Sep 14, 2023 · 51m · big-technology
Google's DeepMind Wants To Make Human-Level Artificial Intelligence, Says Its Chief Business Officer
0:00 / 0:00
▶ Watch on YouTube →
gold bands on the timeline = statements, start to end. Hover to read, click to jump. CC turns on captions
Google DeepMind Chief Business Officer Colin Murdock discusses the quest for Artificial General Intelligence, technical milestones in multimodality and planning, and Google's strategy for translating foundational scientific breakthroughs into commercial products.
How this conversation actually went
Every chapter scored 0–10 on four independent dynamics. Hover any point for the reasoning behind the score. Alex holds 25.7% of the talking time here. How this is scored →
Alex as informed peer 5.3 Guest teaching 3.8 Guest disagreement 1.8 Alex pushing back 4.8
speaking balance: gold is Alex, purple is the guest (3 minute bins)
Sharpest disagreement ▶ 31:44 Guest rejects Manhattan Project comparison in favor of Apollo
Colin deflects Alex's premise that AI leaders are drawing self-aggrandizing nuclear bomb comparisons, reframing DeepMind's institutional narrative around the collaborative spirit of the Apollo space program.
Hardest push from Alex ▶ 31:57 Host rejects guest's deflection on Apollo framingAlex refuses to accept Colin's Apollo reframe, directly pushing back that nuclear comparisons are everywhere in AI leadership rhetoric and nobody talks about space projects.
Biggest teaching moment ▶ 41:34 Guest outlines AlphaFold's concrete scientific impactColin breaks down how AlphaFold mapped 200 million proteins to save an estimated billion years of PhD research, providing concrete scientific use cases in antibiotic resistance and plastic-eating enzymes.
Alex holds their own ▶ 18:15 Host leverages Yann LeCun interviews to define core intelligenceAlex brings deep domain credibility to the table, recalling multi-year interviews with Yann LeCun to anchor the discussion around prediction and planning as the fundamental benchmark of AGI.
the scores for every segment, with the reasoning behind each
| Chapter | Topic | Alex as informed peer | Guest teaching | Guest disagreement | Alex pushing back | Why |
|---|---|---|---|---|---|---|
| The Mission for AGI and Transforming Scientific Discovery | 5 | 4 | 2 | 5 | Alex pushes Colin to clarify why AGI is necessary beyond solving discrete problems, directly challenging Colin when he brings up AlphaFold by noting that AlphaFold is merely a narrow AI system. Colin elaborates on how general intelligence enables exploring vast, unknown problem spaces across sciences. | |
| Human-Machine Collaboration, Responsibility, and AI Governance | 6 | 3 | 2 | 7 | Alex challenges Colin on the philosophical implications of AGI, questioning whether corporations should steward human-like intelligence and pointing out the contradiction in Colin describing AI both as a child-like intelligence and merely a tool. Colin clarifies his view of AI as collaborative tooling developed under strict governance. | |
| The Technical Path to AGI: Generalization, Planning, and Memory | 7 | 4 | 1 | 3 | Alex displays solid domain literacy by referencing years of discussions with Yann LeCun on prediction and planning, as well as Shane Legg's definition of intelligence. Colin provides technical illustrations of generalization via MuZero video compression and simulated robotic learning. | |
| Gemini, Multimodality, and Training Methodologies | 6 | 3 | 3 | 6 | Alex brings up Constitutional AI and pushes Colin on why AI industry executives continually invoke Oppenheimer and Manhattan Project analogies. When Colin claims DeepMind prefers the Apollo project comparison, Alex immediately pushes back, stating nuclear comparisons dominate the discourse. | |
| The DeepMind-Brain Merger and Productization Pipeline | 4 | 4 | 1 | 3 | Colin walks through DeepMind's commercialization pipeline and internal matchmaking process between core research models and Alphabet business units. Alex inquires about the organizational balance between search and broader research initiatives. | |
| AlphaFold's Scientific Impact and Commercializing Drug Discovery | 4 | 5 | 2 | 5 | Colin details how AlphaFold mapped 200 million protein structures, saving vast amounts of research time across pharmaceuticals and enzymes. Alex asks whether spinning out ventures like Isomorphic Labs turns Google into an incubator, prompting Colin to demure on broad corporate strategy. |
Statements from this episode (10)
DeepMind CBO: AGI will traverse branching scientific problems at unimaginable scale
“Once we solve one problem that typically opens up and shows this kind of branching set of new problems, the AGI system will be able to kind of trans translate across that problem space at a speed and scale that we can't even imagine today.”
DeepMind: Human Go players improved performance using AlphaGo
“What we discovered, however, was that once we then made AlphaGo more generally available, the humans actually used AlphaGo to improve their performance.”
Murdock: DeepMind's MuZero algorithm dramatically reduces YouTube streaming bandwidth
“We were able to take an algorithm developed for games. That was a master in chess and use that to dramatically reduce the bandwidth requirement to stream YouTube videos.”
Murdock: DeepMind's Flamingo generates search metadata across YouTube Shorts videos
“We discovered Flamingo was also able to look at that video for you, essentially watch all the videos for you and add data, metadata to those videos, such that when you're searching for those videos, it's much easier, much, much easier to find the videos that y…”
Murdock: AI research has not yet reached the limits of model scaling
“We haven't necessarily reached the limit of making these models bigger either. We're not, no one's quite sure where that limit is.”
Murdock: Google's Gemini accepts and outputs both text and images natively
“One of the really important areas it's touching on is what we call multi-modality. It's a bit like the human senses you just described. We can kind of use all our human senses together and combined to achieve the goal where we're setting out to. So it will bri…”
DeepMind plans to release Gemini models in multiple sizes and scales
“We're hoping to develop models of different sizes and scales. So there'll be kind of different sizes of these Gemini models, which can then be applied to different use cases, depending on what's important.”
Murdock: DeepMind compares its AI mission to the Apollo project
“I think the comparison that I often hear is actually of the Apollo project, the space project... Well, maybe, maybe I hear it more because that's actually how Google DeepMind, we often think about it and talk about it.”
Murdock: DeepMind mapped all 200M known proteins using AlphaFold
“We've used AlphaFold now to map all two hundred million proteins known to science, or two hundred million proteins known to science, and we've made that available to everyone.”
Murdock: Merging LLMs and reinforcement learning will drive major AI breakthroughs
“I don't know exactly what the breakthrough will be, but I'm really excited about the union of these LLMs plus reinforcement learning. And I'm excited about that because I think there's a lot more to come from reinforcement learning. And I know at Google D mine…”