Panel: The Future of Quantum — Annotated Lecture Notes & Key Takeaways

If you couldn't make it to the QMIT launch panel, the short version is this: four people who work on quantum from completely different angles sat on a stage at MIT and got asked, over and over in slightly different words, how close we actually are. The answers were far more careful than the usual conference fare, and a surprising share of the hour went into pushing back on the story the public has been told rather than adding to it.
 

Setting Up: A Chemist Moderates a Quantum Panel

The moderator, Troy Van Voorhis, heads MIT's chemistry department, and he spends his opening minutes explaining why that isn't strange. His own field used to go by quantum chemistry — meaning classical machines running algorithms that imitate quantum behavior in molecules and materials. He joked that the field is currently shopping for a new name, which got the first laugh of the session, and said the QMIT launch sits right on top of a goal he's had for years: genuinely understanding the quantum mechanics of molecules and materials.
He also made a framing choice that shaped everything after it. The three panelists are among the youngest faculty at MIT, picked on the explicit logic that they're the ones who'll be around when the field gets wherever it's going. It's a nice bit of stagecraft, but it does mean you're hearing from people early in their careers rather than a cross-section of opinion.

Harry Zhou is joining EECS in March and is currently a postdoc at Harvard — though he amended the introduction to note he did his PhD at Harvard and is based mainly at QuEra Computing, the neutral-atom startup mentioned earlier in the event, while keeping a Harvard affiliation. His work is on fault-tolerant architectures. He rattled off the platform landscape he's designing against: neutral atoms, trapped ions, superconducting qubits. He also did his undergrad at MIT.

Anand is an associate professor in EECS, in the theory of computation group, working on the complexity theory of quantum algorithms. He got his PhD at MIT under Aram Harrow. Tsuen Wan Choi is the Weiskopf assistant professor of physics, working on non-equilibrium quantum systems and theories of quantum measurement, and he has no prior MIT connection — he applied as an undergrad and was rejected, as he put it, twice. That got the biggest applause line of the opening.

When each was asked what draws them to the field, the answers converged in a way I didn't expect. Harry described the design problem directly: given how much theory and hardware progress has arrived at once, how do you architect fault tolerance so the machines get somewhere useful? Anand came at it from physics — he started in theoretical physics, drifted from quantum gravity into quantum information, and gave what I thought was the cleanest formulation of the day. Theoretical physics and theoretical computer science are both asking what's fundamentally possible in the universe we live in; computer science just narrows it to what physical machines can compute efficiently. What hooked him is that every classical model of computation anyone has proposed seems to land at the same fundamental power, while quantum appears to break that pattern. Working out precisely how it breaks has even fed back into the foundations of quantum mechanics.

Tsuen Wan described his program as split evenly between basic science — reconciling statistical physics with quantum physics, which he says is best done in the language of quantum information — and applications like better measurement and better simulation. Then he took the split back. He doesn't really experience it as two programs; it's one pursuit of learning how to manipulate and control many-particle systems that behave collectively.
 

Wildest Dreams: Sensors That Think, Answers You Can Check, Simulations You Can Trust

Troy invited them to dream, and the three answers were refreshingly specific.

Tsuen Wan went first and opened by admitting he's biased toward his own current work. The idea, developed with a colleague whose name was inaudible on the recording, is to fuse quantum sensing with quantum computing. The standard pitch for sensing is that entanglement buys you sensitivity. He wants to go further: wire the sensors into a quantum computer so the data gets processed quantumly and in real time, rather than measured and then handed off to a classical machine. He claims a provable computational advantage for at least one example he thinks is broadly applicable, and named two things he'd love to see come out of it — sharper gravitational wave detection, and a next-generation quantum-enhanced NMR.

Anand split his answer across two horizons. The nearer one borrows the central asymmetry of P versus NP: some problems are brutal to solve and trivial to check, and that structure may carry over to quantum. A quantum machine could produce something no classical machine could, while the answer remains verifiable. Shor's factoring algorithm is his model case — factoring is believed hard classically, but you confirm the result by multiplying. What he wants is an experiment where a classical computer can check the output and we're confident no classical simulation produced it. He said there's been real progress and partial results, but called the current state inconclusive, and guessed the clean version lands within roughly a decade. On the longer horizon he's watching for surprises in the complexity landscape, pointing to an idea from Oded Regev that started in cryptography about twenty years ago and is only now being converted into algorithms that might speed up some optimization problems.

Harry seconded the surprise point and singled out decoded quantum interferometry as something that genuinely startled him. Then he gave what I found the most honest passage in the whole panel. In grad school he worked on many-body dynamics — with Tsuen Wan, as it happens — and he was persistently bothered by whether the effects he was seeing on noisy hardware were real physics or artifacts, since dephasing can generate signatures that look a lot like the phenomenon you're hoping to find. That discomfort, he said, is a large part of why he moved into quantum error correction: he wanted some assurance that he was seeing the system and not the noise. Now the hardware is approaching fault tolerance, so simulations can actually be trusted to teach us something new, in many-body physics first and then in application-relevant areas.
 

Why QMIT Exists — and the Application Nobody Can Name

Troy's own answer to what QMIT adds was modest: as a chemist he wouldn't otherwise be sharing a stage with two EECS people and a physicist. Anand's answer was funnier and more pointed. At computer science conferences, he said, ask anyone what quantum computing will be good for and they'll wave their hands and say simulating chemistry — then admit they haven't opened a chemistry book since high school and don't talk to chemists. Simulating physical systems is plainly going to matter, but the people who know the algorithms don't know the systems, and vice versa.

Harry extended it to co-design. Finding an end-to-end problem that someone in industry genuinely cares about is itself hard, and the hardware platforms differ enormously in which operations are cheap and how much can run in parallel — something you can't see from the architecture side or the application side alone.

Tsuen Wan was the most enthusiastic, and also said the thing I'd flag as the most important sentence of the hour. The field is intrinsically mixed, he argued — his colleague Seth Lloyd introduces himself as a "quantum mechanical engineer" because he sits in mechanical engineering — and you can't draw a line between what the engineers on the panel do and what the scientists do. In his telling it's actually inverted from how it looks. QMIT, for him, is the first structure at MIT that pulls the whole research and education spectrum under one roof. And then, immediately after praising the field's transformative power across computing and sensing, he said we don't have a single pinned-down application. The only path to finding one is getting engineers and scientists into the same room.

That admission sits oddly against an hour of forward-looking optimism, and nobody on stage tried to reconcile the two.
 

Where the Hype Outruns the Hardware, and Where It Lags Behind

Troy asked where public perception and expert perception diverge most, and what the honest read would be. This section was the strongest of the panel.
 

The qubit-count illusion

Harry took the hardware numbers. Press coverage celebrates devices with hundreds of physical qubits, and the public reads that as hundreds of usable qubits. Not remotely close, he said. The demonstrations people are excited about — he cited Google — burn a large share of the machine's resources producing a single logical qubit, and even that is only just working. The resources needed for deep computations remain a long way from the headline figures.

Then he pivoted optimistic, and this is the part worth holding onto: two or three years ago nothing had been shown operating below the error-correction threshold, and now several platforms have. Once you're comfortably below that threshold, scaling turns exponential, so the next few years may move faster than simply extrapolating qubit counts would suggest.
 

The rocket-ship analogy, and why I half-doubt it

Anand offered an analogy he explicitly disowned — he can't remember where he heard it and apologized in case it belonged to someone in the audience. If you'd tracked the space race by plotting maximum altitude reached, progress would have looked stuck near a hundred miles for ages and then suddenly leapt. Wrong metric; you should have been watching engine size. In quantum, he suggested, qubit count is the altitude and threshold behavior is the engine, and on that measure the field is moving well, possibly faster than people expected.
I want to flag this as the weakest reasoning of the session, because the move is hard to argue with and easy to abuse. Pick your own metric, then report that you're doing well on it. Being below threshold is a genuine and hard-won milestone, but it's a statement about error rates, not about anyone being able to do anything useful. No one named a threshold value, a target logical-qubit count, or a date, so the claim that we're accelerating rests on a curve the audience never got to see. The analogy is vivid; I don't think it carries the weight he put on it.
 

The other gap: what quantum won't do

Anand was much blunter in the other direction. The public assumes quantum machines will speed up a broad range of optimization problems, or machine learning. He said there's little evidence for that. There are real and massive speedups for certain problems — factoring being the evidence — but you can't shove your favorite NP-hard optimization into a quantum box and expect something that beats the penalty you take from the machine's far slower clock rate. His read: the field is still doing basic science on what these devices can and can't do. It isn't sitting on an application waiting for hardware to catch up.
 

Two gaps pointing in opposite directions

Tsuen Wan reframed the question as multidimensional rather than a single gap. One side is overestimation — the public believes current machines are further along than they are, and news coverage repeats figures that are technically true but misleading, like a calculation that supposedly took some number of minutes. The other side runs the opposite way: the public also underestimates what the underlying science might eventually do, precisely because nobody knows where the boundaries sit or how fast things could shift. So he declined to say whether the total gap is large or small.

I wish someone had pushed him on the misleading-numbers point, because "technically true but misleading" describes most quantum press coverage, and he named the pattern without naming a single example.
 

What Would Count as Proof, and What Kind of Thing Quantum Is

Troy then asked what headline would make them think the phase transition is near, or already behind us.
 

Usefulness, loosely defined

Tsuen Wan went first and chose usefulness as his metric, immediately clarifying he doesn't mean profitability. His argument is that current devices already produce measurements we learn physics from, and the process resembles data science: you run an experiment partly on a whim, get something unexpected, and it turns into a serious line of work. He raised the obvious objection himself — sometimes the same insight could have been reached without a quantum simulator — and answered it by saying we probably wouldn't have reached it, because we wouldn't have done the experiment.

That's a counterfactual you can't test, and I think he half-knew it. What's clear is the aspiration attached to it: he wants that usefulness to spread out of his corner of physics into other engineering fields, then out of academia entirely and into industry.
 

Verification, and progress you can bank without the hardware

Anand's answer was the most concrete, and it's the one I'd actually watch for. He likes a minimum bar that Google has proposed: run the same computation on two machines with different architectures and you should get the same answer. He noted they've claimed a computation verifiable that way, but as far as he knows nobody has independently confirmed it. More results in that mold, on progressively harder problems, would raise his confidence — two unrelated devices agreeing on a signal that no classical method reproduces would be a real event. Above that sits a much higher standard he mentioned in passing, an interactive proof that the computation was done correctly.
Then he made the point I found most interesting: theory can move us closer to an application without a single device improving. His example is quantum money, the idea of exploiting the un-copyability of quantum information to build currency that's unforgeable in principle. The research program is to construct it from weaker and weaker cryptographic assumptions, the prize being public-key quantum money, where anyone can authenticate a note without holding a secret. Nobody has managed it from assumptions the community broadly trusts. If someone did, he'd count it as major progress in quantum information science even though no such banknote exists, because the hardware to hold the states isn't there yet.

Harry added a caveat that's easy to miss: these verification tasks usually come as a parameterized family, so a demonstration only carries weight once the instance is large enough not to be trivialized by its own small size. Imperfect as a proxy, but a proxy.
 

Not the end of humanity

Troy closed on science fiction, noting that in most stories inventing a quantum computer is one step short of the apocalypse, and asking what metaphor they'd use instead. It's a slightly leading question — he set up the answer he wanted — and all three took it to varying degrees.

Tsuen Wan predicted that whatever explosion happens behind the scenes will reach the public as something smooth and nearly inevitable. His example was energy: a word we use constantly, barely registering as technology, yet an abstract quantity you can't touch or hold that took centuries to formulate. He imagines a future where people gripe about entanglement the way they gripe about battery life or a slow transaction — "my phone is so bad, entanglement connections," which got the laugh he was aiming for. Transformative, in his framing, but not disruptive.

Anand located the real disruption risk in cryptography. Quantum machines break factoring-based schemes like RSA; candidates exist to replace them and people are reasonably confident; NIST is working on standards and rollout, and he said we should already be moving. Then: who's to say another Shor won't come along and break one of those? A world without public-key cryptography would look drastically different from ours, given how much of the internet and communications infrastructure depends on it. He called the scenario unlikely but possible.

Harry went at it from the frontier angle. It's a computational regime we've never had access to, and he reached for the Higgs boson as a comparison — a fundamental-science hunt that also spun off things like the internet. You enter a regime you know is different, you get to do science you couldn't otherwise do, and the consequences can be interesting "or maybe perverse," cryptography being his example. What we need, he said, is both: learning to use the tools and staying alert to what they do to us.

One caution on that last analogy, since it's the kind of thing that gets repeated: the web came out of CERN generally, not out of the Higgs search specifically. It's an offhand comparison in a closing answer and I doubt he meant it as a historical claim, but don't write it down as fact.

The through-line of the whole session, if you want one, is that the people building this field are noticeably more cautious about timelines than the people funding it, and noticeably more excited about the basic science than about any product. Whether that's intellectual honesty or a convenient place to stand while the hardware catches up is a question the panel raised without answering, and I'd rather flag it than resolve it for you.