In today’s environment, the government is continuously searching for ways to modernize and take risks. Embracing what Artificial Intelligence and Machine Learning have to offer is more important now than ever. With new trends on the rise, agencies are learning what they can do with this technology. In this podcast, NVIDIA’s Adam Thompson discusses Artificial Intelligence and Machine Learning, its importance over the last year and what we can expect to see in the next year as part of Carahsoft’s community blog series focused on mission critical IT trends in 2021.
Speaker 1:
On behalf of NVIDIA and Carahsoft, we would like to welcome you to today's podcast focused around government IT trends to look out for in 2021, where NVIDIA's Adam Thompson, a senior solutions architect, will discuss the topic of artificial intelligence and machine learning, it's importance over the last year, and what we can expect to see in 2021.
Traci Rasdorf:
Well thank you again for joining me today, we're going to be talking about how artificial intelligence and machine learning have impacted the federal workspace this past year, and sort of what that will look like in 2021, since we are now full swing into January. So to start off, can you first go into what exactly you think artificial intelligence and machine learning are, and how they connect?
Adam Thompson:
Yeah, absolutely. One of the interesting things about these terms is they're sort of like those Russian nesting dolls of each other. So when we think of artificial intelligence, this is a term that actually was originated, and I think the first paper was published in the 1950s. So artificial intelligence has been around for a while, and it's really just a term to say that we're getting computers to do human-like things. And so if you think about computers doing human-like things, that human-like thing doesn't necessarily have to be all that challenging. So you can have rules based approaches to performing a human objective. For example, if you're playing a game of checkers that's a confined game in a confined space, and so you can say if this piece gets moved to this location then the AI can counter it with another move, and you can program all of this up, and so you essentially have a solution to a problem with a whole bunch of, in the computer lingo with "if" and "else" statements, if this is true, do this, else, do that.
Adam Thompson:
But what we found is that as the research progressed and the work on artificial intelligence matured, not all problems are confined like that. Sometimes we need to have better ideas of true characteristics and true features about our data, and this started the machine learning vertical. And what machine learning is doing is when you try to find specific features in your data to make decisions in a given application. And so what this really means is if you're trying to tell the difference between dogs and cats, in an image, you can train a machine learning technique to say, okay, I know in general, and remember this "in general" is going to be a defining theme of artificial intelligence and machine learning because we're trying to make generalizations about our data, but in general, dogs have longer tails than cats, and they have different types of fur, they have different shaped ears, they have different shaped noses.
Adam Thompson:
And so a machine learning algorithm would go in and they try to look at things like edge detections and colors and contours to classify and delineate between dogs and cats. And so this is fundamentally different than the if/else statements and approach that I mentioned with artificial intelligence, but remember, doing this data-driven approach with machine learning, it's still getting a computer to do human-like things, which is to tell the difference between dogs and cats in this case. So saying, I was thinking about this a little bit saying artificial intelligence and machine learning is pretty similar to saying desserts and pies. So a pie is a type of dessert and ignoring meat pies, you could say all pies are desserts, but not all desserts are pies. And so that's sort of in an encapsulated form, the difference between artificial intelligence and machine learning. So again, machine learning is a subset of AI.
Traci Rasdorf:
Yeah. That definitely makes sense. Those are really great explanations. I'm pretty new to the topic, so those visuals definitely helped me better understand it, and especially knowing that machine learning is kind of like a subset of artificial intelligence. So yeah. Thank you. Now that we have a pretty good basis for what they are, what do you think the biggest benefits of AI machine learning are?
Adam Thompson:
I think the biggest benefit is just how do we make sense of this deluge of data that we have? And this isn't necessarily constrained to the federal government or for public sector, but I was thinking about, I went to college about 10 years ago, I was talking with my wife about just what cell phones looked like 10 years ago. You would always say to your friends "do you text?" And now you can watch high definition movies on your phone. And so the amount of data that's being delivered to even just something that we all carry around in our pockets, has just exploded over the past 10 years and enlarged obviously. And so we don't see this trend decreasing, and as we have more collectors within the public sector, collectors are essentially mechanisms for getting data and understanding data, and as we put more radios in the world, we have more interconnected cell phones, we're going to need to have a sense of how do we just sift through this data to find things that we care about? How do we take our browsing behavior to influence advertisements for things that we want to buy?
Adam Thompson:
I'm redoing my house right now, and part of this redecoration is trying to figure out like, Oh my gosh, without hiring a home designer, what art do we just want to put on a wall? You know, something simple like that. And I was browsing Instagram, just doing my normal post-work sort of zoning out. And lo and behold, here's this targeted advertisement with this artwork that are these like really cool birds that I just immediately loved. So this is really something that's fascinating, it's taking things that I've liked on Instagram, it's taking all of this activity to recommend something that I would like, and within the federal government, we have other types of goals here, which is how do we use all of this data to find new threats? How do we use all of this data to identify and prosecute fraud? How do we use all of this data to better and more efficiently send out retirement checks, for example.
Adam Thompson:
And so really the end goal here is just organizing data, sorting data, and then another key point is augmenting the operator. So I think that sometimes we hear terms like artificial intelligence and machine learning and even things like deep learning. And we think that, wow, AI is going to be applied to my particular domain in my particular problem. And I don't, I can take a step back, I don't have a job anymore.
Adam Thompson:
But, but the reality is that we should view AI as a tool to help us be more effective. So it's about automating boring things. If you go into your job and you're always clicking on the same four buttons why can't you have a helper, like a computer helper to go and do that for you. So you can do something a little bit more productive. And so, again, it's automating that boringness that is really transformational, not only to the public sector, but also to the commercial sector as well.
Traci Rasdorf:
Right, yeah. I can definitely see how that kind of thinking would be beneficial, especially in today's environment. And that kind of leads me into my next question. So this year has really seen a lot of change and a lot of adjustment, obviously this past year in 2020, you kind of touched on this, but what do you think are the most significant artificial intelligence and machine learning trends that you've seen within the federal workspace this year?
Adam Thompson:
I think it's a really good question. And the healthy reality of artificial intelligence, particularly in the public sector, is it's still really new. We've seen in the commercial industry, and if you follow any of the researchers who are building AI systems, or the companies like Google and Facebook who, you know, you go to Google translate and I can speak my terrible Spanish and it translates it to English and you say, wow, that's amazing. I think as we apply this to the public sector, we've had certain wins that have been just fantastic. The United States, postal service has deployed AI algorithms to do things like package sorting, better optical character recognition, address verification and then hazardous mail identification. So when people put those little skulls and crossbones saying that, Oh, this has batteries, lithium batteries on the package, that those get automatically detected and put in a different work stream.
Adam Thompson:
So there are definable successes that we've seen deployed within the federal ecosystem. But one of the challenges is as we look to apply these fundamental machine learning and AI tools to a given problem set, is we quickly learn that there's a bigger problem with what we call dev ops. And so what dev ops is all about is, we have all this data, but is this data well structured? Do we have an overall software and people and hardware architecture to start being productive with machine learning? Do we have the ability to apply off-the-shelf models for things like image classification or object detection and run these experiments, and then keep track of these experiments so that when you're trying to figure out what model is best for your problem, if you try a given model and you finagle with the parameters a little bit and try to do some fine tuning, if your third runs is better than your fifth run, you would want the ability to go back and deploy that third run.
Adam Thompson:
And so there's a bigger software architecture that has to be applied to these problems. And the federal government is starting to put standards and processes around this. And one of the organizations that's really taken a spearhead of looking at these new technologies is the JAIC: the Joint Artificial Intelligence Center, set up by the department of defense. And so a lot of it is kind of the plumbing work of AI. And I think that is certainly one of the trends. Another trend that I feel passionately about, particularly within the public sector is AI explainability and making sure that the AI systems that we build are inclusive and diverse.
Adam Thompson:
I think that we're all familiar with some of the dangers and biases that can exist in data. And so we need to be really careful as we build these systems that the AI platforms and algorithms that we deploy are indicative of the goals and the morals that we set as a society. And so we certainly want to make sure that that's one of the highest priorities, especially as the public sector space.
Traci Rasdorf:
Yeah, I can definitely understand that. That's definitely somewhere where you want to be extra careful. So those are excellent points as sort of a followup to that, as we transition into yet another really fast paced year and ever-changing year. I know you said AI is still fairly new, so how do you see these trends kind of growing and what new trends do you see coming up in the new year?
Adam Thompson:
I think a lot of the victories are in the quiet moments. And I say that because we're going to have these big transformational programmes like the postal service, and that's fantastic, but think of all of the operations and mission who are dealing with deluges of data, who has a development team who says, Hey, we've heard about this AI thing, why don't we try to apply some of these tools and fundamentals to our given problems set? And so I think of this as using object detection with aerial imagery to try to better triage and better count objects of interest. I think of this, if we look at radio signal processing, how do we use AI to help us identify nefarious or new signals that we've never seen before? And again, this sort of replaces or augments the standard operating procedure of just someone staring at pictures or someone's staring at video and trying to do all of this stuff manually.
Adam Thompson:
You know, I always joke that as we look at AI techniques, the computers didn't get in a fight with their spouse the previous night, the computers got plenty of sleep or didn't get any sleep, I guess it's true with computers. And so there's certain advantages to having this AI as a copilot. And so I think that as we go into the future with these AI and machine learning techniques, we're going to see more small scale successes on programs of interest like this, that show transformational change, and then those... it'll set up a system where we can stand on the shoulder of giants to affect broad-scale organizational change within our community.
Traci Rasdorf:
I definitely can understand that for sure. Fantastic. I wanted to circle back to point that the events of 2020, particularly the Coronavirus have drastically changed way that businesses have to conduct their work. So I was wondering how do you feel AI and machine learning have impacted the way companies now work, who have had to make those transitions like mainly working remote, for example,
Adam Thompson:
That's a great, that's a great question. I obviously worked for NVIDIA, which is a supplier of one of the compute elements of building AI systems, which are graphics cards. And our CEO Jensen Huang said that it used to... You used to work mostly in the office and it would be a treat to work from home. And so in the future, this is going to be inverted, right? It's imagined that people will mostly work at home and it will be a treat to go into the office. But there's... We've all transitioned, and I know, I know the public sector space with, with things like security classifications and all that other stuff is particularly challenging to truly, always work from home. But a lot of us in this community who have transitioned from mostly in an office to mostly at home, we've seen some significant strengths and weaknesses of this new work relationship.
Adam Thompson:
And I think the place where artificial intelligence and machine learning could really help is, again, going back to this idea of having a little assistant there. How great would it be? I know that we all get like thousands of emails, right? And how great would it be to have a little helper that can read your email that said, it's suggested that you have a meeting at this time, so essentially you would have AI who could do some natural language processing that would go into your emails and it could look for keywords, like: "can you meet" and "times", and then you would just have like a dashboard to give a thumbs up or thumbs down on these meeting invites. I'm pretty sure, at least personally, like 80% of the emails I send is: "Hey, I'm available Tuesday at 4:00 PM."
Adam Thompson:
So having a helper there would be really great. Another space here that would be interesting is how do we make this distributed workforce feel more connected and also maintain that sort of work life balance, the relationship that we have.
Adam Thompson:
And so we can envision a situation where AI and machine learning could really target that part of the equation as well. So I think that what COVID-19 has really done to industries is almost put an adrenaline shot in the arm, that we really need to embrace these new technologies, like artificial intelligence and like machine learning and embrace them for what they are to modernize and to take some risks, because for so many companies and for so many organizations, 2020 was just, it was a kick in the pants to modernize. And so once we're on this path of modernizing the way that we view work, the way that we view IT, the way that we view building algorithms and admissions and all that other good stuff, hopefully the appetite is there to do more and to push more.
Traci Rasdorf:
Yeah. I definitely think that that appetite would be there, like you said, and I absolutely understand your example of combing through emails. I'm sure lots of people listening also can. And to your point about balancing work and home life, it's also really important to note since especially so many people are working from home right now. So you've definitely given me a lot of great information here. So to sort of wrap up my questions, we've touched on some great examples, especially with the growth of AI within the postal service. But I was wondering if you had any more examples of any sort of use cases that apply to how AI and machine learning have been helpful in the government space that maybe you haven't touched on yet?
Adam Thompson:
I think one of the interesting components that we haven't touched on is the really exciting fields of autonomy and sort of predictive maintenance. So this is sort of log parsing and cyber use cases. So again for these three pillars that I've just introduced, so like handling log information, like fiber, handling log information, like predictive maintenance, so "do I need to fix my plane yet or not?" And for autonomous systems, so "how do I steer this little car without someone at the driving wheel?" Are all really data-hungry, data-dependent types of problems. And so I think those are really good areas of future work because if there's one thing we all love about our federal government, it's the ability to produce information and logs. And we need to make sure to establish that feedback loop, that all of this information that we record and all of this information that we collect can be used to make better systems that cost less to the tax payer, and that are more effective to the folks who are owning and operating these missions.
Adam Thompson:
So I do think that these... Just to focus on one of these autonomous systems, one of the components of AI and machine learning, and I've spoken about this a lot in this podcast, is the ability and the need for AI to be a coach, but we also need to have a finger on the pulse of what our near peer adversaries are doing. And so if you have an adversary who's doing fully autonomous systems, we need to have the ability to match some of that as well. So future artificial intelligence and machine learning systems will have to keep in mind, particularly within the defense space, what our near peer competition is also doing. So we make sure that we don't fall behind because we don't want a situation like that. You know, America needs the hegemony within the new base of artificial intelligence and machine learning.
Traci Rasdorf:
Yeah. I think that's a really great point, definitely not wanting to fall behind in this area. All right, well, thank you for providing such in-depth responses throughout the podcast and especially expanding on those last couple of examples. So I just wanted to give you the opportunity to see if there's anything else that you'd like to touch on or talk about that we haven't covered in our conversation.
Adam Thompson:
I think this is a really good overview and introduction to some of the key concepts of artificial intelligence and machine learning, throughout this podcast. One of the things that I wanted to adequately and accurately portray was not only some of the benefits of AI, but also some of the things that pose the challenges, right? So this is something that if I say, Oh, I've got a ton of data, Oh, I've got a problem to solve, it's not necessarily like you can just sprinkle AI on your problem and everything will be fine, 45 minutes, that there is a process behind building algorithms, and then going back to the diversity and inclusion component that I brought up, making sure what you built is what you want to deploy, and make the decisions that you want to emphasize.
Adam Thompson:
A lot of this comes down to the biases of your data. So if you're trying to find, as so many government programs are trying to do, needles, important needles in haystacks of data.
Adam Thompson:
So say I have a haystack that has a hundred total needles in it, but only one of those pieces of hay, I guess a hundred pieces of hay, instead of needle, only one of those elements is actually something that you care about. So if you say everything in that data set is either, it's something that I don't care about then you're accurate 99% of the time, right? Because you have a hundred total examples, you have one thing you're looking for. And so this is the problem of how to buy datasets. And so we really need to make sure that we take these considerations in mind as we, as we deploy AI systems and AI algorithms there in the future. But again, I'm really excited about what the future holds. I think this is one of true technology transformations that we're going to see within our lifetime, but with great power comes great responsibility. And we want to make sure that, that as these systems get deployed, they get deployed in a safe and effective manner.
Traci Rasdorf:
That's an excellent point. Every success can come with its challenges, so I think it was great for you to bring that up, especially for any sort of future improvement that we'll have in this space. All right, well, thank you, Adam, for all of your time today, I definitely had an awesome time talking with you. Is there anything else that you had any questions on or any other points that you wanted to make?
Adam Thompson:
No, I think that's it, Tracy. I really appreciate the support and I had a great time as well.
Speaker 1:
Thanks for listening. If you would like to read our blog, highlighting the main points of today's discussion, please visit www.carahsoft.com/community and follow along for more insights into government IT from industry thought leaders. Thanks again for listening, and have a great day.