Jun 28, 2017 · 52m · y-combinator
Ex Machina's Scientific Advisor - Murray Shanahan · Y Combinator
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
In an extended Y Combinator interview, AI researcher and DeepMind senior scientist Murray Shanahan reflects on his career evolution from 1980s symbolic AI to modern deep reinforcement learning, his scientific advising on Alex Garland's film Ex Machina, and the philosophical challenges surrounding artificial general intelligence and human perception.
How this conversation actually went
Every chapter scored 0–10 on four independent dynamics. Hover any point for the reasoning behind the score. How this is scored →
The partners as informed peer 1.9 Guest teaching 3.8 Guest disagreement 0.3 The partners pushing back 0.1
speaking balance: gold is the partners, purple is the guest (3 minute bins)
Sharpest disagreement ▶ 32:17 Playful pushback on code bug accusation
Shanahan humorously pushes back against the host's suggestion that his Easter egg Python code contained a bug, insisting it strictly met specification despite suboptimal efficiency.
Hardest push from the partners ▶ 26:02 Probing the simulation and fourth-wall interpretationCraig challenges the interpretation of the cut visual effects sequence, asking whether Garland avoided it to prevent a meta simulation trope.
Biggest teaching moment ▶ 39:00 Clarifying the purpose of Asimov's lawsShanahan educates the audience and host on how Asimov's laws are fictional dramatic devices exploring moral contradictions rather than viable computational frameworks for AI.
The partners hold their own ▶ 8:29 Kasparov's insight on cognitive adaptation to chess enginesCraig demonstrates strong synthesis by articulating Kasparov's observation that modern grandmasters reshape their own cognitive patterns around engine analysis.
the scores for every segment, with the reasoning behind each
| Chapter | Topic | The partners as informed peer | Guest teaching | Guest disagreement | The partners pushing back | Why |
|---|---|---|---|---|---|---|
| Early AI Research and PhD Thesis | 1 | 4 | 0 | 0 | Craig sets up open-ended prompts regarding Shanahan's doctoral work in the 1980s. Shanahan educates the host on early Prolog logic programming and introduces the frame problem. | |
| Career Trajectory and the Frame Problem in Modern AI | 1 | 4 | 0 | 0 | Shanahan details his intellectual journey away from symbolic AI toward computational neuroscience, then into modern deep learning. The host listens as Shanahan illustrates how the frame problem persists in Atari games. | |
| Interruption and Reflection on Garry Kasparov Talk | 3 | 3 | 0 | 0 | Both host and guest reflect on Garry Kasparov's talk with Demis Hassabis. Craig contributes observations about chess engines reshaping players' minds, while Shanahan elaborates on AlphaGo's Move 37. | |
| Scientific Advising on 'Ex Machina' and Philosophical Foundations | 2 | 4 | 0 | 0 | Shanahan discusses advising Alex Garland on Ex Machina and explains the philosophical roots of embodied cognition and Wittgenstein's view on attributing consciousness based on interaction. | |
| Ex Machina Script Changes, Ending Ambiguity, and Python Easter Egg | 2 | 3 | 1 | 0 | Shanahan shares behind-the-scenes script details including the removed VFX perspective and his Python Easter egg containing his book ISBN. He humorously defends his coding efficiency when Craig teases him about a bug. | |
| Comparing 1950s AI Predictions, Modern AGI, and Asimov's Laws | 3 | 5 | 1 | 1 | Craig asks audience questions about 1950s AI visions and Asimov's laws. Shanahan clarifies that Asimov's laws are narrative plot devices rather than practical engineering principles for modern machine learning. | |
| Deep Reinforcement Learning at DeepMind and Hybrid AI | 2 | 4 | 0 | 0 | Shanahan explains the breakthroughs and sample inefficiency of deep reinforcement learning at DeepMind, advocating for hybrid symbolic-neural architectures. Craig links this to points raised during Kasparov's talk. | |
| Consulting on Sci-Fi Theatre and Interactive Art Projects | 1 | 3 | 0 | 0 | Shanahan describes consulting on theatrical and artistic projects involving neuroscience and the perception of human motion in point-light robotics, while Craig provides supportive wrap-up commentary. |
Statements from this episode (12)
Shanahan: AI research evolved from an isolated niche into intense public attention
“Certainly when I was a PhD student and when I was a young postdoc, it was a fairly niche area. So you could just kind of like beaver away in your little, little kind of corner. Doing things that you thought were intellectually interesting and being reasonably …”
The AI frame problem is determining relevance without being computationally overwhelmed
“The frame problem in its largest guise is all about how how a thinking mechanism or thinking creature or a thinking machine, if you like can work out what's relevant and what's not relevant to its To its ongoing cognitive processes and how it isn't overwhelmed…”
Shanahan abandoned classical AI in 2000 over lack of AGI progress
“By kind of the turn of the millennium, I'd more or less abandoned classical AI because I didn't think it was moving, moving towards what we now call AGI, artificial general intelligence, the big vision of human level AI.”
Shanahan: AlphaGo's Move 37 revolutionized top-level Go strategy
“So we've already seen that with our, with AlphaGo in the match with Lee Sedol. So as you probably know, there was a famous move in the second match against Lee Sedol, move 37, where all the commentators, all these sort of nine Dan masters were saying, saying, …”
Shanahan: Physical embodiment is the origin of human intelligence
“No doubt when it comes to human intelligence and human consciousness Our physical embodiment is a huge part of that. It's in, it's where our intelligence originates from because what we, what our brains are really here to do is to help us to navigate and manip…”
Shanahan: Ex Machina's script was 95% complete before his advisory involvement
“I had no input on that side at all. So the script was already, and the plot was already, the whole script was already, you know, 95% you know done, you know, when I first saw it.”
Ex Machina's on-screen code prints the ISBN of Shanahan's book
“So when you look at this code on the screen, it's just gobbledygook, but something to do with prime numbers. If you run it prints out ISBN equals and the ISBN of my book in bottomers in the inner life.”
Shanahan: Supercomputers will reach human brain scale computing within two years
“Well, depending on how you calculate it, we're pretty close to human brain scale computing already in the world's fastest supercomputers. And we will get there within the next couple of years.”
Shanahan: Raw speech recognition has essentially been solved
“So speech recognition has more or less been cracked. The, they're just the process of turning the way raw waveform into text into, so that that's been cracked.”
DeepMind's DQN is arguably one of the first artificial general intelligences
“To my mind, DQN is in a sense, one of the very first general intelligences because it learns completely from scratch. You can throw A whole variety of problems at it, and it, you know, it doesn't always do that well, but in many cases it does pretty well.”
Shanahan: Symbolic AI ideas can be rehabilitated into deep reinforcement learning
“It made me realize that there were various ideas from symbolic AI that could be rehabilitated and put into deep reinforcement learning systems in a more modern guise. And so that's the kind of thing that I'm most interested in.”
Shanahan: Humans mistakenly project consciousness onto machines without actual agency
“We see someone there when there isn't, and of course that, for me, that was very interesting because it made me think about when we do that with machines, where we often, we do Maybe, you know, we think that there's someone at home when there isn't.”