g., genetic modification, implants).

MIT and the AI Community

• Regrets not asking more questions of deceased AI pioneers (Marvin Minsky, Seymour Papert, Rick Leitner) during their lifetimes.
• Notes that Minsky co-authored the 1968 Perceptrons book which destroyed neural network research, despite his own PhD being on neural networks.
• Criticizes tech giants (Google, Facebook) for "locking away" top talent in corporate labs, preventing open academic discourse and slowing field-wide progress.
• Remembers the early 1980s at MIT AI Lab, including a humorous anecdote where Donald Knuth watched while he and a student repaired a mainframe, causing them to accidentally install a chip backward.

Advice and Personal Outlook

• Advises young researchers to make "unsafe decisions" and embrace failure to achieve real impact, rather than chasing safe, incremental publications.
• Believes mortality is not scary, but the fear of losing cognitive faculties ("dribbling") is a significant concern.
• Expects his legacy to be his unfinished book, hoping it changes one person's perspective enough to spark a future breakthrough.
• Views the meaning of life as the emergence of order from disorder in a random universe, finding fun and wonder in this absurdity despite its likely eventual oblivion.