What Is Machine Learning? (S2:E50)

February 9, 2023

Today, we’re bringing back Mike Taylor, Identiv’s VP of Global Sales. Our chat sheds light on machine learning, discussing how it differs from AI, unpacking some of the myths surrounding the technology, and determining exactly why it is such a vital tool for modern businesses.

Full Transcript

Voiceover (00:01):

You're listening to Humans in Tech. Our podcast explores today's most transformative technology and the trends of tomorrow, bringing together the brightest minds in and outside of our industry. We unpack what's new in physical access, identity verification, cybersecurity, and IoT ecosystems. We reach beyond the physical world, discuss our digital transformation as a species, and dive into the emerging phygital experience. Join us on our journey as we discover just how connected the future will be and how we will fit into that picture. Your host is Leigh Dow, VP of global marketing at Identiv.

Leigh Dow (00:43):

Thanks for tuning in. Today we're talking to Mike Taylor, vice president of global sales. Thanks for being on here today for the Humans in Tech Podcast.

Mike Taylor (00:51):

Great to be here, Leigh. I always enjoy it.

Leigh Dow (00:53):

Today we're going to be discussing machine learning and how it's rapidly changing the face and pace of business as we know it. Machine learning for our audience, if they're not familiar, is a subfield of AI, artificial intelligence, and it's broadly defined as the capability of a machine to imitate intelligent human behavior. So AI systems are used to perform some pretty complex tasks in a way that's similar to how humans solve problems, yet really requires little to no human interaction. So a lot of times, people are confused. They confuse AI and machine learning. What's the difference.

Mike Taylor (01:27):

That's a great question, and I think you outlined it great. So machine learning is a subfield of AI, and really, what machine learning is, Leigh, is it gives computers the ability to learn something without being programmed to respond a specific way, so it allows the computers to gain that data and then turn around and make decisions off of it. And then AI, on the other hand, as you said, is really using a system in a complex way and using it to really solve problems. And I think we'll talk more as you get into the podcast about what it leverages and how it works.

Leigh Dow (02:09):

Well, one thing that's kind of funny is, so the difference between AI and machine learning, we've done a podcast segment on medical devices where we talked a bit about different AI and machine learning systems within the medical device industry, and we got into how in physical security, there's so many applications for both, and there's such a lack of understanding of what they are and how they work, even in the legislative bodies that are required to make decisions about how these technologies are even able to be used. So it's really good to break it down into something that, I think what you just did, which is something that everyone can understand the difference between the two.

Mike Taylor (02:55):

Absolutely. Yep. And it's funny. Hearing you talk about governance and who's going to make those definitions and who's going to manage it, I shudder a little bit, because our government is very slow on the draw when it comes to some of that stuff.

Leigh Dow (03:10):

Oh, 100%. That's what the conversation was, is how there's not deep subject matter expertise in the people who are making these decisions.

Mike Taylor (03:17):

Yeah. So it's really incumbent on these companies to manage it themselves. And again, there's great companies out there, Apple, Google, but they set the parameters, and they can be very loose guidelines. So yeah, I think it's a great subject to discuss.

Leigh Dow (03:35):

Why does machine learning require such high quality input data and models?

Mike Taylor (03:41):

Well, I've heard it said in the past, and I think it's a really good analogy. It's almost like gardening. You have to fertilize it to grow. And I think there's the old saying when it comes to whether you're on a diet, whether you're feeding a CRM, junk in, junk out. So I think it's really critical that you get the best quality data coming in, and that way you have a higher likelihood of a better outcome. There's a million algorithms out there that data scientists use.


And it's interesting, because when it comes to doing AI and machine learning, there's certain algorithms they will use, and there's actually some they will purposely omit so that they can really manage how that data comes through. And it allows them a simpler model to get a little bit better data in the door. When you start with solid data, then the output from that's going to be very clean and crisp. If you use anything that's either muddy or not really clean... I guess I just keep going back to that analogy. It's got to be extra clean. But the better data, then the better decisions the machine can make based on the data that comes in, which will give you, ultimately, much better predictions.

Leigh Dow (05:01):

Which is such a challenge, right, because who's ever worked anywhere where the data system you work with is 100% clean?

Mike Taylor (05:09):

As the leader of a sales team, if that person knows where that exists or can tell me how to make it happen, please call me. My number will be at the end of this call, because I would love to talk to you. Yeah, it is a challenge. It really is.

Leigh Dow (05:22):

Right? And so like you said, kind of the garbage in, garbage out. And I've even seen stories recently, things that are nothing connected to our industry, but different AI and machine learning experiments and things like that. I read one the other day where they were trying to use Twitter, which is why anyone would do this in the first place, but trying to use Twitter to inform a machine learning application, which automatically, the trolls jumped in and tried to skew the data that was being fed to this particular application, which of course had not good outcomes. Yeah, so that garbage in, garbage out's really important, and especially when you think about systems that are making decisions without any human interaction. So when you think about the discussion and information about machine learning and how it's supposed to make life easier for all of us, how is it really being used in real life?

Mike Taylor (06:23):

Great question. And what's funny is, it's being used quite a bit today. I just think a lot of times people are... I wouldn't say oblivious. They just don't pay attention to it. So there's things like voice search technology, image recognition, automated translation, even self-driving cars. It's being used today. I just think sometimes, people get one picture of it in their head, which is, i.e., the Hollywood version of computers are going to take over the world, and they're actually bad mouthing or talking about it. The trolls come out and say how bad it is, and then you look back and you're like, "Hold on. You were on Twitter spewing all this hate and negativity around it, but you were using voice technology to do it, which was based off of the very thing that you're slamming." So it is really already ingrained in today's world. I just think a lot of consumers don't realize what it is.

Leigh Dow (07:17):

Well, I used to work with a woman who's an anthropologist at Intel, and her whole job was to figure out, how are people going to use technology in the future? And she gives this amazing speech about how in almost every country around the world, robots are friends, except for in America. So you brought up the Hollywood. So she would always talk about how Hollywood movies, robots are always coming to kill us.

Mike Taylor (07:42):


Leigh Dow (07:42):

Right? Or the how in Space Odyssey, right? It's never some positive impact when it's depicted in an American Hollywood movie, but it does have really positive outcomes when it's used appropriately. And I also think that, like you're saying, these different capabilities that are just part of your day-to-day life, and they're designed to be non-intrusive, right? They're designed to just be something that you don't have to think about.

Mike Taylor (08:10):

No. Again, you're spot on. And everyone likes easier. Everyone likes better, bigger, faster. So as we get that, sometimes it's one thing to say, "Hey, I love being able to cook a bowl of soup in 12 to 15 seconds, but I don't want microwaves in my house." So there is a trade-off there, but at the end of the day, these things are designed to make our lives better. Are there risks? Yes. Are there things that need to be pulled back? Yes. But at the end of the day, we will certainly, as a whole society, truly benefit from this technology.

Leigh Dow (08:51):

What types of problems does machine learning solve?

Mike Taylor (08:55):

Machine learning problems, really what we're seeing, Leigh, especially in the security space, we gather a lot of data, and so really, the big question is, what are we doing with it? Can we maximize it? What are we trying to do? I mean, in short, if you look at it, machine learning problems typically involve predicting. They take previously observed outcomes using past data and say, "If this happens again, what is most likely going to be the future outcome?" I like to call it as, it's almost like the easy button for predicting the future, if you will.

Leigh Dow (09:30):

Well, I really think, too, that... I know you and I attended the last ISC West, and I wasn't at GSX, but at ISC West, I did notice that there was already a lot more messaging around AI and machine learning. And I think just as an industry, we have to be really cautious about making claims with respect to what the technology can really do today or what we're really willing to let it do just yet. And so I thought it was really interesting just seeing that it's definitely becoming more and more prevalent as a part of the offering and as a part of the industry, but I don't hear a lot of conversations about, what are the claims? Are they false claims? Which ones are real claims? What can it really do, and where is it really working?

Mike Taylor (10:18):

Yep. I think you're spot on. And within our industry, people tie machine learning and AI to analytics, and they are certainly not the same thing. However, there's that perception because the analytics came to the security industry, and they were definitely not ready for primetime. A lot of people that went in on it got burnt, and so now there's a hesitancy to use it. So I think there's a lot of people that want to talk surface. "Oh yeah, AI, machine learning, yes, yes, yes, we're in," but if you actually push them on, "What is it? How are you actually using it in your business," there's a lot of crickets still. I think if we as an industry go all in and truly work with great partners, whether it's universities, whether it's these huge tech companies, to figure out how to implement what they do on top of all the data that we gather, it could be a home run. But I just think we're at the infant stage of having those conversations.

Leigh Dow (11:26):

Well, and I sit on the SIA Facial Recognition Work Group, and I joined that work group because of my background working in legislative bodies. I just thought it would be a really interesting one to join, and that one in particular is part of the government affairs part of SIA. And the reports that I see from them and from the people who are actually doing the education and testimony on the Hill and stuff like that from SIA are really interesting because, to your point, it's like people don't really understand this technology yet. When you push them on it, on what they're really delivering, or why they're protesting it, in some instances, a lot of states are trying to legislate these things, but they don't really quite understand them or what the real uses and risks are in implementing them. So just for me, it's been a really interesting conversation to be a part of because there's so much confusion.

Mike Taylor (12:25):

Yeah. Well, and again, thank you to you and the board of SIA. That's a tough spot to be in. And any time you're in the private to public and dealing with government agencies, it's really a difficult situation, because the government's designed to make everyone happy. "I need you to like me to reelect me." And sometimes they're just going to bow to it. And unfortunately, most of the people they're bowing to are the same Twitter crowd. And I think I read somewhere, about 18% of the US population is actually on Twitter, so you're really not doing a really broad audience there.

Leigh Dow (13:06):


Mike Taylor (13:06):

And so when you were talking earlier about, "They ran analytics on Twitter," and I thought, "Okay, if this was Jeopardy, my answer would be, what is to spew hate."

Leigh Dow (13:16):


Mike Taylor (13:18):

Twitter is where people in their parents' basement go to complain. So it is a tough discussion, but the only way we truly will gain and move forward is if we have these tough discussions. And we see advances across Europe. They are a little bit more quick to adapt some of this. And from that respect, I'd love to see us kind of follow in their footsteps.

Leigh Dow (13:40):

Right. And then I think that a lot of what is interesting to me on that subcommittee is that the technology of machine learning was really, in part, developed to solve problems that require that unbiased analysis of lots of different factors and generate an outcome. But the truth is that, how do you really, truly remove that bias?

Mike Taylor (14:04):


Leigh Dow (14:04):

It's so difficult to do.

Mike Taylor (14:07):

And then the concern and the question is, can that bias somehow be fed or built into the electronic models? Is there a way that the AI will actually recognize some of that? And that's a risk you run, because if you're predicting outcomes based on past data, then maybe it's skewed, but-

Leigh Dow (14:27):

Yeah, because what is your data set, right?

Mike Taylor (14:30):

Correct. So again, junk in, junk out. It's really critical that you take really good, clean data points and use that going forward as you build your model.

Leigh Dow (14:39):

I also think that this is an area where there needs to be a lot more discussion about ethics. I haven't seen a lot of discussions about ethical data capture or the ethics around managing data or creating unbiased data or stuff like that, so I definitely would like to see SIA and other industry bodies like that explore those topics a bit more.

Mike Taylor (15:04):

I think it's critical. And I think it's really one of the things that is a bit of the wild, wild West right now still with data today. If we capture data, I mean, we are a third-party manufacturer, so we're not retaining the data, but we have the ability to retain data for our customers. What do the customers do with it? Who owns that data? And those are all really what is the meaning of life type questions.

Leigh Dow (15:31):

Right. Exactly.

Mike Taylor (15:32):

I'm certainly not smart enough to give you the correct answer on it, but I think if we don't have the discussion and if there's not bodies leading the charge to have those discussions, then we're going to fail.

Leigh Dow (15:44):

Well, even those discussions, if we only have them within the security industry, then we're not involving the people who can guide us in those.

Mike Taylor (15:51):


Leigh Dow (15:51):

Or at least, I don't mean we can't make those decisions because we're not all smart, professional people, but because you need those other voices to make decisions that really impact humanity.

Mike Taylor (16:04):

Absolutely. And again, the security industry is a subset of the technology that's out there, and in the grand scheme of life, it's a very small subset. So you're right. I mean, we can't do it just within our industry. I do think we can be an industry that rings the bell and has a big voice and tries to carry the day and get these other technologies on board, but if we look at it just in context of the security industry, we're missing.

Leigh Dow (16:33):

So what are some of the factors that you think are really driving the growing popularity of machine learning?

Mike Taylor (16:40):

I think there's probably two or three key factors. I mean, certainly, the availability of data in huge volumes is unbelievable. I mean, people do not realize the digital footprint that they leave behind. And all of that data is collated. So there's not only large volumes of it, but just the range of it, everything from if you walk up to someone and say, "Give me your password," they're not going to do it, but they'll log in on their phone to a website on a public Wi-Fi and hand out that information.


So I think there's large volumes of data with wide-ranging topics. I think computing power has never been more cost effective. You've heard stuff, but my iPhone today has as much computing technology as the first space shuttle. It's just unbelievable how affordable the access is to computing power. And then certainly, my favorite, high speed internet. God bless Al Gore. He created it. But just the access to truly high speed data, meaning you're getting data truly in real time, these are the kind of factors that make it easy to develop computer models, certainly to take that data, analyze it.


And again, as we move forward with big data, companies are looking to leverage that to, what I often say is, they want to do more with less. And the belief is, I can sort my way through with technology, and I can build models, and I can build all the stuff so that I need fewer people to actually run my business, making me more efficient, thus making me more attractive to investors and making everybody more money.

Leigh Dow (18:36):

Do you think, can machine learning predict the future?

Mike Taylor (18:41):

Hey, listen. As much as anybody, I'm a big fan of that magic eight ball, but this is certainly way beyond that. I do believe models can predict the future, but I think it's all relative. I mean, some of the relevance comes from connecting to past events. So for example, there are some machine learning models that we can use to predict stock prices in the future, but the only thing we're leveraging that is, what has that stock price done in the past? The same thing with weather. So we can see the future to a point. We only do it from looking backwards, which seems kind of counterintuitive.


But I do believe that a lot of things that are going to come, we can predict. And some of it's high level, like will there be a major earthquake on the West Coast? Yes. That sort of thing. But predicting more of when it's going to happen, I think we can get there. I think we're pretty close today. I just think one last piece of it is really setting the expectations. We live in this world where we expect, because of technology, to know everything, exact, at all times. So if someone predicts, "Hey, there's an earthquake that's going to happen, and I can guess down to the week that it's going to hit," is that a success? To me, it would be. So I think we can predict it. I just think we've got to be realistic on what our expectations are of the outcome.

Leigh Dow (20:10):

That's a lot to unpack.

Mike Taylor (20:14):

Yes, it is.

Leigh Dow (20:15):

Well, thank you so much for joining us today and giving us insight into machine learning. It's just a really fascinating topic. It's got so many legs and so many open question marks to explore.

Mike Taylor (20:26):

Yeah, I think it's certainly something we could almost run as a series and over the next couple years, just watch it evolve, see how it not only impacts our company and the security industry, but certainly society as a whole. So I look forward to it, and I certainly appreciate the time.

Leigh Dow (20:42):

If you enjoyed this podcast, please like and subscribe for me. We drop a new episode every Thursday.

Voiceover (20:47):

The problem isn't security, it's awareness. Velocity Vision is the future of visual surveillance. An intelligent video management solution that delivers real-time situational awareness in an open security platform. Integrate with your existing systems, verify your environment in one pane of glass, and increase the efficiency of your security operation. Get full control of your environment when and where you need it. Learn more at identiv.com.


Get access control anywhere, anytime, for less money out of pocket. Highly secure Freedom Cloud is a cloud-based Access Control as a Service offered through a cost-effective subscription model, allowing users to control, manage, and maintain their physical access control systems via Freedom's intuitive, always up to date, browser-based web administration. Learn more at identiv.com. Physical security, identity verification, the IoT. The hyperconnectivity of our lives will only grow more pervasive. As technology becomes more automated and experiences more augmented, it's up to us to preserve our humanity and use new tools and trends for good. The only question is, are we up for the challenge?