If It Can't Solve Quantum Gravity, It Isn't AGI
A physicist draws a high line for AGI: it counts when it solves quantum gravity. The conversation also clarifies the real relationship between quantum materials and AI — the materials a search turns up still have to clear synthesizability, and what survives may be zero.
The argument · tap a timestamp to hear it
Too much learning paralyzes; naivety is what lets you ask
The hardest part of going from student to researcher is not understanding the knowledge but posing the original questions that push its boundary forward. Gopakumar describes this transition as ‘naivety plus experience’: learning too much can actually paralyze you, and you need to keep a bit of naivety to dare ask the questions that look like they have no handhold. An advisor's job is to teach you to break a big problem into steps you can make progress on today without losing sight of the ultimate goal. His own advisor, David Gross, was very good at this — by his account, Gross's first student produced the work that won the two of them the Nobel together, and Witten also won a Fields Medal. He now runs the same method with his own students: he assigns the first one or two problems, ones he has no answer to either, works alongside them into their third or fourth year, and only then has them start asking their own questions.
— Rajesh GopakumarSpacetime is not a backdrop but an emergent approximation
Use ‘water is made of discrete atoms’ to understand string theory: the continuous spacetime we can touch is very likely just a low-resolution approximation, at macroscopic scale, of some deeper structure. Gopakumar stresses this is not pure speculation — quantum mechanics itself does not permit treating spacetime as an infinitely divisible smooth background. Add black holes (Hawking and others found they have many anomalous properties) and the breakdown of the Big Bang at the singularity, and physicists have to answer where spacetime comes from. String theory happens to supply a natural framework: spacetime emerges from something more fundamental, and gravity along with the known interactions fits inside it. This explains why he treats ‘solving quantum gravity’ as such a high bar for intelligence: it demands not just the ability to work problems but the construction of a new conceptual framework.
— Rajesh GopakumarDesign materials backwards: fix the properties, then find the structure
First, a long-cycle yardstick: quantum mechanics was originally meant only to explain the atom, and the devices it now underpins account, by his estimate, for roughly 35%-40% of US GDP — while the commercial ripples only became visible forty or fifty years after the theory appeared. He calls the current stage the second quantum revolution: quantum materials like graphene, found ‘by accident’, are already commercializing, and the next step is inverse materials design with AI — define what properties the material must have first, then search for what structure delivers them and how to synthesize it, bypassing traditional trial-and-error chemistry. He also revealed a conversation with Demis Hassabis: DeepMind's next holy grail is finding a room-temperature superconductor, and Hassabis told him at the outset not to talk about quantum gravity.
— Rajesh GopakumarBeing stuck is normal — keep three or four problems running
Getting stuck is not the exception, it is a researcher's daily life; the transition from student to researcher especially brings a sense of being crushed. Gopakumar's first strategy is a portfolio hedge: push three or four problems at once, so when one dies you switch to another, and probabilistically at least one will move half a step. His second is to go teach — the moment a light appears in a student's eyes, that satisfaction can pull you out of a trough, and he has repeatedly had the experience of ‘coming back from a lecture and the stuck problem suddenly opens up’. His third is to deliberately learn something new he has always wanted to study but never had time for, so he returns to the old problem from a new frame of reference.
— Rajesh GopakumarTalent multiplies rather than adds; headcount is never what's missing
Why does top-tier science concentrate in a handful of places rather than distributing evenly? Gopakumar gives the mechanism: conversations among people in the same location, using each other as sounding boards, keep the brain in a constantly stimulated state even without formal collaboration — and once collaboration does happen, the different skills two people bring multiply rather than add. He reaches for a physics analogy: photons in an ordinary light bulb are disordered, photons in a laser are coherent with each other, and a group of people who achieve coherence will likewise produce a lazing effect. He says many Indian institutions perform below what their talent deserves; what is missing is not headcount but this resonant atmosphere. Hence ICTS's founding principle is simple — hire only first-rate people, give them resources and freedom, and don't let them lower their ambitions.
— Rajesh GopakumarAI can accelerate research but cannot make conceptual leaps
AI has just taken two Nobel Prizes back to back, but Gopakumar splits ‘AI for science’ and ‘science for AI’ into two different things. The former he accepts as an overdrive gear: AI is accelerating one narrow class of task within research. But the ‘theory’ AI produces today is far from his conceptual-level standard — the best physics runs on conceptual leaps, not pattern recognition. Running the other direction, he is more bullish on science for AI: among the people at Anthropic who found the scaling laws there were physicists by training, and Hopfield was himself a statistical physicist, using simple physical models to simulate the brain. Along that road, physicists could work with computer scientists and neuroscientists to carry AI from language models to physics-inspired world models. This division is worth remembering.
— Rajesh GopakumarPast the synthesizability gate, AI's new materials come to zero
The most clear-eyed cold water of the session came in the audience Q&A: Google claims to have discovered a large number of new materials with AI, but once you filter for feasibility and synthesizability, Gopakumar's line is ‘I think left with zero’ — zero remain. His stance is not a rejection of AI materials science (someone in the same room cited Microsoft's DFT model with only 385,000 parameters, bypassing traditional hybrid methods); it is a reminder that the whole chain is still long: after you predict a crystal structure you still have to grow it and scale it, and every step can filter the surprise away. He says India should also place bets along this chain. The remark is a useful reverse calibration on the ‘AI discovers new materials’ narrative.
— Rajesh GopakumarIn their own words · checked verbatim
I think I'll believe that there is AGI when it solves quantum gravity.
Rajesh Gopakumar0:02
he was asking me what do you think are the big unsolved physics question and he said don't tell me about quantum gravity
Rajesh Gopakumar40:45
a laser beam is much more powerful than your regular light bulb. The laser beam has a coherent set of photons and that's the Bose Einstein effect.
Rajesh Gopakumar55:08
the point was that you you you are not just a cog in the wheel like you are might be in some of the welloiled machines of Stanford or Princeton or Harvard or some places.
Rajesh Gopakumar58:10
as a nation I mean we are a many millennia old civilization we should be thinking in in in terms of the long term investments and that's what I think is going to to to lift the country.
Rajesh Gopakumar1:10:19
I think of it as a little genie on my uh table uh which uh I mean I never used to code.
Rajesh Gopakumar1:26:38
Figures
| Share of US GDP from devices using the laws of quantum mechanics | roughly 35%-40% | 35:44 |
| Naturally occurring elements | 92 | 8:17 |
| Non-US citizens in his Princeton theoretical physics PhD cohort | 10 out of 18 | 13:24 |
| Problems Gopakumar advises running in parallel | 3-4 | 50:59 |
Glossary
- Quantum gravity
- The theoretical goal of unifying general relativity with quantum mechanics; no complete theory exists yet, and string theory is a candidate framework.
- String theory
- A framework positing that elementary particles are different vibrational modes of one-dimensional strings, attempting to unify gravity with the other interactions and explain how spacetime emerges.
- Bose-Einstein condensation
- Large numbers of bosons entering the same quantum state at low temperature; borrowed here as a metaphor for the lasing effect when talent falls into phase.
- Synthesizability
- Whether a computationally predicted new material can actually be synthesized — the gate where AI materials discovery most easily fails.
How to listen
Founders and investors working on AI for Science, and engineers and scientists weighing whether to join a large platform or pursue hardcore research with a startup mindset.
The first 25 minutes of education-and-training stories can be played at double speed; start mid-episode if you want the arguments.