Tech on Fire: Michael Azoff, Analyst, GigaOm

Today we have the fourth in our series of podcasts for Tech Trailblazers called the #TechonFire interviews, in conjunction with GigaOm. This time Michael Azoff gives an insightful explanation as to the difference between the terms artificial intelligence, AI, and machine learning, ML.

Chief Trailblazer Rose Ross asks him all about the last 10 years of AI and, more excitingly, where things will go in the future. Topics covered include how quantum computers could change the encryption landscape and how we will see new hardware accelerate the ability to process data whilst still secure using fully homomorphic encryption.

Michael and Rose also talk about AI in the home and the conundrum of how you never seem to have enough remote controls, whilst also often having too many. Watch the full podcast here:


Also available on:

Interview transcript

Rose Ross: Well, hello, everybody, and welcome to the Tech on Fire podcast and video cast. My name is Rose Ross, and I’m the founder and the Chief Trailblazer at the Tech Trailblazers, which is an awards for tech startups. And I’m delighted that we are doing a podcast series with our friends at GigaOm, and talking today to Michael Azoff, who is very much in the AI space. So, I am delighted to welcome you, Michael, to Tech on Fire. Hello, how you doing?

Michael Azoff: I’m very well, thank you Rose. It’s a pleasure to be here.

Rose Ross: Fantastic. It is great to have you here, and AI is a big topic. We introduced the category to the Awards probably about four or five years ago, and it’s grown in popularity. And I know that’s not going to be a surprise to you, Michael. But to people who don’t know you already, it’d be great to get a little bit of a background on you and what your beat is, from a GigaOm perspective, and the kinds of things that you’re going to be keen to chat about.

Michael Azoff: Sure, sure. So, I go back many, many years, as you can probably tell from my lack of hair. Well, I got interested in AI, actually, after that winter in AI that people talked about in ‘86, when backpropagation was a way of training the deep layers in neural networks. I know this is getting very technical, but up to that point, neural networks, which was a branch of AI, were very simple, and that breakthrough allowed some very deep and complex type of neural networks to be played with. So, that drew me into it. I was doing research in solid-state electronics, so I had that background in academia. I got drawn into neural nets, and I spent quite a bit of time working with neural nets and using them for forecasting, and I wrote a book on the topic.

So, that was the era that I was deep into that. And then I went and did other things, and then in the last 18 years, I’ve been working as an analyst at Informer, and then more currently at the moment at GigaOm. And when I saw that deep learning had erupted with the use of accelerators at Nvidia’s GTC conference, that really sparked my interest again, and that drew me to covering AI as an analyst. So, that spanned a lot of years there! But AI has been a thread throughout my interests. And I do other things as well, but let’s talk about AI.

Rose Ross: Yeah, lets talk about AI because my burning question, and I said I only had one question for you. I may have been lying, but this is going to be my first opening question. For me, I found that people tend to use AI, machine learning, deep learning, quite interchangeably. You can see them appearing, slash, slash, slash; can you just give a layman’s guide to what is AI, what is machine learning, and what is deep learning? Obviously for me, I understand them all. But I’m sure you would explain it a lot better than I would.

Michael Azoff: And I think it’s a good question. Because when whenever you see coverage of AI, people talk about AI and ML, and you wonder why are they using two acronyms, and what does it all mean, and how are they different, and so-on. So, that’s a good question. What I like to do is what I call level setting and explain what I mean, when I talk about AI, because you can say AI to somebody and they’ll have an idea in their head, but is it the same idea that I have, and somebody else has? So, let’s level set, what do I mean by AI?

It’s a very big topic, it’s a big research topic. It really took off from the mid-40s, and there were two main schools of thought, there was Marvin Minsky at MIT with symbolic AI. And then there was the connectionist school of thought, and that’s where neural networks fit in. Using data to drive designs that are inspired by the neural networks in the brain. So, you have two very different camps, and for many, many years, as I mentioned, backpropagation came in ’86, but prior to that, the ideas were very good in the connectionist world with the neural networks, but there wasn’t really a lot of progress and then backpropagation changed a lot of that, and we’re still… when we look at deep learning today, we still see the repercussions of backpropagation of training very deep and large complex networks. So, AI is all of that.

Machine learning is a branch within AI, and the symbolic research is another branch. And then neural networks are a part of machine learning. So, machine learning is all about using data, having designs that feed on data, and come up with some intelligence, such as pattern recognition, and so-on. They all fit within each other. So, you’ve got AI, you’ve got machine learning as a branch within that, you’ve got neural networks as a branch within machine learning, and then you’ve got deep learning as a branch within neural networks.

So, the reason why people talk about AI and ML, is because especially with deep learning being the foremost ML technique of today, is that there’s a lot of AI that’s quite interesting and not neural network based, so Bayesian networks. And there’s a lot of amazing research being done at Numenta, a non-profit research organisation that’s doing some really interesting things. Jeff Hawkins has recently published his book called ‘A Thousand Brains’, and they’re doing a lot of research on sparse networks.

Rose Ross: Just quickly, whats a sparse network?

Michael Azoff: Well, a sparse network, a sparse versus a dense neural net. So, with a dense neural net, you’ve got a lot of neurons, everything’s connected to everything else. It’s a very, very dense network. With a sparse network, you may have thousands of neurons, but only a few of them are actually lit up. And that’s actually how the brain works, the brain is very sparse, that when you get a signal coming in, maybe it’s a visual response, the eyes seeing something, only a few neurons fire at any given moment. And so the brain works in a very sparse manner. And so that research is looking at models that are sparse and have some amazing properties.

The reason I mentioned all of this is that AI isn’t just about deep learning and deep neural networks. Amazing though the research, especially from DeepMind at Google. They’re doing some great things. But there’s a lot of other things, and I think also we need to look at the difference between AI research, especially as it’s done in academia, and what is coming out of the research labs and going into industry, because most of what we’re focused on today is that AI that is being used in amazing things in applications, whether it’s vision, audio, Amazon Alexa, that sort of thing. And this is using this technology but is it going to get us to the point where two people talk about general AI, intelligence in an artificial form that is equivalent to a human being.

And I would say, most people in academia will say that deep learning is great, but it’s not going to get us to general AI. And so that’s why there is a lot of interest in other methods and other pursuits, such as the work going on at Numenta. So, it’s all about what you’re trying to achieve, and what is the best type of technology to achieve that end. So, there’s a lot of great applications, for example, if you’re trying to look at fault finding in a factory, looking at hearing the vibrations, and detecting something different. If you’re seismologists, you’re looking at patterns trying to find interesting things in the data. If you’re navigating a drone autonomously, or a train or a ferry between two ports, or even autonomous driving, as a lot of researchers, as you’re aware, are trying to get to that point.

So, this technology has a lot of promise, can do a lot of things, but I think there’s more to it, is what I’m trying to say, when it comes to actually achieving the research aim, which is to eventually get to that holy grail where we can emulate the intelligence of a human being. And I’m not defining what that means, because that in itself is quite a complex question!

Yes, so that’s why you see AI and ML because AI is more than just ML, and there are techniques in AI that’s not necessarily machine learning. So, hope that helps.

Rose Ross: That’s cool. Yeah that’s really helpful, thank you. So, one is a branch of another, and then the other is deeper again. That makes perfect sense, but I think it also helps to clarify from a lay person’s perspective, because as you say AI is such a huge element. And I mean as soon as you say AI, I go all iRobot and Isaac Asimov’s Laws of Robotics and all that. And by the way, you have a great name. If you ever want to go into science fiction, you’re almost halfway there!

Yes, so from that perspective, I think that’s really useful to sort of set the scene for what’s going on. Obviously, AI has quite a long history, both in a science fiction ‘what could happen’ way – and what actually has been done, neural networks, all these types of things. And when you’re talking about the sparse networks, well that makes sense, the brain would be like that, because you imagine if you’re powering on everything, I mean, our heads would be really hot and probably explode. And some people say that my head sometimes is a bit like that anyway.

So, from that perspective, its kind of interesting, isn’t it, because you just divert, you put your lights on when there’s something to do, and the rest of the time just relax a little bit.

Michael Azoff: Yes, so one of the things that people find is the way the neurons are connected in the brain, is something called the small world model. So, there is this game we play, is it the Kevin Bacon game? It’s how many steps to Kevin Bacon, or how many…

Rose Ross: Ive never played this, where have I been all my life!

Michael Azoff: Well, how many steps does it take to get from yourself to a US President, you know, these sort of games. And surprisingly, there are quite a small number of steps, like six. So, this is the small world model, and what’s interesting…

Rose Ross: Is this degrees of separation?

Michael Azoff: Yes. So, when you’re looking at how the neurons are connected in the brain, it turns out that they emulate this small world model. So, you have – near neighbours are quite clustered, but then you’ve got a hub neuron that has a very long connection to another hub. So, you have these clusters, these hubs, and you have a single neuron that will connect to the hub of another cluster. And that is a very small world type of model. So, we see that kind of pattern happening in the brain. These ideas are sort of feeding into sparse models.

And then you also have a spiking neural net. This is something that’s in only the last couple of years, we’re seeing startups produce hardware accelerators that are using spiking neural nets. So ,most neural networks that people work with have neurons that have continuous values. But in the brain, when an event happens it triggers a spark in a neuron and that that spike travels along the axon, and travels across synapses from one neuron to another. And that’s such a travelling spike through time. And so spiking neural networks emulate that type of characteristic.

The spiking neural networks have been around when I got interested, I mentioned ’86, they were around in ‘86. But it’s taken a long time for people to actually do something useful with them. And all of a sudden in last two years, we’re seeing a whole burst of startups coming on the scene and producing chips that are based on spiking neural nets. So it’s really exciting, that they’ve finally cracked it and are able to train neural networks with these spikes. So, that’s fascinating.

They tend to be used for edge computing where it’s very low power, and there’s analogue models, so they’re not just digital. So, when we talk about neural networks, we’re really simulating an idea, an algorithm on a computer. And we need accelerators like GPUs and FPGAs to enable these systems to compute and training in a reasonable amount of time. But they’re really just simulations, they’re simulating the idea of a neural net, that’s what the algorithm is.

But analogue models, they actually embody the neural network physically, and you have spiking neural nets, with analogue models. So, you have all these different combinations, you can have spiking neural nets on a digital simulation, or you can do it analogue. So, you’ve got all these combinations, and all these startups are trying out these different ideas.

So, there’s a lot of burst of creativity going on at the moment. It’s something we’ve been writing about a GigaOm, I’m just finishing a report on edge AI accelerators that is looking at these spiking neural nets, they’re called neuromorphic models, so we’re covering that space. And as I mentioned, edge AI, where you’re doing inference mode, so inference mode is when you’re getting a result, you’re using the neural network in production, you’re feeding in data, and then you want to see what results you get after you’ve done all the training.

So, neuromorphic is very hot for that, and it’s very low power, and it’s often very tightly connected with the sensor. And then you’ve got other chips that do heavy duty work. So, sort of Alexa, where you’re looking for a command word that will be done by, for example, a neuromorphic chip would do that. Very low power, very small chip, and it just does that function, looking for a trigger alert, a command word. And then you’ve got the bigger chips that do more heavy processing.

So yeah, I think when we talk about AI today in practice, with the technology that we see around us, like being able to talk to an Alexa or Google Home or one of those devices, or eventually we’re going to be talking to our washing machines!

Rose Ross: I already have a smart dishwasher, by the way.

Michael Azoff: Oh, that’s good.

Rose Ross: I haven’t worked out how to do it! Let’s face it, I’ve got a teenage son. I can’t even get him to put the dishes in the dishwasher. Now that’s the revolution, is when we can actually get something like a little robot that will pick everything up. I have a fun idea, because you kept saying Alexa and I was waiting for her to pop up because

Michael Azoff: She’s not in my room, she’s in the house.

Rose Ross: Well mine is.

Michael Azoff: Oh, and she hasn’t woken up?

Rose Ross: Oh here we go. I thought we could ask her what she thinks artificial intelligence is. That’d be fun, wouldnt it.

Alexa: ‘According to term paper, artificial intelligence, AI, is the branch of computer science concerned with making computers behave like humans’.

Michael Azoff: Well, that’s one view, I wouldn’t say that’s definitive.

Rose Ross: Oh, controversy, controversy! AI specialist disagrees with Alexa on what AI is! I feel theres a story there! I feel there could be acould we have a face off between human and Alexa? But this is the problem isnt it because she can only pick up information

Michael Azoff: That someone’s written.

Rose Ross: … Thats already out there. So, she doesnt hold your own term. She is just searching.

Michael Azoff: Yeah. I mean, researchers talk about behaving like a human. So, when you want to be human-like, do you want to be truly human-like, or do you just want to be rational? Do you want to think like a human or do you want to be able to act like a human? So, these are different dimensions to what we might expect out of an artificial intelligence. And they all play into, perhaps, they’re all part of ultimately we want all of those things.

I don’t know, do we want an artificial machine to be so human that it’s irrational? I don’t know.

Rose Ross: Well, it’s fine because I’m happy to be the role model for that one. Follow me, Ill show irrationality and impulse, here we go. But, I think it’s an interesting thing; I mean, I’m already now, when I go to my mum’s and I work there, I kind of turn to Alexa to go, Alexa, turn the light on, or Alexa, put the radio on, or whatever it is – she’s now gonna ask me, what do you want me to do, because you keep asking me questions too quick to react, how quickly we get used to them being around.

Michael Azoff: Yes.

Rose Ross: And it wont be long before I’ll be quite happy when there’s a little robot doing my dusting.

Michael Azoff: Yes, yes. I’m sure we’ll get to that point. We have vacuum robots. And I think that’s quite a successful little market of its own, and I’m sure we’ll see a lot more of this technology. One of the difficulties for example, with having a robot, I think, Elon Musk – was it Elon Musk, or Apple – were talking about creating a robot for the home. But, one of the difficulties is being safe around the robot because obviously…

Rose Ross: Well, I’d trip over it all the time.

Michael Azoff: They would need to be soft and flexible if you do collide with them, and I think that’s something that the researchers are working on. So, I’m sure at some point, we’ll see more of these things in our home, yeah.

Rose Ross: For sure, for sure. Well, it’ll be interesting. I mean they’re infiltrating already, aren’t they? Got the lights on it, I get all my music through it, I haven’t connected it to the telly yet. But you know, it’s not it’s not gonna be long before that will be voice commands.

Michael Azoff: Yeah, we’ll have more connected technology for sure. At the moment, I have about five different controls for my TV, which is quite ridiculous.

Rose Ross: Well, thats showing off now, Ive only got two.

Michael Azoff: I’m not showing off. It’s a pain. I want to have one! I want to one intelligent control.

Rose Ross: What on earth is going on with your telly that youve got five controls?

Michael Azoff: Erm, I have a Roku box. I have a DVD Blu-ray player; I’ve got a BT recorder. All of these boxes have different controllers.

Rose Ross: All mod cons.

Michael Azoff: Yeah, and I really just want one to control all of it. And then the television has its controls, so not forgetting that.

Rose Ross: Well youd be better than me, I was away for like a night and my son managed to lose the remote control for everything. It was an absolute nightmare. Theres nothing worse, because tellys today you cant actually do anything manually with them, you can just about turn them on and thats it! So, I bought some cheap ones.

Michael Azoff: We’ve got all this great technology, but we want simplification. So, nobody in the house can work out how to get my hi-fi system working. This is where technology is too complex. So, somebody said if you need to read a manual, it’s technology. When it’s just part of the background and you use it with ease, then it’s part of life. And that’s what technology needs to migrate to. If you need to read a manual, then it’s no good!

Rose Ross: The only reason you read the manual, and obviously everybody going to go, she’s definitely not a technologist, is when I can’t work something out myself. Or Ive broken it!

Michael Azoff: Yes. I think that’s a failure, somebody has designed a wonderful box that does some wonderful things, but they’ve done it to the way they want to design it, not to the way that we want to interact with it. And I think there’s a big disconnect between what is made available to us and how we really want to make use of those objects. We want ease of use, and yeah. So, I think there’s a lot more to be done, I think for technology to be done right. And it applies to AI I’m sure. AI may help actually.

Rose Ross: Well, that’s what I was thinking, that’s sort of like you make up for the lack of human intelligence with some artificial intelligence.

Michael Azoff: Well, I don’t think it’s lack of human intelligence. I think it’s lack of understanding by designers, as to actually how to make our life easier, instead of more complicated. And I think that’s part of the problem.

Rose Ross: I certainly would like it so that it was very simple for my mum to use, were coming up against things which are just partly to do with her age and her ability to see things, and obviously technology, I mean she has a mobile phone which she can just about work out how to phone somebody on it. But I think thats increasingly due to eyesight rather than intelligence. But voice command stuff for people who can’t see very well, or don’t understand the technology, if you make it simple

Michael Azoff: Yes, and I saw a device advertised that you could scan on a page, and it will read it to you. So, people who are just learning the language… that would be a great help. And, yeah we see more and more useful things around us.

Rose Ross: Well, we hope so. But then I did solve my TV problem, I bought two, maybe three spare remote controls on the basis that

Michael Azoff: You’re going to lose one or two!

Rose Ross: if you lose one your life does not go on hold, because obviously this is lockdown. I mean this is serious stuff, you can’t go out and have a social life, you are relying on TV-based entertainment, unfortunately. But anyway.

Well thats good, I mean, we’ll see the future moving in that direction. So, you’ve talked a little bit about startups, and obviously startups is something that we are all very passionate about. And obviously GigaOm is very much looking at the innovative space of a technology, so startups are obviously going to be an important part of that. And you talked about the ones who are doing chip acceleration, which is being utilised in edge computing, and it is something that we’ve seen. I’ve actually talked to some of your colleagues, in previous broadcasts, is that we are starting to see AI become a big part of, you know, things like cyber security, things that can be automated, things where you can maybe take a human out of part of the solution and give it some intelligence, and then free up the human beings to do stuff once that first level, what I call almost first touch support, or first touch part of a product, ability.

Michael Azoff: Yes, so there’s some very fascinating technology. In fact, the call I had just before yourself Rose, was with a very interesting company, using photons to accelerate something called the Fourier transform. But the reason why it’s fascinating…

Rose Ross: Im glad weve kept with a really simple topic for me there, Michael.

Michael Azoff: But I the reason I want to mention this, you have secure data in storage, right? You can send encrypted data as a communication, WhatsApp for example, as you know, encrypts data. But what happens when you actually want to do things with the data; you have to decrypt it to do the things you want to do. And then when you’ve got a result, you then you encrypted and then send it. But while it’s decrypted, it’s vulnerable, people can snoop and look at the data. And it’s not secure. So, there is something called fully homomorphic encryption, which allows you to work with encrypted data.

And, for example, you can multiply two bits of encrypted data and get a result that’s meaningful, but still encrypted. it’s completely hidden, then that’s what really homomorphic encryption does. But the trouble is the breakthrough around in that field was around 2009. It’s relatively recent, and there’s been an explosion of work, taking this technology forward. But, the biggest problem with fully homomorphic encryption, FHE, is that it’s very slow to do those computations when it’s encrypted. And this is where accelerators come in and make it possible to do this calculation in real time. This company I was speaking to, Optalsys, they’re UK based, actually. And there’s another company called Cornami in the US. Suddenly, you’ve now got these chip companies that are able to accelerate this technology. This is something that’s going to happen in the near future, where you’ll be able to secure data not just in storage and in flight, but alsp during compute, completely secure all the time. And that’s, that’s gonna be a massive, massive technology.

Rose Ross: That’s pretty crazy.

Michael Azoff: It is.

Rose Ross: Storage and security are areas that I’m very familiar with. She says, after pretending she’s totally technically illiterate, but um, and you know, we’ve done a lot around encryption of tapes, encryption in transit, you know, at rest. Dealing with people who were working with recovering data that had been at the bottom of the ocean on tapes, you know, since World War Two. All sorts of interesting stuff like that. Fascinating. But the thing is, obviously, those are the two states that encryptions happen, at rest and in transit. That’s what people are worried about, you know, obviously, in transit was the first one to worry about, with the ability to intercept transmissions. And but during computers, it is bonkers.

Michael Azoff: But, part of the problem is that the thinking with security now is that you’d have to assume the bad guys are in your network. That’s the level of thinking today. You need to do the right things on the assumption that the bad guys are in your system, and how do you protect. So, this kind of technology, working with encrypted data, and being able to work with it without having to decrypt it, it’s a massive breakthrough. There are some other technologies that are adjacent to this, there’s something called confidential computing. This is the chip that companies like Intel and AMD and others are working on where they have Trusted Execution environments actually on the chip.

And they have these enclaves where you can send your confidential data into these enclaves, decrypt it, work with it, and then it comes out of the enclave encrypted.

But it’s not quite as fully secure as FHE. it’s horses for courses, you choose the right type of technology for your needs. But for the holy grail of absolute security, FHE is really there. And, and we’re getting nearer to that point. We’re a few years away. These companies that I mentioned, are just in the process of building out these chips, and it’ll be a few years before they’ll be out in the market. But, you know, you heard it here. In a few years time. You heard it here first.

Rose Ross: Well,l that’s what we’re talking about. We talked before we started recording about the context of this, that Tech on Fire was looking back over the last say, 10 years, because the Awards are 10 years old. And then we thought if we can look back 10, can we look forward 10? And that’s exactly what you’re saying, If they’re alive and kicking, and I guess even if they’ve got prototype, they can come in and put their hat in the ring for the award. So, don’t be shy guys there with this incredible encryption, you know, with the whatnot, don’t be shy, put it put out there. But there are lots of these types of things that are happening in security right now, that are being used right now within sort of automation and such like. So, very exciting times.

Michael Azoff: Yes, yes. Yes, this is definitely a ‘watch this space’, this technology, secure in compute.

Rose Ross: It’s some new buzzwords there for me to hold on to there. And also a fight with Alexa who decided that she wanted to get involved in the conversation. I think she’s taken umbrage that you didn’t like her definition of AI. And she’s now going to be feisty and has decided to get in on the act, somewhere along the line.

But I mean, it’s all out there. And the other thing I was thinking about, because this has been a bit of a bit of a sort of a futuristic topic and there’s been talks about encryption and I can’t see how there’s not an element of AI in this whole area, and everybody who knows this in depth will probably go ‘of course it is’ or ‘of course it isn’t’. But there’s been a lot of talk about quantum computing.

Michael Azoff: Oh, yes.

Rose Ross: Do you see that AI is going to be at the forefront of that? Is that going to be something that will run in parallel, and they all kind of mosey along together.,.

Michael Azoff: One of the things about quantum computers is that they can do certain types of computations incredibly fast.

They’re not going to replace our classical computers. But for certain types of computation, they really are in a different realm.

So, the question is, when you look at our AI models, is there a way where a quantum computer can do something useful, whether it’s accelerating the computation or coming up with an intriguing AI model that the quantum computer can take things in new and interesting directions? I’ve seen papers being published on quantum AI, quantum computing based AI. It’s definitely that people are looking at these ideas. And at the moment, we just have emulators, and we’ve got quantum computers that can only run a few qubits. But, I think we’re all expecting, at some point, that we’ll get some quantum supremacy type of computers, quantum computers that can do things that classical computers cannot do at all.

And I think people talk about needing about 1000 qubits to do that. I think once, once we get those sorts of computers out there, I think, we’ll see more of this research. So, I expect to see AI being run on quantum computers.

Rose Ross: Well it kind of needs to. It all fits together. And the interesting thing that I was thinking about there, what you were talking about, that level of security because obviously we’re gonning to want to secure data that’s being processed by… you know, just because you want it fast. Doesn’t mean that you don’t want it securely. Right? You’re not going to compromise one for the other here?

Yes, it almost feels like it’s a bit like the safety belt, in a car. You can push more sensitive data through if you know that you’ve got this encryption during the processing part. Yes, and because quantum is so fast, a kind of a throttling back and forth. One will be really fast, and one by its very natural will slow it down. But overall, you’re going faster?

Michael Azoff: Yes, quantum computers definitely have turned the whole cryptographic community upside down, if I can put it in those terms. But, there’s good news as well, because there are technologies like FHE, that if they’re done in a certain way, so lattice-based FHE is known to be quantum proof. So, this is one of the things that DARPA in the US is running this competition for FHE accelerators, that they all need to be quantum proof. Because basically, these are going to become the next standards in cryptography, international standards. And they need to be quantum proof. If it’s not quantum proof, okay, it’s not going to be any of any use. So, there are these technologies that are understood to the best of our knowledge. So, you’re getting into the realm of NP-complete systems. And, you know, if you can show one example that can break this type of restriction, then all the others are broken as well. To the best of our knowledge, these lattice systems are quantum proof.

But you can’t guarantee that these things may change in the future. But to the best of our knowledge today, these lattice-based FHE systems will be quantum proof. A quantum computer will not be able to break it.

Rose Ross: Well, interestingly enough, the NSA kind of agrees with you. The National Security Agency. Although when I heard the headline, I had a little bit of an ‘Ohhhhh’ moment, and thanks to our friends at The Register here. And apparently, there is an FAQ from the NSA, about quantum computing and post-quantum cryptography which says, and I’m quoting from the article in The Register not from the PDF itself: “has to produce requirements today for systems that can be used for many decades in the future. With that in mind, the agency came up with some predictions, but generally speaking, is that they don’t know when or even if a quantum computer will be able to break today’s public key encryption”.

Michael Azoff: Right. So, as long as they are using some of the ‘to the best of our current knowledge’ known quantum proof, as long as you’re using those kinds of standards, you’re safe. But, some of the oldest standards will be broken. So, that’s just one of the hazards with this new technology.

Rose Ross: Well, it’ll be interesting, how it all plays out. Proof is in the pudding with these things, isn’t it? If it holds up under the rigour of attacks, then it’s all good. Let’s keep our fingers crossed on that one.

So, what else have we got to look forward to in the realms of AI over the years? You know, what are we going to see in our entrants? You know, bottle along with their entries and pop them in the virtual box, and get them looked at.

Michael Azoff: I think if we’re talking about practical AI in the world, in the real world, as opposed to the efforts in research. So, from this sort of practical point of view, I think we’ll see a lot more, we talked about gadgets and trying to make our lives simpler at home, I think we’ll see a lot more AI in the home, in the office, in the factory. There’s going to be a lot of edge AI. So, edge AI and inference mode AI is all about putting AI into everyday use.

And with 5G rollout, there’s going to be more connectivity, there’s going to be a lot more data being processed, some of it locally, some of it on the cloud. Just, you know, in the next 10 years, we’re gonna see more chips embedded in objects and devices, and they’ll be more intelligent, for sure. You’ll be able to… the human machine interaction will improve.

They’ll be doing more advanced type of analysis and computation. And hopefully, you know, it’ll be a safer world as well, because all this effort to bring AI into our transport is all about safety. We should be living in a better world as a result of this technology.

Rose Ross: Let’s hope so.

And anything else you want to add on that? On innovation, startups, and AI over the coming…

Michael Azoff: I mean, stay tuned to what we do at GigaOm because I’m covering this space. And I’m talking to a lot of startups. And there’s a lot of amazing ingenuity out there. And we love to cover it. So check out our reports.

Rose Ross: Fantastic. And I know that a couple of your colleagues have stepped in as judges. So, if you have the time, we’d love you to help judge who’s coming forward on the AI side of things. So, that’s another way for people to get your attention, get their entry in front of you. Because then you can have a look and go ‘Oh, that’s interesting, I’d like to find out a little bit more about that one’.

So, that’s good. Well, brilliant. Thank you so much. It’s been absolute pleasure, Michael. And just to wrap up, it’s the Tech on Fire podcast that we’ve been doing with the team at GigaOm, which is just with mind-blowing stuff that we’ve covered, a real smorgasbord of different technologies. Thank you very much again, Michael, for coming and joining us. And you’ve been listening to the Tech Trailblazers Tech on Fire podcast and video cast. My name is Rose Ross. And I will be coming back to you with regards to this very soon.

But in the meantime, get your entries in if you’re a startup, we want to get you in front of Michael, other guys from GigaOm, and other amazing judges, who are all keeping an eye out on what you’re up to. And also, you can find out more about that at You can follow us on Twitter @techtrailblaze and you can also find us on LinkedIn. Thank you very much.

Michael Azoff: Thanks a lot, Rose. It’s been a pleasure.