Hey everybody, Dr. John Shovic here back for another little talk. Today we’re going to chat about Artificial Intelligence for the absolute beginner. You don’t need an advanced level of knowledge to understand this. This is more of an informational talk just to give you a handle on what all the hype is about and what all the hype really, isn’t about. As many of you know, in addition to being the CTO of SwitchDoc Labs, I’m also a Professor in Computer Science at the University of Idaho in Coeur D’ Alene, Idaho, and my specialty is in AI and Robotics. I’m focusing on building and manufacturing robotics laboratories and applying AI techniques to various hardware in there. But that’s way beyond what we’re going to talk about today. Today we discuss, what is AI? What does it mean, and what does it mean for us going forward as a society?
Remember this is an introduction for absolute beginners. We’re going to talk about what is AI. Then we’re going to talk about some of the techniques of AI, and finally, we’re going to talk about some of the applications for Artificial Intelligence.
What Artificial Intelligence is, and it isn’t
First of all, AI isn’t one program that someone is trying to put together to emulate the entire human mind and make a person out of a computer. There are people trying to do that, but they’re a long ways off. Where AI is very successful today and very, very useful in a collection of techniques that are used to do various things that are associated with Intelligence. For example, learning. You know, learning is associated with Intelligence. But, flatworms can do learning to, but we’re talking learning that is more applicable to the human experience.
So that’s one area where there are fascinating techniques. Artificial Intelligence is a collection of methods that are used to emulate or simulate known parts of Intelligence. You know, there are different opinions out there and saying humankind Intelligence isn’t the only Intelligence in the world. That could be, maybe machines could have different kinds of Intelligence. Those are all philosophical questions and something that will be answered eventually in the fullness of time. But right now AI techniques can make people money and can also really improve human life and do particular things. Remember what I talked about is AI is a collection of techniques. So let’s get technical for just a moment. These techniques are a advanced form of statistical and mathematical models that engineers, professors, and scientists have created in an attempt to duplicate certain parts of Intelligence.
What are these techniques? Well, there’s a wide variety of them, but they typically fall into a few categories. The first thing we use with AI is Heuristics. Now, heuristic sounds like a big word, but it isn’t. Heuristics are rules of thumb we apply to gauge certain situations. Humans do this all the time. We extrapolate for specific incidents to general incidents and heuristics are part of those. Now. What is a heuristic? An example might be if it’s dark outside, the sun has gone down. Well, that’s a heuristic. It’s not always true, but it’s useful engaging whether it is daylight or night by the fact that it’s dark outside. Maybe someone’s put a shade over your window. Perhaps it’s cloudy out, but our heuristic says as a rule of thumb that will give us right answers.
There’s a whole series of programs which are roughly lumped into Expert Systems. They take rules of Heuristics, rules of thumb, and then apply them to a lot of different situations. This is actually what I did my dissertation in “Lo those many years ago.” We’ve been working on Artificial Intelligence for a long time, and you know, it’s always a technology of the future. Well, some of it’s coming true now. Some of it is useful. That’s is a heuristic. Building an expert system, taking your knowledge and condensing it down to a set of rules of thumb, that allows us to get to certain places. One place where this works well is in emulating the way a doctor diagnoses diseases. We can build expert systems that could do a pretty good job of diagnosing diseases. 20 percent are tough, and the doctors have to work hard to figure that portion out.
The other 80% can be done by an expert system. Why hasn’t that happened yet? Well, it is happening. In places like Israel, they’re doing a lot of this. In the United States, we’re behind regarding applying those. This is also true in a place like Japan. It’s not that we don’t understand the techniques, it’s more about the way the medical system works in the United States that makes adoption very, very slow. Part of that is litigation and tort law.
Oh no, I forgot to tell you my joke. This electron walks into a bar, and he sits down at the bar, and the bartender comes over and says, uh, what’ll you have? Well, I don’t want anything. I’m just really un-pleased to be here. The bartender looks at him and says, don’t be so negative.
Vector Machines
Now let’s get back to Artificial Intelligence. There’s another category of Artificial Intelligence which is called vector machines. It sounds complex, but it’s just taking large amounts of data and splitting it into different classes, and they call that vector analysis, a state vector, or vector machines. But what you’re doing is looking at a big selection of data, big data and then going in and coming up with heuristics and other statistical analytical techniques to separate it into various classes so you can determine things.
Another big part of AI which is used a lot and getting used more and more are machine learning techniques. As we talked a little while ago, learning something is a real hallmark of Intelligence. Whether you’re a flatworm or a human. Humans are good at learning something new and then extrapolating from what they already know and putting it into new situations. Machine learning follows this pattern.
What you do is take a picture showing a square and a circle, and you feed picture after picture to this statistical model while it’s built of neurons or artificial neurons or whether it’s using some other of the many statistical and mathematical techniques for doing this. You feed it a picture of a square, and you tell it this is a square. You say, yes, we’re successful. That’s a square. Put a circle in your tell it; it’s not a square. It’s a circle, and you keep feeding data in it. Training the machine, teaching the machine, teaching this piece of software. How to identify the difference between a circle and a square without telling them? Well, you got to look for four corners and what not just by feeding him true information, feeding false information, and telling them which are which this machine, the statistical algorithm, the AI or whatever you want to call it, learn from that.
And then we’ll take that and extrapolate. Now when you feed it pictures after it’s been trained. It’ll identify what a square is and what is the circle at a very high level of confidence just like a child would be once you teach them how to do that. They do a very good job. That’s a machine learning technique, and you can apply that to many, many different problems. Video problems, noise problems, data problems. They use techniques like this to identify planets orbiting other stars by looking at the Kepler space telescope data. There are a lot of different applications. Machine learning is one of the applications that in my Advanced Robotics class I’m teaching, both Undergraduate and the Graduate sections we’re actually applying machine learning to some robotics problems, but it’s kind of like teaching the robot to identify a square or circle, and we’re actually using one of our students babies building blocks as the test to do this.
So it’s rather basic, but it illustrates the technique.
Natural Language Processing
Now, another area where AI is becoming very interesting and very useful is natural language processing. That means I’m talking to you; you have a machine take what I’m saying, parse it and try to understand part of it. I use the word “understand” very loosely because these programs aren’t intelligent. They look like they’re intelligent and they can take what they’re looking at, and it’s produced usable results. But when I say understanding natural language, that’s a big deal, but taking my words, translating the text, parsing them, figure out what I’m talking about and supplying an answer that’s pretty good. You may have one of these AI systems in your house, maybe you’ve heard it Alexa or the Google Home or something like that. That’s what those guys do.
I’ll tell you from experience, Alexa is not very smart. I’ll come back to that in just a moment.
There’s a whole other category of Artificial Intelligence, and that’s decision making support. Using Markov models and all sorts of things like that to help people and machines make decisions. A decision like whether that’s a circle or that’s a square, that’s a decision. But also maybe how to navigate around a particular item. We haven’t even talked about self-driving cars yet, but we will in just a moment. So we’ve talked about what is AI. It’s a collection of techniques, and we’ve talked about what some of those techniques are, machine learning, natural language processing, things like that. So let’s talk about the applications because that’s really where this stuff is coming into its own. There are AI applications all over the place, but this isn’t generalized Artificial Intelligence like a human.
These are specific programs with specific purposes, doing specific tasks and they do those tasks well, and they’re valuable techniques. That’s one reason why I think it’s very important for any software engineer or any engineer to understand how to apply Artificial Intelligence techniques. You no longer have to be a super mathematician to use machine language algorithms. There are packages out there available in the cloud. Amazon’s cloud, Google’s cloud all over the place. You can use it. You can even get machine language programs that you can download for your Raspberry Pi. They may not work very fast, but they will work even on a Raspberry Pi or Arduino. We’ll be talking more about that in the future for sure. So what are some of the applications? Well, let’s start out with robotics, and one of the big problems with robots is having them understand their environment enough so they can deal with changes in their environment.
For example, in a production application a block is coming down on the other side of the conveyor belt, not entirely where it should be or it’s the wrong color or something like that. A robot is sensing an environment and then making a plan. This is tough. Creating an idea of what to do about that wrong color block coming down your conveyor belt. Picking it up and throwing it out, or picking it up and putting in a different box or handing it off to a human which can figure out what to do. That’s kind of like having a robotic vacuum. When it runs into something too big, it says hey, help me here. I don’t know what’s going on. Now you understand I’m anthropomorphizing a very dumb machine that’s smarter than a calculator, but is not very smart. You must be very careful when you talk about AI not to anthropomorphize so much.
Anthropomorphize means I’m talking like it’s a person. I’m making it look like a person when it’s a machine, and it’s not a very bright machine. Someday we may have bright machines, and personally, I’d be very excited about that. It would make everything a lot better in my opinion. I think the coming AI Revolution, which is already here, is going to make a big difference in the quality of our lives. It’s going to be to the plus.
We mentioned Alexa a few minutes ago. Well, you think, wow, Alexa is brilliant. I can ask her a question. She gets me the answer. Well, I’ve written some Alexa code, and I have students doing Alexa code. Plus we’ve released products that talk to Alexa on SwitchDoc Labs. For example, SmartPlantPi our plant watering system. Now you can communicate back and forth with it using Alexa using your hardware at home.
Here is one of our tutorials: https://www.switchdoc.com/2018/01/tutorial-voice-time-smartplantpi-alexa/
How about the all in one OurWeather station? You can ask what the weather is at your house. Any place you have an Alexa adequately wired up to your account. That’s neat! But when you get down and look at really what’s going on with the vast majority of these apps, Alexa is good at turning spoken words into text. But then the program kind of breaks down. It’s not very smart from that point. You have to match a particular set of texts to a specific set of actions. For example, when I asked OurWeather, “Alexa ask OurWeather temperature,” it parses OurWeather and temperature sends the word temperature to what I call a lambda function, which is just a server-less thing running on the Amazon cloud somewhere. Anyway, it goes to this program, and this program says, “Oh, I have received the intent, it’s called temperature.”
So it says OK, I’ll get the current temperature and provide that information. And then it puts it in a statement in a string that says “The outside temperature is 84.3 degrees”. Looking out the window. I wish it were 84.3 degrees. But, hey, the snow is falling. What can I say? So it takes that text., sends it back to Alexa, and Alexa converts that text to speech and puts it on your Alexa machine. So Alexa says “The outside temperature is 84.3 degrees”. You’ll note that while the natural language processing is impressive on both ends, there’s no real Intelligence on the back end. All it did was know you’re sending text messages. Now that’s not true of all Alexa apps or but you have to remember is that you can provide the appearance of Intelligence without actually having Intelligence in the system will come back to that in just a moment.
Here’s a link to the OurWeather Alexa tutorial: https://www.switchdoc.com/2018/01/tutorial-voice-time-ourweather-and-amazon-alexa-part-1/
Digital assistance. “OK, Alexa, make me an appointment to do so and so.”. That’s pretty straightforward. But you know, Alexa is just replacing the keyboard interface to go into Outlook to do the same thing. Again, it’s the perception of Intelligence, but it’s not intelligent. But it’s still advantageous and cool! I have a bunch of Alexa’s. I like Alexa’s.
Well, now another application of AI techniques is going to be upon us much sooner than I initially predicted. You know, GM is launching a real self-driving car this year. Wow! You can buy a car that drives itself. This is a fantastic thing because driving in an environment with a car when your car is a robot, in this case, using all sorts of AI techniques, machine learning, navigation aids and things like that, that’s amazing!
It can navigate on the street. That’s pretty impressive. You know, 20 years ago when people started talking about self-driving cars, they were saying, wow, you know, what we can do, we can build highways with a ton of sensors on the sides telling you where you are on the highway. And then we can put automated cars on the road, you know, we’ll need to keep the riff-raff, human drivers off. But we can put computerized vehicles on the road. And this is what I find so remarkable because the way it’s coming out, our machines have gotten good enough to move into our world. We don’t have to construct an artificial highway just for automated cars. No, these self-driving cars work in our environment with human drivers around them. I just wonder, if we get a few automatic cars out there and there’ll be people that want to bully the automatic cars, you know, like slowing down in front of them to make them slow down and things like that.
I wonder if we’re going to have anti-bullying laws so you can’t abuse robots that way. By the way Coeur D’ Alene, ID has a law that you can’t injure a robot. It’s the first municipality in the country to pass that law. What good is that? I don’t know, but it was cool that they did. OK. So what about business planning? Looking at data, looking at big data like the kind of data people get off, Facebook, things like that to figure out what the customers want and how to provide it to them. All sorts of different applications are possible. And then you can do some design analysis where it can look at the design you’ve done and suggest better things, you know like “well, this isn’t too good, but this is good.”
Summary
Ooh? I never thought of that. Now here I am anthropomorphizing that robot again. So let’s summarize. We talked about what AI was. We talked about the kind of AI techniques. And we talked about some of the applications that are in use today. AI is here, and it’s doing things. These applications, these techniques are being used all the time. But one thing I want to warn you about is the media, the news media, and the government. They can be just as bad as the news. Media tends to overhype things. They say, “well, Google invented AI that fought with itself,” you know, they’re anthropomorphizing what is a bunch of mathematics dickering something out.
So the media tends to over-hype small results and the academic community does the same thing. If you’ll look at this particular software interaction and if we stand on our head and maybe open the window by three inches and get some air flowing through here, perhaps we can say these two AI’s were fighting each other. They weren’t fighting each other. They were mathematically constructing something out of conflict analysis or something like that. So watch out for media hype. But on the other hand, self-driving cars that should be hyped. The fact that we can do something like that and years earlier than anybody expected is impressive. So AI has a lot of excellent techniques, and it can be applied to problems that will make your life easier. I like Alexa, does it make my whole life easier no, but it does some cool things for me.
I love the way it can change the lights in the house to red, which drives everybody else crazy, but I still think it’s cool.
So I want you guys to think about AI techniques. This is the future, and it’s going to be glorious. Now, I’m not a Luddite. The Luddites were people in pre-industrial England who were breaking all the looms because the looms and the industries, we’re robbing their cottage industries where they’re making clothes by hand. Rising tide lifts all boats and honestly, we’re so much better off than we were in the 1700’s, there’s just no question about it.
I believe AI is going to lead to a renaissance where people are doing more productive, more exciting jobs and more importantly there’s going to be more jobs than ever. That’s typically what automation has done, but that’s the subject of another talk.
I hope you enjoyed our little talk about Artificial Intelligence. I just want you to be a bit scared of those words, but not a lot scared. That was the whole point of this talk. Anyway, thank you very much.