D-Wave Founder's New Startup Combines AI, Robots, and Monkeys in Exo-Suits | Hacker News Reader

D-Wave Founder's New Startup Combines AI, Robots, and Monkeys in Exo-Suits

spectrum.ieee.org

86 pointszxcvvcxz61 comments

These d-wave guys are the greatest scam geniuses in history. First they make a "quantum computer" and now they're building trainable exo-suits. I don't even know where they come up with this stuff.

I'm going to have to ask you to expand on this comment and provide some context for why you think they're "scam geniuses"?

There is more-thoughtful discussion of the D-wave controversy in this thread - look for dsacco's comments.

They have consistently dodged providing any kind of real evidence and managed to survive this far purely on marketing hype. I think someone recently posted about the Bogdanov affair. D-wave is the analog of the Bogdanov affair for quantum computers.

So far D-Wave has failed to provide any direct or indirect evidence that their "quantum computer" actually achieves quantum speed up. An investigation by a scientist from ETH Zurich who was able to use their machine came to the conclusion that their quantum algorithm does not outperform a classical system.

It's hard to think of a reason to refuse to publicly prove that your quantum computer is actually working if it really is, if it isn't it makes a lot of sense of course.

Concerning the purchases of Google and the US government: These are probably more like research grants for D-Wave, as the buyers expect that at some time there might come a working quantum computer from D-Wave, and buying a prototype (even a faulty one) puts them in a good position to be the first to actually get a working version, while helping the company to stay afloat. By the way, Google invested 200 million USD in a different quantum computing technology endeavour led by a UC Santa Barbara professor, which would make little sense if they already had a working quantum computer.

The value of D-Wave as a company is not only defined by their ability to build a working quantum computer though, as there could be significant value in the patent portfolio they built / acquired and in their expertise in building and assembling superconducting chips and systems, which are both highly non-trivial tasks.

Their way of communicating with the public is of course highly misleading, which is why many people in the scientific community have very little respect for them, as this is viewed as highly unethical and harmful behavior in science (but not necessarily in the business world).

The UCSB professor you're referring to is John Martinis [0] and though the physical structure differs (they're using small arrays of "Xmons" instead of typical transmons qubits to minimize decoherence and thus error) they're still working within the realm of adiabatic quantum computing/quantum annealing, the same general avenue D-Wave is pursuing (with a "digital" twist) which is why Google invested in both of them.

[0] http://web.physics.ucsb.edu/~martinisgroup/publications.shtm... [1] https://research.googleblog.com/2016/06/quantum-annealing-wi...

Sorry that's just not correct (to my knowledge at least), as John was always working on gate-based quantum computing and is currently pursuing designs based on resonator-coupled Transmon qubits (the X-mon refers to the geometry of the qubits, which provide four capacitive ports to couple them to resonators and fast flux lines). The next milestone for their collaboration is the demonstration of "quantum superiority" using a 40(-ish) qubit processor to perform an algorithm that is beyond the reach of current supercomputers. The real effort comes afterwards though as the system needs to be scaled up in order to be useful in any real-world task.

The quantum annealing example from the Google research blog seems to use normal quantum gates to simulate the nearest-neighbor interactions in a spin chain. The nature article is currently down but it seems they use "Trotterization" to simulate the spin chain using one- and two-qubits gates.

It is indeed, correct. The annealing example I linked is from his team (here's another one with full citation with Martinis at the end there [0]). And if you want further proof that Martinis and the Google Quantum team are pursing quantum annealing look no further than the Adiabtic Quantum Computing Conference that was held this summer where their team held several talks including "Building Quantum Annealer v2.0" [1].

And I didn't touch on your earlier (erroneous) comment on the scientific community's perception of D-Wave, but I think it needs to be said that in actual professional circles the research they're performing isn't met with as much derision as they seem to garner in these more causally informed settings. It's hyped and a difficult subject to understand, so that's fair, but I suggest really informing yourself if you're going to go out there and make the claims you're making.

[0] http://research.google.com/pubs/pub44292.html [1] https://aqccreg2016.eventfarm.com/events/index/7fff5387-0000... http://1qbit.com/insights-inside-googles-annual-adiabatic-qu... [2] (Another article reviewing QA with Martinis and team) http://arxiv.org/abs/1512.02206

I actually did my PhD on superconducting quantum computing with Transmon qubits, so I really wouldn't say that I'm only "casually informed" about the subject.

Again, their digital adiabatic quantum computing relies on a gate-based quantum processor, which provides a universal set of qubits gates and thus can in principle simulate any Hamiltonian, quantum annealing is just an algorithm that runs on it.

And I don't want to be overly negative, but the perception of D-Wave in "more professional" circles was not very enthusiastic in my experience, which to a large extent is also D-Waves fault.

[...]in actual professional circles the research they're performing isn't met with as much derision[...]

Yes it is.

What's wrong with doing pure research as a Startup? Why does pure research have to only be the domain of Universities? I think it was a great experiment that failed, but that probably goes for something like 95% of great experiments.

Lots of pure research goes nowhere. It's not like Geordie is defrauding the old and disabled. He's working with really really really smart people with lots of money.

Honestly, we need more of this and less of SnapChat startups.

What's wrong with doing pure research as a Startup?

Because there is no reliable way to fund it. Said another way, the time horizon for repayment is too long and too risky for traditional, non-corporate, non-public investing.

Lets assume you have a breakthrough with research, now you have to make a product to sell. Which means you need to likely build a new market, or even in the best case of an existing market change the team composition to fund product development.

Those are all risks that Banks and Venture Money (except in very rare cases) aren't willing to take.

There's nothing wrong with doing pure research as a startup, in fact I'd love to see more of these kind of startups and I'm thinking about starting one myself!

What's harmful (IMHO) is making spectacular claims (look we have a working quantum computer!) and then not back them up by evidence. It is of course okay to sell your product aggressively as a startup, and sometimes to oversell it as well (fake it until you can make it), but I really have my doubts that D-Wave can "make it" in the near future, as there are just so many difficult problems to solve for which we don't even have a viabe approach yet (e.g building large scale qubit coupling schemes), and trust me most quantum computing research groups have people that are at least as smart and dedicated as the people working at D-Wave.

So the high profile customers (US government among them) have fallen for some tech con artists, right?

This is a common fallacy. The US government is not that great at decision making actually. Mistakes abound and not just tiny mistakes but mistakes that consistently cost billions. The history of the fighter jet program is quite interesting in that regard. The healthcare.gov debacle is another recent example. I mean this isn't a ding against the government. Any large enough organization is terrible at making good decisions consistently. So just because the government and some other high profile people have invested in d-wave doesn't lend them any more credibility in my opinion.

The other high profile guys I'm assuming you mean are Google and IBM. It would be interesting to know how d-wave "quantum computers" have actually made any kind of dent in their respective companies other than pure marketing hype so that they can say they have "quantum computers".

It's perfectly reasonable for a large organization like some part of the USG or Google or IBM to simply buy one to see what it's like. If there actually is a there there (perhaps even one not known to the developers) they can learn something. And if there's something real (even if not useful) they can gain a leg up on users of the next generation system, perhaps from a different company who actually does know what they're doing.

And if it is a flop, well, the marginal cost is pretty minor. When I was at MIT we bought all sorts of crazy things some of which turned out to be useful in unexpected ways once you actually had the equipment in your hands.

A good example of this was the EmDrive which really seemed like a scam to me but might have a there there. I wouldn't have believed it at all except that NASA, a seemingly independent organization where this is concerned, saw enough that it's apparently worth looking further. If it had been the 100% scam I assumed (and that's still not clear) I'd still be OK with the money spent.

At the moment I still don't believe D-Wave and dinnertime discussions with friends at Google have not caused me to change my opinion.

That's fine. I don't disagree with that approach. What I find faulty is the incorrect reasoning applied based on that behavior. It is suspiciously close to appealing to authority type of argument. Just because a large organization takes a chance on something doesn't mean there was actual planning involved in the process other than as you said "let's just see how this pans out".

If they actually made that policy, that they will buy any <500k widget that promises the world, then I bet the market could churn out pseudoscience at astonishing rates.

The market already churns out an astonishing amount of pseudoscience. You don't need any special policies from big corporations for that. Magical thinking is one of those human things that seems to be deeply ingrained and requires sustained effort, practice, and training to avoid. Science and scientific thinking does not happen on its own. No one is born a scientist.

Whenever D-Wave comes up on HN I'm always impressed to see how many posters who casually read one Scott Aaronson blog (not even the most recent ones mind you) think they're more qualified than legions of classically trained engineers and physicists from Los Alamos, Lawrence Berkeley, ETH Zurich, MIT Lincoln Lab, NASA, Lockheed Martin, Google, et. al.

D-wave isn't selling a "universal" quantum computer, but the platform they are and have continued to deliver has shown real, tangible progress with respect to quantum tunneling (even though their coverage has been typically questionable, the best layman's explanation of their most recent results happens to be from Ars [0]). No we're not getting a "universal" quantum computer anytime soon and no they're never going to be accessible to average consumers (so remain skeptical of anyone who says otherwise) but those claiming D-Wave is a scam/con are absurd.

[0] http://arstechnica.com/science/2016/02/is-d-waves-quantum-pr...

All modern computers are based on quantum effects, but Intel does not call it's i7 CPU a quantum computer. The D-Wave really does not fit the established criteria for a quantum computer.

However, they really want the publicity from calling it a quantum computer so they continue to do so.

There was also problem where the speedup might have happen with classical computing. They were comparing their algorithm for their architecture against a single CPU implementation done same way in one paper. I've seen FPGA's get 3 digit speedups on problems they could brute force when redone for their architecture. Something the size and cost of D-Wave machine could easily use something like FPGA's, ASIC's, and/or NUMA links. I'm not sure it would achieve the exact results but I'd like it ruled out that they didn't build a MPP of ASIC's for the specific problem.

this is gratuitously negative and against HN policies and you should not write comments like this here (especially using the word scam). a neutral phrasing might be "I believe that the D-Wave guys do not themselves believe in their innovations, and are perpetrating what would otherwise be considered scams."

Please bear in mind that Google as well as NASA have bought D-Wave computers. Whether the effect is real or not, they do not exactly transfer money to scammers, and are not easy, unsophisticated "marks" (targets of scammers.)

"Hey Google, it's your cousin!" is not how these sales take place. Use of a word like scam is inappropriate, especially given that sales have been made, and the ordered equipment has been delivered and is being used. (By people who run Google and who send things to space.)

Please be a bit more judicious in your phrasing.

Have you ever been involved in an M&A meeting? You'd be surprised how often "Hey Google, it's your cousin!" kinda thing actually happens. I'm surprised by the amount of faith and assumed competence on the part of these organizations among the HN crowd.

The comments I'm seeing as response I agree don't seem to be conducive to an actual discussion so I'll rephrase but the point about the intelligence and infallible decision making abilities of the large organization being used as counterclaims do not really add up. Extraordinary claims require extraordinary proof is the usual bar for claims that go above and beyond current state of the art and science. D-wave has consistently failed in that regard. I find their approach unscientific and unethical for the kinds of claims they are making for whatever technology they have sold to whatever organizations. In the long run no one wins when such tactics are used.

Oh, all I meant with my exmaples is that it should be enough for you to pause to phrase judiciously and carefully, for good discussion here.

Unfortunately I think you're outside of the edit window though. the phrasing you just used (in your second, longer paragraph) is fine to me! It should have been your original phrasing (both IMO and by HN guidelines.)

Thanks for following up.

Has D-Wave made any actual feasible technical achievements WRT quantum computing or are they just the Theranos of that domain?

Heh, harmless theranos.

It's hard to say. Industry experts like Scott Aaronson have gone back on forth with regards to whether or not it's 1) truly a quantum computer and 2) capable of real improvements over classical computers for anything nontrivial. Point #2 is perhaps more important for practical purposes.

Aaronson is currently back in the skeptical camp after a brief stint of "cautious enthusiasm" in 2012. Umesh Vazirani, Aaronson's PhD advisor and the grandfather of quantum computing, has been largely dismissive of D-Wave and doesn't believe their computers will be capable of much even if they are truly quantum.

The technology is apparently impressive enough that Google, NASA and Lockheed Martin have partnered with D-Wave to test its capabilities, but to date it has not done anything revolutionary. I'm sure they have done something, but it's not clear what trade secrets they have can be separated from marketing, and their public work has been reproduced on classical computers.

There have also been a number of articles in Nature Physics, Science and elsewhere by different authors going back and forth on points 1 and 2 above. The current consensus is that D-Wave is not creating true quantum computers; but they have developed classical, pseudo-quantum computers with legitimate entanglement implemented, and which do not currently offer any real quantum speedup or improvement over classical computers.

That said, I mostly follow quantum cryptography, so I haven't read this research in a while. I am personally happy that D-Wave exists and is trying to push forward the field - I think comparisons to Theranos are uncharitable. But I am skeptical of their real advancements.

Edited to clarify after reviewing the research again.

> That said, I mostly follow quantum cryptography,

Which is also a field full of bullshit and nonsensical claims, and it still has to prove that it provides a solution for any real problem.

If you're interested, the Quantum Key Distribution [0] sub-field has already likely bore fruit that is a "ready in the medium-term" useful solution to a real problem. It's probably already deployed somewhere in the national security/defense space.

[0] https://en.wikipedia.org/wiki/Quantum_key_distribution

DJB is still skeptical of Quantum Key Distribution.

https://sidechannels.cr.yp.to/qkd/holographic-20160326.pdf

Which problem exactly does QKD solve? (It is used by some banks afaik, but that it's used doesn't mean it solves any problem - it just means someone bought the marketing claims from ID Quantique and co)

A quick response (so forgive the brevity): Let's say worst case, that a quantum computer does not currently exist, but its basic logical operation has been fleshed out to a reasonable extent. Now, what if a practical implementation shows up in the 5-10 yr timeline. The issue is that some quantum computers could, in theory, pose a real threat to many currently used classical encryption systems. QKD + one time pad could provide, in theory if not already also in practice, a reasonable stand-in solution to counter this perceived threat.

This is not true. Quantum cryptography is frequently conflated with concepts that are not possible and don't make sense, but quantum key distribution is a legitimate area of research. In fact, it has already been experimentally implemented (Los Alamos, UCambridge, UGeneva, etc).

It might not be mature enough yet for widespread adoption, and it doesn't revolutionize all parts of a cryprosystem, but it is definitely progress in the narrower problems of key distribution and perfect forward secrecy for one-time pads.

Unless I'm misunderstanding you and what you mean is that the field has many people opportunistically using the term for marketing things that are not really quantum cryptography; if that's the case, I agree with you.

> but quantum key distribution is a legitimate area of research

If you used that with the meaning "we don't know how to do anything useful with it, but it is worth looking because there may be some use", I would grant you this point. But:

> In fact, it has already been experimentally implemented (Los Alamos, UCambridge, UGeneva, etc).

No. The existing protocol for quantum key distribution is insecure bullshit.

Maybe it's a legitimate area of research. But I hear frequently obvious bullshit claims from these researchers, like that they want to build a quantum internet.

And I don't see how quantum key distribution solves any problem. It's kinda circular - it requires an existing authenticated channel to work, which needs math-based crypto. At the same time the whole field is built on the claim that they're needed because they don't trust math-based crypto.

"It might not be mature enough yet for widespread adoption, and it doesn't revolutionize all parts of a cryprosystem, but it is definitely progress in the narrower problems of key distribution and perfect forward secrecy for one-time pads"

I'm not sure. The best minds I've talked to in high-assurance security don't believe it. One, Clive Robinson on Schneier's blog, predicted we'd see classical attacks on it. Something like that happened. I pointed out there would still be software, endpoints, interception, subversion, etc for bypass whereas the tech itself was little understood versus effects of regular electronics. Tried and true beats novel and new if it's high-security I always say.

No, these devices are not trustworthy or even necessary that I can tell. High-assurance key exchange is a solved problem if the organization is the simpler, hierarchical designs likely to use QKD. Many solutions available with the most reliable being a courier and key-handling hardware you can trust that only has part of the secret + another transmitted over various medium with it all using several layers of symmetric to work (eg my polymorphic schemes). All on dedicated, low-TCB devices.

I don't know what QKD costs today but I bet it's still cheaper to courier the pre-shared keys and use tamper-resistant hardware. Do you have numbers on it in terms of amount of keys negotiated over time and at what cost for hardware w/ HA setup?

> It's hard to say. Industry experts like Scott Aaronson have gone back on forth with regards to whether or not it's 1) truly a quantum computer and 2) capable of real improvements over classical computers for anything nontrivial. Point #2 is perhaps more important for practical purposes.

I down-voted this. I have only a layman's understanding but from what I have read particularly #2 is demonstratively false, D-Wave has not achieved any improvements over classical computers for anything (including it's very own domain problem, which Alex Selby's classical algorithm outperforms D-Wave on). [0][1]

[0] http://www.scottaaronson.com/blog/?p=2555#comment-974407

[1] http://www.scottaaronson.com/blog/?p=1400

I should clarify - I agree with you in that #2 is demonstrably false. But Aaronson has gone back and forth on his skepticism for D-Wave.[1] He is now rightfully back in the skeptical camp, however.

My main point was that acknowledged experts have gone back and forth on this and there have been conflicting publications about D-Wave's computers. Now that I review the research again, it looks like it's been experimentally shown to have legitimate entanglement implemented in the computers, but the consensus is indeed that there is no quantum speed up or improvement or classical computers.

Thanks for the reply though, I've edited the original comment.

[1]: http://www.scottaaronson.com/blog/?p=639

The parent said capable of real improvements over classical computers, not whether it has achieved it. Of course some hold the opinion that it's at least capable of it (Google, NASA, D-Wave employees) but the jury seems to remain out.

No point to refer to "industry experts". Either data shows speedup, or it doesn't. Things will become clear after they release the next version [0]. If there's x1000 speedup, as claimed, it's a real deal, no matter if it's "really quantum" or otherwise.

[0] http://www.dwavesys.com/media-coverage/techrepublic-quantum-...

>If there's x1000 speedup, as claimed, it's a real deal, no matter if it's "really quantum" or otherwise.

Except not, because constant time speedup isn't interesting, 1000x classical computers will get you the same effect probably for less $ than d-wave. The entire point is that a genuine quantum computer should achieve an asymptotic speed-up not a constant factor one, which no one really cares about.

Scott Aaronson's posts on D-Wave are worth reading in full, including the discussion in the comments. He's a brilliant guy and his intellectual honesty is commendable. In other words, he's willing (and excited) to be proved wrong.

> The technology is apparently impressive enough that Google, NASA and Lockheed Martin have partnered with D-Wave to test its capabilities

As someone who has actually worked in the research divisions of NASA and Lockheed, I wouldn't assign this form of social proof any value. There is absolutely nothing to infer from fact that NASA and Lockheed bought prototypes. I don't have personal experience with Google, but it is not unlikely that the same effects are at work.

Google is having success with their DWave[1], test results on the Google Research blog[2].

[1] https://www.technologyreview.com/s/544276/google-says-it-has..., "Google Says It Has Proved Its Controversial Quantum Computer Really Works"

[2] https://research.googleblog.com/2015/12/when-can-quantum-ann..., "Quantum annealing significantly outperforms its classical counterpart, simulated annealing. It is more than 10^8 times faster than simulated annealing running on a single core".

I was talking with D-Wave about a year ago about applying their system to some defense projects I was involved with and I went through some training with their systems integration team.

The way their system seems to work is that you have to transform your problem set into a specific D-Wave function that will be topologically graphed into the D-Wave chipset. From there is finds the global minimum immediately. The challenge is really in having a problem that fits this type of transformation and doing it in a way that it's actually valuable. So it's quite a bit of work just to do the transform and we couldn't see an easy way to take some of our general problems and turn them into the graphs needed to work.

In theory their stock functions seem easy to implement but when you investigate further it's a pretty narrow set of problems that fit.

> The operator can use foot pedals to control the robot’s movement and a virtual reality headset like the Oculus Rift to experience what the robot is seeing.

Maybe it's a bit off topic, but I'd like to put here fiction sources (which I know) that explore this kind of setup:

1) A Cognitive Discontinuity story by Andrej Karpathy (2015): http://karpathy.github.io/2015/11/14/ai/

2) Sleep Dealer movie (2008): https://www.rottentomatoes.com/m/sleep_dealer/

Regardless of the actual/implied progress made by D-wave to build a quantum computer. The fact that its CTO of 17 or so years decides to move on to greener pastures / take on a side mission, does not bode well for D-wave

That's what I was thinking. Mechanical exoskeletons are not innovative - They already exist. Quantum computing on the other hand; that's highly innovative.

Why would someone who is supposedly at the forefront of this highly innovative field throw away years of experience and start again from scratch in a comparatively much simpler field.

This doesn't inspire much confidence in D-Wave.

One way to gather all the data for training using deep learning, is run millions of robots. But that may happen if tele-operated robots become a big industry in of itself - which economically could work.

So the big issue is: what is the state of the art in tele-operated robots ? And what's missing for it to become a thing ?

We are definitely not going to do deep learning for robots that way: we'll run simulations inside virtual worlds.

http://spectrum.ieee.org/automaton/robotics/artificial-intel...

Why did the founder leave d-wave?

Is it not possible for someone to have more than one startup?

As someone who did it, it's really hard! You really want to be all-in on both, which of course is impossible.

Monkeys in Exo-Suits. So ripe for parody as a Gavin Belson moonshot.

I know a few years ago there was some amount of skepticism on whether the Dwave machines were atually quantum computers. Has that skepticism been put to rest? Can anyone comment?

edit: I missed the comment below regarding the designation "universal" quantum computer. I guess that was the crux of the skeptic argument?

Isn't this a lot like Disney's robot:

http://spectrum.ieee.org/automaton/robotics/robotics-hardwar...

That’s not a robot, that’s a “waldo” – a puppet that is being controlled remotely.

what's the difference between an exo-suit and puppet?

I was only objecting to calling Disney’s “Jimmy” a “robot”, because is isn’t one; nothing else.

Reply on news.ycombinator.com