Ethics and Analytics: What We Mean By Analytics


Unedited Google transcription from audio.

Hi. I'm Stephen Downes. Welcome to another episode of Ethics Analytics. And the Duty of Care. Today's video part of module one, which is the introductory module to the course, is ethics and analytics, what do we mean by analytics?

So analytics generally is thought of to be related to data and related to decision. Making for example, here to channel writing, it's thought of as the science of examining data to draw conclusions and when used in decision, making to prevent present paths or courses of action. But this is, by no means.

The only way of thinking about analytics, we can also think of it as the overall process of developing actionable. Insights through problem definition and the application of statistical models. That's what Cooper writes in 2012.

The focus of this course is going to be learning analytics, that is to say the application of analytics, which will continue to talk about here and learning or educational context. So as apply to learning and education and even looking at this definition, we see there are different aspects of analytics that we can focus on everything from data environments, contexts to the objectives of analytics.

And learning the methods, and who is involved to the stakeholders.

When you stand learning learning analytics, is typically described in terms of its objective, which overall is to increase the chances of student success. But in practical day to day sense, might mean anything from basic reports and log data through experimentation and results of trials to the organization of students and faculty to the transformation of an organization.

The way it offers its classes the way it presents materials. Even through to a sectoral transformation. This model here is called the maturity of learning analytics deployment model.

But there's also what might be called a scientific goal to learning analytics looking more deeply at the subject which is to say the learner and trying to approach an understanding of how that person learns by studying the the mechanisms of analytic systems in order to understand the mechanisms of human development and human cognition, There's a this idea that these might work handing hand to develop, shall we say a science of learning.

But generally, we want to do more than just understand. We want to optimize learning, we're looking to do what we're doing better. And so, this involves is George Siemens, says the measurement collection and analysis and reporting of data about learners and their contexts.

In this course, I want to take analytics to mean something very broad. There are different ways of thinking about analytics and it's easy to get distracted by focusing on a, fairly fairly narrow perspective. But let's look at some of the different questions we can ask Gartner. For example, offers a model of analytics that moves from basic information management through to optimization at the basic level.

We ask what happened? Then we get a little more diagnostic now we ask why did it happen? Then we try to predict what we'll happen. And then finally we look for efficacy or agency how can we make it happen? Of course the answers to these questions are going to depend a lot on who you ask typically when we look at learning analytics and by typically I mean generally or in the majority of studies that I've looked at are the majority of reports that I've looked at the focus of learning analytics is described from an institutional perspective, and we read things like slate and tate here, learning analytics offers the potential to provide educators with quantitative intelligence to make informed decisions about students learning.

So they're used by the people who provide organize and present educational or learning opportunities. But what we mean by learning and it's learning and education might be very different depending on who we're asking.

For example, here, we have an institutional leader saying it might be the case that we keep them students. We retain them, but also, we were able to provide them with better support. They're looking at it from this, institutional point of view, the teacher might be saying, you can reflect on what works and what doesn't, which should I keep doing?

What do I need the change? The student meal is looking, at analytics, from a more personal perspective. I'm always curious about which areas I'm struggling in and which areas I am doing bettering. These three domains are important to an understanding of analytics as a whole. Not only the institution of domain sometimes called academic analytics, which looks at operational financial decision, making student retention topics, like that not only teaching and pedagogy learning design, learning design, curriculum, recommendation and materials, of course, paths etc.

But also learning, from the learners perspective, learning strategies feedback, dashboards all of these aspects of learning analytics play a role in the subject that we're talking about. In this course, Another way of looking at the different types of learning analytics, is to look at the different areas which analytics is used and we're going to see a similar tripart division of the field, the UC Berkeley Human Rights Center research, team for example, divides AI tools into three categories.

Learner facing teacher facing and system facing or institutionally facing. It should be clear though for our experience with little the learning management system that the same tool might face all three of these sectors. At the same time, just three ways of looking at the same data. In fact, as we look at it, as we look at the history and developments of learning analytics over the years, there are numerous types applications and domains of analytics research and education.

We can look at online systems. Neural networks students, paper learning education, study virtual learning and more. All of these are different perspectives for different frames. We can attach to learning. And in fact, you know, as I prepared for this particular piece of work looking for these frames, looking for these ways of characterizing analytics, I saw model after model after model lens.

After lens after lens, there are many ways of categorizing and typifying learning analytics. None of them is probably best for any given application. You could probably choose the one that fits your purpose. Most tightly, but if we're going to understand the subject of the ethics of learning analytics, we want to construct.

This is broadly as possible. The wider definition avoids. The difficulties trying to come up with a narrow definition, but also it makes sure that our look at the ethics of the subject is complete that we're not ignoring potential ethical applications or ethical implications. Simply because the practices outside the scope of learning analytics attempting to avoid that.

If the question comes up, we'll include it as part of learning analytics and then we'll sort it out from there.

Now analytics is part of artificial intelligence. Artificial intelligence has its own subdivisions and ways of breaking it down. This is a useful way of breaking it down. As we can begin with artificial intelligences, self as a term, meaning something like building machines and software that can mimic intelligent behavior.

May be mimic is the wrong word. The subset of that, is machine learning where instead of providing the AI with explicit instruction or explicit rules, we focus on giving computer systems, the ability to learn from data without being explicitly programmed. And then a subset of machine learning is deep learning, which is as neural networks to shall we say, learn a representation of a data set and but near on networks here, what makes the deep learning deep, is the idea that these neurons, these connected units are organized and layers, and it's the layers that makes the learning deep.

So, to description of the topology of the network, and not say the idea that it can have deep thoughts or something like that.

So we're going to take analytics. Broadly to include artificial intelligence. We're gonna think of artificial intelligence, broadly, a software and possibly hardware systems designed by humans given a complex goal to act in a physical or digital dimension but perceiving their environment through data acquisition interpreting, the collected structure to run structured data to reason or process that information and decide on the best action, where that action could be.

Any of a number of things, including a prediction, including a categorization, including a representation and more. We're not going to limit it to action as in action verbs, and we're gonna keep the focus a wee bit narrow in the sense that we're going to focus much less on AI.

That's based on symbolic rules and much more on AI. That's based on narrow networks. In other words, machine learning and deep learning. And the practical reason for that is that most of the field has turned away from symbolic or rule-based systems. Now there's a caveat there and the caveat is that in many applications, you'll find a blended approach with a neural network being used and then rules being employed to apply constraints on those rules.

Will certainly consider such systems because such systems have ethical implications. If you apply air, don't apply a rule that clearly has and ethical implication. But we're not going to be thinking of the rules based systems from say, the 1970s in the 1980s as examples of current artificial intelligence or analytics.

Now generally through the course, I'm going to use the global terms analytics and AI or artificial intelligence interchangeably. When SAI, I mean analytics, when I say analytics, I mean AI. When I use either term, I'm talking about learning analytics more specifically. In general, we can think of these general uses as being fairly loose when we need to be precise.

We will be. So if we need to say machine learning as opposed to other methods for example as opposed to expert systems we will if we need to make the distinction between supervised and unsupervised learning we will if we need to make the distinction between convolutional neural networks deep learning etc, we will.

But generally, when I'm using the global terms, they'll just be loosely applied. Finally what makes this different? Why does you know the topic of analytics and artificial intelligence in learning create a range of ethical questions. We have an encountered before. Well, back in the 70s where and I'll wrote that the dangers of digital technology stem from three major effects and, and just pretty much captures a lot of it.

First of all, there's scale a computerization, enables an organization or today an individual to enlarge their data processing capacity substantially. We look at the modern artificial intelligence systems and they're looking at a billion data points. This is something that we just could not do in the you know, even 10 years ago, even 20 years ago, just could not do that.

Exceptional from the perspective of a human brain, which does do that. So, the things we can do with computers are in that sense different from the things that we could do with machines in the past. Yeah, that's kind of like the difference between weapons and weapons of mass destruction.

There's a significant difference between what you can do at the former and what you can do at the latter. Second thing that makes digital different is access. We talked earlier about the idea of context to collapse which is to say the case where something that you have written. Perhaps intended for a specific audience being available to a worldwide community and then hearing back from that worldwide community whether you want to or not.

That's the sort of thing that digital technology enables so people can act as data much more freely and easily than they used to be able to we read of a case, where an analytics engine is generating faces by simply gathering thousands millions of photographs of people on the internet and using those as input data.

So this access is what makes this kind of AI possible. It's kind of the same thing for individuals we now have access to things like Wikipedia and Google and maps and more. And so we have at our fingertips masses of data that we never did before. And also as we see we have access to the technology that allow us to process that data to regard that data intelligently and to draw inferences from it.

We'll be looking at many examples of that. Finally, third function computerization creates. As these authors wear a towel set, originally a new class of recordkeepers, we now have a much better idea today than they did in the 70s, what this new class of record keepers looks like. And it's not just companies like Equifax, which handle our credit data or health companies, which handle our health data.

It's companies like Facebook and Twitter that handle our messages back and forth to each other which used to be secret used to be privy used to be something that we did, except in the rare case of a wiretap without anybody else looking in. Now, there's this class of companies that has custody over all of our interactions in all of our information.

There's really no way to avoid that in a certain respect and it's the creation of this new class of record keepers with correspond with corresponding power and responsibilities that creates a whole new class of ethical questions. So that overall is what we're looking at for that analytics. Artificial intelligence learning analytics technology.

Generally, we're looking at these different categories of machine algorithm but we're looking at is broadly as possible. And I hope I've given you a sense with the short descriptions of the varieties of different applications and systems that are out there in the next section. In module two, we're going to focus specifically on applications of analytics, in learning applications of AI in learning and we're going to use to begin with the characterization offered by McKinsey looking at the questions that we're answering.

We'll find that McKinsey's characterization, false short of what's actually happening in the field. And we're going to identify and classify a large range of potential applications. Now, the reason why we're doing this isn't to create some theory or model, that best helps us classify and categorize applications of learning analytics.

I know there's a lot of theories that do not, but that's not the point here. The point here is to capture a sense of what the benefits are that we obtain from the use of learning at analytics. And, and it's important to keep in mind, like these benefits are what generates is entire inquiry into ethics.

In the first place, If there were no benefits to the technology then nobody would care. We simply wouldn't use it, but the fact is the, there are benefits and so are discussion of episodes of ethics is going to look at these benefits, and look at the ethical issues in the light of these benefits.

So we have the applications, we have the benefits, and we have the issues. We're going to try to map those out and I've I even now I have no idea what that map is going to look at. And a lot of this course consists of taking a lot of these entities.

A lot of these things like applications and issues and ethical codes and so on putting them in a chart and seeing what we see and as we get to the later sections of the course we get more of what my impressions are. But also importantly, you will be able to develop what your impressions are, and I'm sure they'll be different from mine and that's the beauty of organizing a course this way.

So that's it for this section of the course, there's one more video in module one following this, which is the wrap up discussion held on Friday and then we'll take the weekend off and we'll get back to it on Monday, with module two. So thanks a lot. I'm Stephen Downes.