Prescriptive Analytics

By Stephen Downes
Oct 26, 2021
Transcript of Prescriptive Analytics

Unedited Google Recorder transcription from audio.

So, this is the presentation on prescriptive analytics, module two, ethics analytics and the duty of care and I'm, of course, Stephen Downes. And thank you for joining me. If you're watching you probably not on YouTube right now. I don't see any viewers or if you're watching the recording that comes later.

So I'm just going to set up these slides and these slides will be up and running here in just a second. So and again I have to set up the slide show because Microsoft always resets it to presents full screen, but I've done that now. So I'm going to start and there we are.

Prescriptive analytics. So the topic of prescriptive analytics, essentially is a branch of artificial intelligence and analytics that makes recommendations for me. It'll say, you know, we have an example here of marketing analytics work this deal, not that one. So this product, not that product. Do this task, not that task etc.

And we can see how this is going to have some obvious applications and teaching and learning both from the perspective of a teacher. Making suggestions about what to do next in the classroom for example, and from the perspective of a learner, what to study, how to study, where to study it when to study it.

So as with previous discussions of applications or uses of analytics, I'm breaking this down into a number of different topic areas or categories as I'm calling them. This breakdown is purely arbitrary and of course we could think of many other ways of dividing up different applications of learning analytics but that's what I've chosen for this point.

So the first and foremost of these and of course you had to be expecting, it is learning recommendations and this breaks down into two major categories of prescriptive analytics. On one hand, we have content recommendation systems. Now, content recommendation goes back, several decades. And there have been a variety of different approaches to contact her content recommendations over the years.

These days, it's based on a collaborative, filtering and other AI metrics that identify what people like you who are in your program, who are similar, aptitudes, maybe similar learning styles. Any range of properties what they have done well with, then the system recommends, a similar sort of thing for you a more difficult challenge.

But one that has been of interest in the AI and analytics community, for some years is learning path recommendations here. We're not just recommending learning resources on a case by case item by item basis. But rather we're recommending a path through a variety of different alternative resources or even topics for learning in order to reach a learning outcome or a learning goal.

So these kind of recommendations may even select courses or lessons or topics. They may simply outline a learning path or it may offer personalization parameters. Again maybe based on learning style, it may be based on mastery learning maybe even on things like time limitations or knowledge background, they're used by course generators course or learning objects sequencing systems and they can be used for both online and offline.

Education. The second major category and this is related to the first is adaptive learning. These are AI systems. That look at how a person is performing in the context of a course or a learning environment and ads this system to that, we see this morning games actually, then we do and educational applications.

A gaming engine will track a person's progress closely and adapt, the presentation of challenges enemies, whatever to the person in order to challenge them. But not too much. Similarly with adaptive learning you're trying to challenge but not to much the individual learner. You want them to push themselves in order to succeed but you want not to push them so hard that they give up.

So you have to of an artificial intelligence that monitors, what the student is doing, and then adjusts the difficulty of a content accordingly. Another class of prescriptive analytics is, especially useful, in large, open online, courses. And that's adaptive group formation and maybe educational scenarios, the use of small groups is encouraged in order to enable each person to take part in the discussion to engage in small group, dynamics conversation, feeling more, like they're participating in contributing to, the course, rather than simply receiving content.

This is not something you can do with a single group of a thousand people or more. So you need to break down your course into small groups. This however, can be difficult people who know each other, make cluster into groups, but you're going to have a especially online a lot of outliers as well.

People will be participating to different degrees. You'll have people who are very engaged and active and other people who are more lurkers or as we call them legitimate. Peripheral participants, indeed, in your group, you may have people dropping out of the course, part way through. And so there's the risk of a scenario where a person is the only act of member left in their group.

So there's a need to have adaptive group formation that takes into account. All of these and additional variables. And to manage the group for nation as the course continues. In order to ensure that people have a useful and productive small group experience

Related to this. And similar to this, in many respects is placement matching. This would be used by systems or applications that are looking to provide real world experience. For example, an educational program looking at co-op placement, may use an AI system in order to match potential students, potential co-op employees.

In other words with potential employers, There's another case in the government of Canada where the treasury board of Canada pilot a project with us called Micromachines, where government employees would be considered for short term assignments, in other departments, just to broaden their experience and help them learn new skills.

And an AI was used to match these potential short-term placement people with the term placement opportunities. Any place where you need to match a person with a resource or a person with an opportunity, you're going to have the need for an opportunity to use artificial intelligence and analytics to make this more effective and especially a lot more efficient.

And that brings us naturally to the topic of hiring. I've been involved in a number of AI for hiring or AI for job interview projects, over the last few years. Right now, we're seeing corporations already, using artificial, intelligence, and automation in order to push the hiring process. I haven't seen a case where MAI has outened higher to person yet, typically what happens is that these systems work hand in hand with human recruiters making suggestions creating the shortlists, you know, matching a pool of candidates to a specific job profile or opportunity.

That said, these systems can do everything from phone interviews, posting ads, screening resumes, even prescreening candidates. So that when the human actually gets into the loop, they're looking at much less time and resources in order to accomplish a hiring goal. Similarly artificial intelligence or analytics can be used to make pricing decisions.

We're seeing that a lot already. In other industries for example, in airlines AI supported applications help with what is called differential pricing where this system. Calculates how much a person would be willing to pay for an airline ticket based on their background? The time they're purchasing the ticket other factors, how frequently they fly, whether they've purchased business, class tickets in the past etc.

And then makes the offer accordingly companies like Uber adjust their pricing according to demand, we've all heard of Uber's famous or perhaps, infamous surge pricing, whereas demand rises sodas the price and this is calculated not by a person sitting in a room. But by a fairly sophisticated analytics engine, the same sort of thing can be done in a learning environment.

Both for the pricing of tuition or other fees for access to learning as well as the pricing of resources such as books and applications notes. And other sorts of support, probably I don't have direct evidence but probably this is already happening in the publishing industry. There have been examples where companies that publish online newspapers or magazines or journals have been using AI to determine whether or not to put up a paywall blocking access to a resource based on whether that AI thinks that a person would be willing to pay for the resource as opposed to say, somebody who's just casually browsing as a result, I get a lot of these payrolls, but I'm still not willing to pay for them this feeds into a general set of applications around decision.

Making generally AIs, can support decision decision making at pretty much any step of the process. Typically AI will be used to handle the comprehension of big data, which is what we saw. And things like descriptive analytics, and even diagnostic analytics, basically setting the stage for a person to make a decision.

The AI can also map out the set of possible actions for the actual decision major to consider This person. Now will take into account other information that the AI might not have access to, and then I actually make the decision, Although we can imagine catway a scenario where the reason of other information available.

And so the that is making the suggestions about possible actions, maybe actually in the best position to make the business position. So those are overall some applications that constitute the area of prescriptive analytics. Perhaps you can think of more cases, or more types of cases where an artificial intelligence engine or an analytics engine can make a recommendation, or make a suggestion to you.

Certainly there are many cases and learning and development where these tools can be applied. If you do find such a case, then I recommend you go to the applicate, the list of all applications, in module, two of the course, and submit them as your own suggestions. And that way we can get as comprehensive a list as possible for the different prescriptive applications of AI and analytics.

So that's it for this video. The next video in this series will be a look at some applications that are generative in. That is applications that we use in order to create things. If you're watching live and you might be, I'm going to be doing that right away. So, give me about two minutes then reload, the reload, the activity center website.

Otherwise, if you're not watching me live, look for the next video in the ethics analytics and duty of care playlist. Or if you're listening to this as a podcast, this will be the next item in the podcast. So that's it for now, I'm Steven Downs. See you shortly.

Force:yes