The Intersection of Quantum Computing and AI: A Conversation with Dr. David Anderson

Our conversation today is centered around the fascinating field of quantum computing, where we explore its potential applications and the current state of research. We're also joined by Dr. David Anderson from NASA's Jet Propulsion Laboratory, who shares his expertise on the intersection of quantum computing and AI. Our conversation takes us through various aspects of this emerging field, including machine learning, neural networks, and the role of expert artificial systems in solving complex problems.

Dr. Anderson's introduction to his work highlights the importance of understanding that AI is not limited to machine learning alone. He mentions that there are many methods considered AI, even before the hype surrounding machine learning took off. Dr. Anderson explains that his team at the quantum AI lab focuses on developing expert artificial systems to tackle complex problems, such as planning and scheduling for robots operating on distant planets.

The intersection of quantum computing and AI is an area of ongoing research, with several approaches being investigated. Dr. Anderson notes that machine learning can benefit from quantum computing, particularly in training new neural networks. He also touches upon the concept of quantum neural networks, which utilize quantum correlations to implement more efficient algorithms. These advancements hold great promise for solving complex problems in the field of AI.

However, Dr. Anderson emphasizes that there is still much to be learned about the application of quantum computing to traditional machine learning tasks. He mentions the challenges involved in compiling problems into a quantum computer and calibrating its performance. Despite these hurdles, researchers are actively exploring ways to harness the power of quantum computing to enhance AI capabilities.

Throughout our conversation, Dr. Anderson highlights the potential benefits of combining quantum computing with AI. He notes that there is still much to be discovered about this emerging field, and he encourages listeners to explore resources such as lecture notes and blogs from prominent researchers in the field.

One of the most significant challenges facing the intersection of quantum computing and AI lies in creating robust and reliable solutions for practical applications. Dr. Anderson emphasizes that while there are many tutorials and resources available for learning about quantum computing, they may not be immediately accessible or user-friendly.

To better understand the intricacies of quantum computing and its potential applications in AI, Dr. Anderson recommends exploring review materials and lecture notes from prominent researchers in the field. He also suggests checking out the blog of Scott Aaronson, a well-known advocate for quantum computing, which provides insightful discussions on various aspects of this emerging technology.

Throughout our conversation, we discussed the importance of ongoing research and collaboration between experts in both fields. Dr. Anderson emphasized that his team at NASA's Jet Propulsion Laboratory is dedicated to pushing the boundaries of what is possible with quantum computing and AI. He encouraged listeners to join him in exploring this exciting area of research, whether through online resources or by reaching out directly.

As we conclude our conversation, it's clear that the intersection of quantum computing and AI holds vast potential for groundbreaking discoveries and applications. By continuing to explore this emerging field, researchers and enthusiasts alike can unlock new possibilities and drive innovation forward. Dr. Anderson's insights provide a valuable starting point for those looking to delve deeper into the world of quantum computing and AI.

Resources:

* Lecture notes from Caltech

* Scott Aaronson's blog

* Resources on quantum computing and AI at twillie Icom

For more information on de'vide or any of the topics covered in this episode, please visit the show notes page. To share your thoughts or ask questions, feel free to comment on our website or reach out via Twitter at @tormal AI.

"WEBVTTKind: captionsLanguage: enhello and welcome to another episode of we'll talk the podcast where I interview interesting people doing interesting things in machine learning and artificial intelligence I'm your host Sam Charrington and we are back first things first happy new year everyone and welcome to 2018 I had a great holiday but I've definitely been itching to get back to the show as most of you know at the end of last year we held a listener giveaway to celebrate hitting one of our biggest milestones to date 1 million plays of this podcast Thanks to everyone who entered we sent out an email to entrants a few days ago so please be on the lookout for that if you haven't heard from us yet just reach out to us at team at Twin Olay I calm so that we can get you your swag next up the details for our January meet up are set next Tuesday January 16th we'll be joined by veteran twimble guest and Microsoft researcher Tim knit gabru Tim net joined us just a few weeks ago to discuss her recently released and much acclaimed paper using deep learning and Google Streetview to estimate the demographic makeup of neighborhoods across the United States and I'm excited that she'll be joining us to discuss her paper and the pipeline she used to identify 22 million cars in 50 million Google Street view images I'm anticipating a very lively discussion segment as well in which we'll be exploring your AI resolutions and predictions for 2018 for links to the paper or to register for the Meetup or just to check out previous meetups visit Twilio comm slash Meetup finally a bit about today's show I'm joined by David venture le science operations manager and quantum computing team lead for the university's Space Research Association's institute for advanced computer science and NASA Ames divi joined me backstage at the NYU future labs AI summit a while back to give me some insight into a topic that I've been curious about for some time now quantum computing we kick off our discussion with a review of the core ideas behind quantum computing including what it is how its applied and the ways it relates to computing as we know it today we then discussed the practical state of quantum computers and what their capabilities are as well as the kinds of things you can do with them and of course we explore the intersection between AI and quantum computing how quantum computing may one day accelerate machine learning and how interested listeners like you can get started down the quantum computing rabbit hole and now on to the show all right everyone I am here at the NYU skirble center backstage just as we're finishing up the NYU Future labs AI summit and I have the pleasure of being seated with David venturelli who's the science operations manager for the research institute for advanced computer science at NASA Ames daveed welcome to the podcast thank you thank you very much it's great to have you on the show and I am really looking forward to this conversation because you are doing a ton of work in an area that I know very little about but keep hearing a lot about and that is quantum computing so before we get into me peppering me with questions wanna you just tell us a little bit about your background and how you got into quantum computer self yes so I'm a physicist I'm at your ethical physicist and I did study quantum mechanics like every physicist does back when I was a PhD students in 2008 seven the words - inside the physics community about quantum computing but it's nothing but it's nothing like what we have today what I did I did study quantum information science studied nanotechnology and of course I have seen in front of me unfolding the opportunities of actually not only validating the theories on pen and paper but now experimenting with real world quantum machines so I joined the quantum artificial intelligence laboratory in 2012 as one of the founding members NASA Google and us are a University Space Research Association decided to create this group to investigate the near-term impact of quantum computing for computational problems of national interest and that was you know what I've been doing in the last five years we managed projects in collaboration with government academia and with of course the principal stakeholders the private sectors as well to understand really if in the next five to ten years there could there could be an acceleration of computing because of quantum technologies great great I guess the place to start is on quantum technologies in quantum computing like I imagine that it is something that many in the audience have heard of but don't fully understand you know not just what the big ideas are but what are the the foundational ideas upon which it's built and I'm hoping maybe you can walk us through some of that yes the idea idea comes from the 80s when one of the gods of physics richard fineman made the observation at the conference pretty casual observation that if we were able to create a computer which worked with the laws of quantum physics then simulating quantum systems would not be very difficult while today it's very very difficult to do that we need supercomputers and bla bla bla so this observation changed the world because the idea of processing informations with the fundamental laws of nature as api's as elementary building block of the algorithms really got every physicist excited some early discoveries on the fact that the mathematics of quantum mechanics is fundamentally more powerful than the mathematics we employ when we process information in standard digits Computers really got everyone excited can you give us some examples of that yes so people have discovered in 96 that you can search a database and unstructured database so we're basically all the information is not arranged by any sorting normally to search that such database you do not have any smart way to do it because you just have to randomly make queries and if you're unlucky you will find what you're looking for is the last attempt of your query and that means that the problem has a linear complexity meaning that if you have n items you need n trials to get to the to the item you want in the worst case scenario now if you had a quantum database and the ability of doing quantum queries and you exploit all these effects of superposition interference entanglement you are able to find the item you want in a square root of the number of items so this means that instead of for example 100 queries you need only 10 and this quadratic speed-up is a very interesting phenomenon which shows by itself that quantum computing is more powerful than classical computing because there's no way you can beat this without quantum mechanics but what people is really after is exponential speed-up so the opportunity to devise algorithmic strategies that are able to accelerate exponentially problems there are some examples in cryptography in in chemistry where we know how to exponentially bypass the bottlenecks of computation so that problems that could be solved today in billions of years they would be would be solved by a quantum computer in seconds but this is not the case for a lot of things that we want to solve and we are investigating and it's so difficult to investigate these algorithms without having a quantum computer that in some sense we want to build quantum computers to understand whether it was worth to build in the first place so is it fair than to say what I took away from that is that quantum computers isn't the idea of kind of going piece by piece and replacing the various components on a current system board with you know quantum versions it's not like you know shrinking down the the physics is it is a fundamental new way of thinking about computers built around the it's a new paradigm it's not incremental improvement it's really a different fundamental way to process information of course quantum effects are already very well taken care of by the silicon industry in integrated circuitry already but in but today they kind of bother you you're because you want to shrink things you want to make them more fat faster now what we're talking about is assigning a functional role to the quantum effect and actually assigning logical states and variables to quantum states and then very finely very delicately orchestrate their quantum dynamics so that we do mathematical operations on our logical assignments and this is a new paradigm it is operating on probability distributions which are weird because they're not over real numbers they're over complex numbers so there are imaginary units involved and it's very interesting mathematically and funny enough for for the listeners of this show you don't really need to know physics to become a quantum algorithm person it's actually linear algebra so as long as you accept why quantum mechanics works you can study it in an after what if you have the mathematical background you can study it fast and just you don't know why it works it doesn't matter but you can start writing algorithms because it's just you know matrix multiplications and things like things like that what are some of the you kind of rattled off some quanta primitives that you know upon which this math and these computers are built can you kind of give us a next level of detail into you know some of the most important ones of course the main paradigm which is explored today is to use binary variables so bits basically but generalized in the quantum world so the idea of using quantum bits or qubits so you do identify a physical system it could be a superconductor or a circuit or it could be an item it could be an electron it could be a fault or anything anything which is sufficiently small called protected from the external environment follows the laws of quantum physics even cats and boxes apparently so as long as you are you're able to find that then you you look at the states of the system and then you identify what is you call zero and what you call one for example it could be in more states in a quantum system yes but you choose two of them okay so you say you say okay this electron can have spin up or spin down for those that know physics or this photon can have polarization in one way or in another this superconductor can have a clockwise current or a counter clockwise current and then you look at how you can act on the system so similarly as you do in chemistry if the system is quantum you can manipulate their state like in chemistry you can do chemistry reaction but this time you need to be very controlled so if it is an 8 ohm you need to shine a laser if it is a circuit you need to apply a magnetic field you know these kind of things and and when you do that you know how your state will change the funds the fun fact is that it is totally true in quantum mechanics that you can have a state which is not 0 not 1 it's kind of a superposition of the two and is represented by this abstract vector in complex number subspace where you represent 0 and 1 at the same time now when you look at it it would be 0 or it would be 1 with some probability but if you don't look at it then it's both you can operate on it like if it is both until you look at it so the algorithm the job of the of the programmer is to figure out a way to operate on this ensemble of qubits in such a way to orchestrate the probability at the end to be the solution of the problem you want to solve but that's a complex job because you know if you have n qubits what you are really describing is a state which has an exponential number of degrees of freedom so and you need to basically describe it with true to the power of n complex numbers and there's no way you can simulate this in a machine you need a quantum computer to experiment most of the time so it sounds like one of the fundamental realities of quantum computing is this chicken-and-egg problem like how do how do we get beyond that yes you're totally right we are getting beyond that we decided I don't know the chicken or the egg what we decided but we decided that it's worth to do it and we're starting there are companies IBM Righetti Google Intel they're all building chips and they're making them available to the research community and people are trying to use them in the short term to learn about quantum mechanics to validate the theories that don't pen and paper I mean divides the last 30 years so now we are we are really using quantum computers as tools for improving our knowledge of quantum information processing yeah so IBM until some of these companies that you've mentioned had been talking about this for ever like in in kind of research mode I think the thing that really caught my attention recently was I think it was their build conference one of the microsoft's recent technical conferences there was i think it was a keynote like their main announcement such an Adela came out and had some researchers and they kind of you know brought out this I'm not even sure what it was and look like a GPU to be honest they brought out this board that was their first attempt at a quantum computer or something you tell me what exactly that was it at significance yeah I know Microsoft as well as IBM as you said like everyone had some quantum computing expertise since the beginning of time meaning since since when the field has as a name and for good reasons I mean good physicist kept an eye on the idea since the beginning and and this company is higher the very good people now what change now is that people are confident that we can achieve great results very fast great results in terms of science not not necessarily on applications yet because in Microsoft for example it's interesting because they decided to they always had a very good group on architectures and quantum operative systems and quantum compilation methods but now they decided to expand and go experimental as well their approach is long term they want to do a topological quantum computing which is a complicated mathematical theory which is allowing this type of computer to be automatically protected against the noise of the external world so it would be amazing if it ever works however as far as we know nobody has it you never even created a single qubit so everyone is really looking at what they are thinking of doing because even the announcement of a single qubit would be would be interesting so let me just hit pause on that so the the cupid is the fundamental unit of this computer it's it's the bit in contemporary absolutely yes and we've not produced one yet no in the approach that Microsoft bonds they didn't now for to give you an assess so iBM has 16 qubits operational allowing researchers to use them Google uses sixteen qubits does that mean sixteen bits just 60s not sixteen you can not even solve a Sudoku okay Google has nine but 2200 under testing 49 announced then Righetti has about eight now the wave has to thousands its oh my god the wave is so far ahead this is the wave is a Canadian company which kind of schooled to the world in the fact that you can try to build a quantum computer and they did before everyone and that's even why my group exists because Google decided to buy one of these machines not machine at one of the d-wave machine yes ok and to host it at NASA and that we operated and it is analogous to you know mainframe computers where a huge black box which everybody bits mm there's a catch there so first of all this is a huge machine it operates with the cryogenics temperature so at 13 millikelvin we're talking about less than the temperature of space it requires a vacuum it requires a helium it's a complex machine but it implements one strategy of computation which is quite unknown meaning that not so so well studied which is called quantum annealing which does not have all the requirements of the other kind of computation but at the same time it's unclear the power of it so they decided indeed to create it to experiment in the end in NASA and Google and us array decided to support the project and then after this Lockheed Martin bought the machine los alamos bought the machines so now they're doing quite well how much is one of these machines cost three price is from ten million to 15 million but it's it's actually I want to advertise in this show everyone can use this machine us array has a program a collaboration program for which 20% of the machine time can be outsourced to any researcher which is submitting a proposal it's just a five pages proposal so the the website were to find it is dot edu / quantum & / RFP request for proposal but so us are a quantum d-wave if you google it people can find and we had about 80 people 80 80 groups 80 research groups from all kind of universities around the world which are running on the machine or they they propose some of them are running some not yet but it's a very successful problem Wow and as all is all of the research that's happening on this machine about how to build more machines like it or they are people doing you know how far we in in terms of doing things with it can we add 1+1 yet with this machine you know we can we can do more than that but not much more than that no you're totally right let's say 90% of the really invested research is of fundamental nature we need to improve our understanding get different mindset and figuring out how to improve these machines themselves at the same time I must say that programming this kind of machines using this kind of machines can often lead to new ideas in classical computing because you need to change the mindset to frame the problem in a different way in order to be attacked by the quantum computers so very more there are a lot of anecdotes of you know people coming up with a quantum algorithm that is beating every possible classical algorithm that every classical algorithm was created before and then the classical guys came back say wait a second and then they improve the debate in the quantum again and so on any of any examples that come to mind yeah but it's very nerdy example there's a recent recent result on hue way away which is baptized by as the quantum alternating operator and that algorithm which is shown to basically have a particular Kaling which seemed to be better to what was before known on a problem which III Lin it's basically connected to satisfiability but but the point is like a very specific problem it seemed that this algorithm was was beating it and then the people and then the classical computer scientist came back and they improved their own algorithm so this is the latest example in general I mean it's it's fun quanta computing because there's so much room for improvement in our understanding that it's really fun for us for instance chemistry chemistry is the holy grail of quantum computing because we can simulate molecules much faster and and people were writing on the back of the envelope calculations seven years ago six seven years ago on how fast the quantum computer would solve for example the molecular ground state of interesting molecules such as ferredoxin I remember and she's a fertilizer it's interesting and people were trying to calculate it and saying okay what still takes of billions of years okay it's not very interesting because ok with the classical computer is impossible to take trillions of years one rupee would take billions ok very very worth doing it but then you know new postdocs look at the problem is saying now wait a second it's not scaling as n to the power of 11 it's scaling as n to the power of nine because I can do this mathematical tricks and then new postdocs and PhD students look at the problem say wait a second why don't I do this transformation before and I represent the problem with gosh of waves instead bla bla bla and then they put it down to a few years well it still you know not very interesting but then in a matter of like I don't know a few months other groups improved improved improved and now we're talking seconds Wow so on that particular project reason is if this molecule the people investigated you know the initial estimates based on quantum mechanics understanding were so improvable that they brought they brought down and to the power of 11 to n to the power of 3 and and make the things much more tractable and so now what does that what are the requirements implementing this algorithm that they've designed in terms of a quantum computer there's our d-wave mm bit thing get us anywhere near that or now so the wave as it was mentioning before they can attack only some specific algorithms community toriel optimization with a specific method and it's unclear what the performance of this machine can be we are investigating them early results is that it's comparable with the Intel chips on these small problems now we would like to investigate them on large bump so that we really could see the difference because now we're still talking about problems you can solve in seconds so it's not really clear the overhead how much it counts bla but but this this is interesting for the way because again having every machine they release is a little bit more powerful in the previous one also on a qualitative level they give you more knobs they give you more physical effects to experiment so that's interesting but to run the algorithm so for chemistry for for database search that I discussed our cryptography you need the computers such as the ones which are created by now by IBM and Google which right now are at this 9 cubed 16 cubed stage because they're much more difficult to build because they give you much more control and they're called the universal digital quantum computers as A+ D way which is an analog quantum annealer little bit different flavor so this is if I could paraphrase the architecture of the this D wave computer is limited in some way so that it can only you can only implement these quantum annealing algorithms and that's a sounds like a smallish subset actually we don't really know exactly the types of algorithms well yes there are uses to build the computers but much more limited in their algorithmic or encoding to give you an idea with 2000 cubits we can more or less solve problems of that you can classical problems you can encode in 50 bits roughly ok it could be 60 sometimes 400 I mean depends on the car and is that like it's not 2,000 bits or the 50 bits is this the entire state of the computer like include like a traditional computer has you know there's state in the CPU there are registers there's all of these different yes no it's let's say no no your your question is on point the end-to-end solvers will be hybrid you know system which have a classical Co process or quantum processor so after all you can decompose your problem pre-process it divide it in chunks so that the quantum computer just solves the combinatorial aspect of it and the other does something in parallel so we're way behind in figuring out what's the best way to solve you know a full problem we are experimenting right now on very specific special-purpose problems and in that case yes we can use only this little memory and but it's improving at every at every stage ok so the next generation machine probably would be much better than the previous one and how many bits do we need to get you on a universal side to be useful yeah that's a good question so it depends on the approach for example Microsoft approach it could be it could be a few hundred for the Microsoft approach but they don't even have one but for this the other approach is like the ones which are superconducting we likely need almost a million Wow almost I mean it we're at 9 yes I mean the scalability is there what is more difficult is the error correction the error correction we have very good theories but we need to get the fidelity of the operations to a certain level of procedure but there's no fundamental reason why we shouldn't be able to do that with good engineering you need to understand that the quantum engineer job is a new job I mean everyone which worked on this is physicist and physicist do not know how to you know build products yeah and so now engineers are getting are getting to this game so I believe that we will have very interesting machines soon and there's not only the big players that are startups which come out fun off of universities for example I own q how to Maryland is doing an amazing job with the ion trap and cube ion q q q thank you okay they're they're trapping atoms with lasers and manipulating them they're very very very good at this there's a University of Bristol group I mean in the UK which has some stealth operation on fully photonic computers so basically a lot of lasers so I believe in the landscape one year from now we'll be already different significantly from the landscape of today Wow wow that's fun so we haven't even gotten to AI yet like what's the intersection between your at the quantum AI lab what's the intersection between quantum computing and AI it's a very very good question so our name is might be a little misleading we have a is a quantum slash ai ai first of all of AI is not only machine learning there's a lot of methods that are considered AI since even before the hype of machine learning took off so we do a lot of optimization it's part of it yeah I do a lot of planning and scheduling which is one of my personal work where we actually have to take decisions of how robots operate on distant planets okay how to take the robots to decisions that's the problem that you need to solve by model-based algorithmics so so that's that's why there's an AI in our name because we do pay particular attention to problems where there's not good solution yet and we need expert artificial systems to solve them but the intersection of quantum computing and AI is interesting even beyond what we do in the quantum artificial intelligence laboratory the intersection could be on a technical standpoint on machine learning there are approaches which are being investigated this is a very new field so we're talking about only few years of research where you can apply some algorithms of quantum computing to gain polynomial speed ups on certain aspects for example training the new neural networks there are approaches where you can use quantum correlations to implement quantum neural networks okay I mean this requires a huge number of qubits of course but people are actually looking at this kind of and and this quantum neural networks might be very good at learning one two problems so again it's a little bit too self referential but but that's that's another approach that has been investigated and I must say there's an intersection on the other point of the arrows that there's a lot that AI can do for quantum computing okay so the payoff will come afterwards quantity will pay back in ten years or something but for now what are some of those ideas do you think even compiling a problem into a quantum computer is a big problem and you need and you need the metallization on sterilization of steroid calibrating a quantum computer is also a painting it's crazy and again laboratories are are are training neural networks and they are they are employing heavy heuristics to be able to do that so there is an opportunity on both ends to imagine what quantum computing can do for AI and how to employ AI to enable quantum computing Wow we're getting to the end of our time but is there a kind of a canonical reference or a place that people can start if they want to learn more about quantum AI or not quantum AI have a quantum computing yes there are a lot of tutorials which have been published over over the years not like build your own qubits no with a 3d printer really it's really not that innocent and accessible as it was few years okay so my suggestion is to look for the reviews and the lecture notes of the most prominent professors in the field I suggest to look at the for example John press kills lecture notes Caltech that's a good start there's a another very very skilled evangelizers of quantum computing Scott Aaronson he has a fantastic blog which is very followed where it discusses a lot of aspects of quantum computing so I'm sure that if the listener to this show are motivated they will find their way and of course please feel free to contact me my email is David a David with an e at the end dot venture le at nasa.gov ok great well David thank you so much for taking the time out to chat with us I learned a ton but there is so much more to learn about this I really enjoyed it thanks bye-bye all right everyone that's our show for today thanks so much for listening and for your continued feedback and support thanks to you this podcast finished the year as a top 40 technology podcast on Apple podcasts my producer says that one of his goals this year is to crack the top ten and to do that we need you to head over to your podcasts app rate the show hopefully we've earned five stars leave us a glowing review and share it with your friends family coworkers the barista at Starbucks or uber driver everyone every review rating and share goes a long way so thanks so much in advance as you know I love to meet Tomah listeners this week I'll be at the CES Show in Las Vegas so if you're in the area and would like to meet up ping me at at Sam Carrington on Twitter last but certainly not least for more information on de'vide or any of the topics covered in this episode head on over to twillie Icom slash talk slash in ninety three of course we'd be delighted to hear from you either via a comment on the show notes page or via twitter at at tormal AI thanks once again for listening and catch you next timehello and welcome to another episode of we'll talk the podcast where I interview interesting people doing interesting things in machine learning and artificial intelligence I'm your host Sam Charrington and we are back first things first happy new year everyone and welcome to 2018 I had a great holiday but I've definitely been itching to get back to the show as most of you know at the end of last year we held a listener giveaway to celebrate hitting one of our biggest milestones to date 1 million plays of this podcast Thanks to everyone who entered we sent out an email to entrants a few days ago so please be on the lookout for that if you haven't heard from us yet just reach out to us at team at Twin Olay I calm so that we can get you your swag next up the details for our January meet up are set next Tuesday January 16th we'll be joined by veteran twimble guest and Microsoft researcher Tim knit gabru Tim net joined us just a few weeks ago to discuss her recently released and much acclaimed paper using deep learning and Google Streetview to estimate the demographic makeup of neighborhoods across the United States and I'm excited that she'll be joining us to discuss her paper and the pipeline she used to identify 22 million cars in 50 million Google Street view images I'm anticipating a very lively discussion segment as well in which we'll be exploring your AI resolutions and predictions for 2018 for links to the paper or to register for the Meetup or just to check out previous meetups visit Twilio comm slash Meetup finally a bit about today's show I'm joined by David venture le science operations manager and quantum computing team lead for the university's Space Research Association's institute for advanced computer science and NASA Ames divi joined me backstage at the NYU future labs AI summit a while back to give me some insight into a topic that I've been curious about for some time now quantum computing we kick off our discussion with a review of the core ideas behind quantum computing including what it is how its applied and the ways it relates to computing as we know it today we then discussed the practical state of quantum computers and what their capabilities are as well as the kinds of things you can do with them and of course we explore the intersection between AI and quantum computing how quantum computing may one day accelerate machine learning and how interested listeners like you can get started down the quantum computing rabbit hole and now on to the show all right everyone I am here at the NYU skirble center backstage just as we're finishing up the NYU Future labs AI summit and I have the pleasure of being seated with David venturelli who's the science operations manager for the research institute for advanced computer science at NASA Ames daveed welcome to the podcast thank you thank you very much it's great to have you on the show and I am really looking forward to this conversation because you are doing a ton of work in an area that I know very little about but keep hearing a lot about and that is quantum computing so before we get into me peppering me with questions wanna you just tell us a little bit about your background and how you got into quantum computer self yes so I'm a physicist I'm at your ethical physicist and I did study quantum mechanics like every physicist does back when I was a PhD students in 2008 seven the words - inside the physics community about quantum computing but it's nothing but it's nothing like what we have today what I did I did study quantum information science studied nanotechnology and of course I have seen in front of me unfolding the opportunities of actually not only validating the theories on pen and paper but now experimenting with real world quantum machines so I joined the quantum artificial intelligence laboratory in 2012 as one of the founding members NASA Google and us are a University Space Research Association decided to create this group to investigate the near-term impact of quantum computing for computational problems of national interest and that was you know what I've been doing in the last five years we managed projects in collaboration with government academia and with of course the principal stakeholders the private sectors as well to understand really if in the next five to ten years there could there could be an acceleration of computing because of quantum technologies great great I guess the place to start is on quantum technologies in quantum computing like I imagine that it is something that many in the audience have heard of but don't fully understand you know not just what the big ideas are but what are the the foundational ideas upon which it's built and I'm hoping maybe you can walk us through some of that yes the idea idea comes from the 80s when one of the gods of physics richard fineman made the observation at the conference pretty casual observation that if we were able to create a computer which worked with the laws of quantum physics then simulating quantum systems would not be very difficult while today it's very very difficult to do that we need supercomputers and bla bla bla so this observation changed the world because the idea of processing informations with the fundamental laws of nature as api's as elementary building block of the algorithms really got every physicist excited some early discoveries on the fact that the mathematics of quantum mechanics is fundamentally more powerful than the mathematics we employ when we process information in standard digits Computers really got everyone excited can you give us some examples of that yes so people have discovered in 96 that you can search a database and unstructured database so we're basically all the information is not arranged by any sorting normally to search that such database you do not have any smart way to do it because you just have to randomly make queries and if you're unlucky you will find what you're looking for is the last attempt of your query and that means that the problem has a linear complexity meaning that if you have n items you need n trials to get to the to the item you want in the worst case scenario now if you had a quantum database and the ability of doing quantum queries and you exploit all these effects of superposition interference entanglement you are able to find the item you want in a square root of the number of items so this means that instead of for example 100 queries you need only 10 and this quadratic speed-up is a very interesting phenomenon which shows by itself that quantum computing is more powerful than classical computing because there's no way you can beat this without quantum mechanics but what people is really after is exponential speed-up so the opportunity to devise algorithmic strategies that are able to accelerate exponentially problems there are some examples in cryptography in in chemistry where we know how to exponentially bypass the bottlenecks of computation so that problems that could be solved today in billions of years they would be would be solved by a quantum computer in seconds but this is not the case for a lot of things that we want to solve and we are investigating and it's so difficult to investigate these algorithms without having a quantum computer that in some sense we want to build quantum computers to understand whether it was worth to build in the first place so is it fair than to say what I took away from that is that quantum computers isn't the idea of kind of going piece by piece and replacing the various components on a current system board with you know quantum versions it's not like you know shrinking down the the physics is it is a fundamental new way of thinking about computers built around the it's a new paradigm it's not incremental improvement it's really a different fundamental way to process information of course quantum effects are already very well taken care of by the silicon industry in integrated circuitry already but in but today they kind of bother you you're because you want to shrink things you want to make them more fat faster now what we're talking about is assigning a functional role to the quantum effect and actually assigning logical states and variables to quantum states and then very finely very delicately orchestrate their quantum dynamics so that we do mathematical operations on our logical assignments and this is a new paradigm it is operating on probability distributions which are weird because they're not over real numbers they're over complex numbers so there are imaginary units involved and it's very interesting mathematically and funny enough for for the listeners of this show you don't really need to know physics to become a quantum algorithm person it's actually linear algebra so as long as you accept why quantum mechanics works you can study it in an after what if you have the mathematical background you can study it fast and just you don't know why it works it doesn't matter but you can start writing algorithms because it's just you know matrix multiplications and things like things like that what are some of the you kind of rattled off some quanta primitives that you know upon which this math and these computers are built can you kind of give us a next level of detail into you know some of the most important ones of course the main paradigm which is explored today is to use binary variables so bits basically but generalized in the quantum world so the idea of using quantum bits or qubits so you do identify a physical system it could be a superconductor or a circuit or it could be an item it could be an electron it could be a fault or anything anything which is sufficiently small called protected from the external environment follows the laws of quantum physics even cats and boxes apparently so as long as you are you're able to find that then you you look at the states of the system and then you identify what is you call zero and what you call one for example it could be in more states in a quantum system yes but you choose two of them okay so you say you say okay this electron can have spin up or spin down for those that know physics or this photon can have polarization in one way or in another this superconductor can have a clockwise current or a counter clockwise current and then you look at how you can act on the system so similarly as you do in chemistry if the system is quantum you can manipulate their state like in chemistry you can do chemistry reaction but this time you need to be very controlled so if it is an 8 ohm you need to shine a laser if it is a circuit you need to apply a magnetic field you know these kind of things and and when you do that you know how your state will change the funds the fun fact is that it is totally true in quantum mechanics that you can have a state which is not 0 not 1 it's kind of a superposition of the two and is represented by this abstract vector in complex number subspace where you represent 0 and 1 at the same time now when you look at it it would be 0 or it would be 1 with some probability but if you don't look at it then it's both you can operate on it like if it is both until you look at it so the algorithm the job of the of the programmer is to figure out a way to operate on this ensemble of qubits in such a way to orchestrate the probability at the end to be the solution of the problem you want to solve but that's a complex job because you know if you have n qubits what you are really describing is a state which has an exponential number of degrees of freedom so and you need to basically describe it with true to the power of n complex numbers and there's no way you can simulate this in a machine you need a quantum computer to experiment most of the time so it sounds like one of the fundamental realities of quantum computing is this chicken-and-egg problem like how do how do we get beyond that yes you're totally right we are getting beyond that we decided I don't know the chicken or the egg what we decided but we decided that it's worth to do it and we're starting there are companies IBM Righetti Google Intel they're all building chips and they're making them available to the research community and people are trying to use them in the short term to learn about quantum mechanics to validate the theories that don't pen and paper I mean divides the last 30 years so now we are we are really using quantum computers as tools for improving our knowledge of quantum information processing yeah so IBM until some of these companies that you've mentioned had been talking about this for ever like in in kind of research mode I think the thing that really caught my attention recently was I think it was their build conference one of the microsoft's recent technical conferences there was i think it was a keynote like their main announcement such an Adela came out and had some researchers and they kind of you know brought out this I'm not even sure what it was and look like a GPU to be honest they brought out this board that was their first attempt at a quantum computer or something you tell me what exactly that was it at significance yeah I know Microsoft as well as IBM as you said like everyone had some quantum computing expertise since the beginning of time meaning since since when the field has as a name and for good reasons I mean good physicist kept an eye on the idea since the beginning and and this company is higher the very good people now what change now is that people are confident that we can achieve great results very fast great results in terms of science not not necessarily on applications yet because in Microsoft for example it's interesting because they decided to they always had a very good group on architectures and quantum operative systems and quantum compilation methods but now they decided to expand and go experimental as well their approach is long term they want to do a topological quantum computing which is a complicated mathematical theory which is allowing this type of computer to be automatically protected against the noise of the external world so it would be amazing if it ever works however as far as we know nobody has it you never even created a single qubit so everyone is really looking at what they are thinking of doing because even the announcement of a single qubit would be would be interesting so let me just hit pause on that so the the cupid is the fundamental unit of this computer it's it's the bit in contemporary absolutely yes and we've not produced one yet no in the approach that Microsoft bonds they didn't now for to give you an assess so iBM has 16 qubits operational allowing researchers to use them Google uses sixteen qubits does that mean sixteen bits just 60s not sixteen you can not even solve a Sudoku okay Google has nine but 2200 under testing 49 announced then Righetti has about eight now the wave has to thousands its oh my god the wave is so far ahead this is the wave is a Canadian company which kind of schooled to the world in the fact that you can try to build a quantum computer and they did before everyone and that's even why my group exists because Google decided to buy one of these machines not machine at one of the d-wave machine yes ok and to host it at NASA and that we operated and it is analogous to you know mainframe computers where a huge black box which everybody bits mm there's a catch there so first of all this is a huge machine it operates with the cryogenics temperature so at 13 millikelvin we're talking about less than the temperature of space it requires a vacuum it requires a helium it's a complex machine but it implements one strategy of computation which is quite unknown meaning that not so so well studied which is called quantum annealing which does not have all the requirements of the other kind of computation but at the same time it's unclear the power of it so they decided indeed to create it to experiment in the end in NASA and Google and us array decided to support the project and then after this Lockheed Martin bought the machine los alamos bought the machines so now they're doing quite well how much is one of these machines cost three price is from ten million to 15 million but it's it's actually I want to advertise in this show everyone can use this machine us array has a program a collaboration program for which 20% of the machine time can be outsourced to any researcher which is submitting a proposal it's just a five pages proposal so the the website were to find it is dot edu / quantum & / RFP request for proposal but so us are a quantum d-wave if you google it people can find and we had about 80 people 80 80 groups 80 research groups from all kind of universities around the world which are running on the machine or they they propose some of them are running some not yet but it's a very successful problem Wow and as all is all of the research that's happening on this machine about how to build more machines like it or they are people doing you know how far we in in terms of doing things with it can we add 1+1 yet with this machine you know we can we can do more than that but not much more than that no you're totally right let's say 90% of the really invested research is of fundamental nature we need to improve our understanding get different mindset and figuring out how to improve these machines themselves at the same time I must say that programming this kind of machines using this kind of machines can often lead to new ideas in classical computing because you need to change the mindset to frame the problem in a different way in order to be attacked by the quantum computers so very more there are a lot of anecdotes of you know people coming up with a quantum algorithm that is beating every possible classical algorithm that every classical algorithm was created before and then the classical guys came back say wait a second and then they improve the debate in the quantum again and so on any of any examples that come to mind yeah but it's very nerdy example there's a recent recent result on hue way away which is baptized by as the quantum alternating operator and that algorithm which is shown to basically have a particular Kaling which seemed to be better to what was before known on a problem which III Lin it's basically connected to satisfiability but but the point is like a very specific problem it seemed that this algorithm was was beating it and then the people and then the classical computer scientist came back and they improved their own algorithm so this is the latest example in general I mean it's it's fun quanta computing because there's so much room for improvement in our understanding that it's really fun for us for instance chemistry chemistry is the holy grail of quantum computing because we can simulate molecules much faster and and people were writing on the back of the envelope calculations seven years ago six seven years ago on how fast the quantum computer would solve for example the molecular ground state of interesting molecules such as ferredoxin I remember and she's a fertilizer it's interesting and people were trying to calculate it and saying okay what still takes of billions of years okay it's not very interesting because ok with the classical computer is impossible to take trillions of years one rupee would take billions ok very very worth doing it but then you know new postdocs look at the problem is saying now wait a second it's not scaling as n to the power of 11 it's scaling as n to the power of nine because I can do this mathematical tricks and then new postdocs and PhD students look at the problem say wait a second why don't I do this transformation before and I represent the problem with gosh of waves instead bla bla bla and then they put it down to a few years well it still you know not very interesting but then in a matter of like I don't know a few months other groups improved improved improved and now we're talking seconds Wow so on that particular project reason is if this molecule the people investigated you know the initial estimates based on quantum mechanics understanding were so improvable that they brought they brought down and to the power of 11 to n to the power of 3 and and make the things much more tractable and so now what does that what are the requirements implementing this algorithm that they've designed in terms of a quantum computer there's our d-wave mm bit thing get us anywhere near that or now so the wave as it was mentioning before they can attack only some specific algorithms community toriel optimization with a specific method and it's unclear what the performance of this machine can be we are investigating them early results is that it's comparable with the Intel chips on these small problems now we would like to investigate them on large bump so that we really could see the difference because now we're still talking about problems you can solve in seconds so it's not really clear the overhead how much it counts bla but but this this is interesting for the way because again having every machine they release is a little bit more powerful in the previous one also on a qualitative level they give you more knobs they give you more physical effects to experiment so that's interesting but to run the algorithm so for chemistry for for database search that I discussed our cryptography you need the computers such as the ones which are created by now by IBM and Google which right now are at this 9 cubed 16 cubed stage because they're much more difficult to build because they give you much more control and they're called the universal digital quantum computers as A+ D way which is an analog quantum annealer little bit different flavor so this is if I could paraphrase the architecture of the this D wave computer is limited in some way so that it can only you can only implement these quantum annealing algorithms and that's a sounds like a smallish subset actually we don't really know exactly the types of algorithms well yes there are uses to build the computers but much more limited in their algorithmic or encoding to give you an idea with 2000 cubits we can more or less solve problems of that you can classical problems you can encode in 50 bits roughly ok it could be 60 sometimes 400 I mean depends on the car and is that like it's not 2,000 bits or the 50 bits is this the entire state of the computer like include like a traditional computer has you know there's state in the CPU there are registers there's all of these different yes no it's let's say no no your your question is on point the end-to-end solvers will be hybrid you know system which have a classical Co process or quantum processor so after all you can decompose your problem pre-process it divide it in chunks so that the quantum computer just solves the combinatorial aspect of it and the other does something in parallel so we're way behind in figuring out what's the best way to solve you know a full problem we are experimenting right now on very specific special-purpose problems and in that case yes we can use only this little memory and but it's improving at every at every stage ok so the next generation machine probably would be much better than the previous one and how many bits do we need to get you on a universal side to be useful yeah that's a good question so it depends on the approach for example Microsoft approach it could be it could be a few hundred for the Microsoft approach but they don't even have one but for this the other approach is like the ones which are superconducting we likely need almost a million Wow almost I mean it we're at 9 yes I mean the scalability is there what is more difficult is the error correction the error correction we have very good theories but we need to get the fidelity of the operations to a certain level of procedure but there's no fundamental reason why we shouldn't be able to do that with good engineering you need to understand that the quantum engineer job is a new job I mean everyone which worked on this is physicist and physicist do not know how to you know build products yeah and so now engineers are getting are getting to this game so I believe that we will have very interesting machines soon and there's not only the big players that are startups which come out fun off of universities for example I own q how to Maryland is doing an amazing job with the ion trap and cube ion q q q thank you okay they're they're trapping atoms with lasers and manipulating them they're very very very good at this there's a University of Bristol group I mean in the UK which has some stealth operation on fully photonic computers so basically a lot of lasers so I believe in the landscape one year from now we'll be already different significantly from the landscape of today Wow wow that's fun so we haven't even gotten to AI yet like what's the intersection between your at the quantum AI lab what's the intersection between quantum computing and AI it's a very very good question so our name is might be a little misleading we have a is a quantum slash ai ai first of all of AI is not only machine learning there's a lot of methods that are considered AI since even before the hype of machine learning took off so we do a lot of optimization it's part of it yeah I do a lot of planning and scheduling which is one of my personal work where we actually have to take decisions of how robots operate on distant planets okay how to take the robots to decisions that's the problem that you need to solve by model-based algorithmics so so that's that's why there's an AI in our name because we do pay particular attention to problems where there's not good solution yet and we need expert artificial systems to solve them but the intersection of quantum computing and AI is interesting even beyond what we do in the quantum artificial intelligence laboratory the intersection could be on a technical standpoint on machine learning there are approaches which are being investigated this is a very new field so we're talking about only few years of research where you can apply some algorithms of quantum computing to gain polynomial speed ups on certain aspects for example training the new neural networks there are approaches where you can use quantum correlations to implement quantum neural networks okay I mean this requires a huge number of qubits of course but people are actually looking at this kind of and and this quantum neural networks might be very good at learning one two problems so again it's a little bit too self referential but but that's that's another approach that has been investigated and I must say there's an intersection on the other point of the arrows that there's a lot that AI can do for quantum computing okay so the payoff will come afterwards quantity will pay back in ten years or something but for now what are some of those ideas do you think even compiling a problem into a quantum computer is a big problem and you need and you need the metallization on sterilization of steroid calibrating a quantum computer is also a painting it's crazy and again laboratories are are are training neural networks and they are they are employing heavy heuristics to be able to do that so there is an opportunity on both ends to imagine what quantum computing can do for AI and how to employ AI to enable quantum computing Wow we're getting to the end of our time but is there a kind of a canonical reference or a place that people can start if they want to learn more about quantum AI or not quantum AI have a quantum computing yes there are a lot of tutorials which have been published over over the years not like build your own qubits no with a 3d printer really it's really not that innocent and accessible as it was few years okay so my suggestion is to look for the reviews and the lecture notes of the most prominent professors in the field I suggest to look at the for example John press kills lecture notes Caltech that's a good start there's a another very very skilled evangelizers of quantum computing Scott Aaronson he has a fantastic blog which is very followed where it discusses a lot of aspects of quantum computing so I'm sure that if the listener to this show are motivated they will find their way and of course please feel free to contact me my email is David a David with an e at the end dot venture le at nasa.gov ok great well David thank you so much for taking the time out to chat with us I learned a ton but there is so much more to learn about this I really enjoyed it thanks bye-bye all right everyone that's our show for today thanks so much for listening and for your continued feedback and support thanks to you this podcast finished the year as a top 40 technology podcast on Apple podcasts my producer says that one of his goals this year is to crack the top ten and to do that we need you to head over to your podcasts app rate the show hopefully we've earned five stars leave us a glowing review and share it with your friends family coworkers the barista at Starbucks or uber driver everyone every review rating and share goes a long way so thanks so much in advance as you know I love to meet Tomah listeners this week I'll be at the CES Show in Las Vegas so if you're in the area and would like to meet up ping me at at Sam Carrington on Twitter last but certainly not least for more information on de'vide or any of the topics covered in this episode head on over to twillie Icom slash talk slash in ninety three of course we'd be delighted to hear from you either via a comment on the show notes page or via twitter at at tormal AI thanks once again for listening and catch you next time\n"