Jan 15, 2020

This week Dan and Ray talk through a real educational problem and how we can solve for this using Artificial Intelligence.  We look at the problem itself and unpick the outcomes, and then discuss the data that institutions would already have, and more importantly don’t have, to use AI and predictive analytics to solve the problem.

At the end we discuss ways that you can move from theory to practice to give it a go, and Ray talks about how he's used Azure's Auto ML (https://docs.microsoft.com/en-us/azure/machine-learning/service/concept-automated-ml) to build his own experiments, and they both talk about what they've learned from the Microsoft AI Business School (http://aka.ms/aibs)

TRANSCRIPT For this episode of The AI in Education Podcast
Series: 2
Episode: 2

This transcript was auto-generated. If you spot any important errors, do feel free to email the podcast hosts for corrections.

Hey Dan, how are you?
I'm good. How are you?
Hey, we're back for another podcast. Dan,
fantastic.
It's going to be a great conversation because I think we should talk about where we finished last time, which was gosh, there's so much data. How do you use it all? And and why do you use the data?
Yeah. And and what are the big challenges in education? And let's try to solve a few now. Is it?
Well, we're going to just solve the big challenges in one podcast.
Yeah. In Well, let's do two even.
Oh, okay. Two two big challenges
in one 30inut podcast.
Okay.
That'll be good.
Wow.
You've been working in education a long time, right? Thanks. It's okay. You look younger than you you sound.
I do on radio. D.
What are some of the biggest challenges that that are occurring in education that you've seen that we can look at today.
Oh well, if you want to talk about one, how about uh student attrition or student retention?
What's the difference between attrition and retention?
One is how much you keep, the other is how much you lose. It's the same number,
right? Okay.
From two different perspectives. So, I'll give you an example.
Yes.
Uh in universities about 80% of students make it through their first year of university. That's your retention number. Your attrition number is In universities about 20% of students drop out in year one.
Right. Okay.
Okay. The two numbers should always add up to 100%.
Absolutely.
But it's not just a university issue. If you take schooling, about eight out of 10 students make it through and graduate year 12. About two out of 10 students don't. In fact, the actual number has improved a little bit now in Australia, but it's still 17% of students don't graduate year 12.
And and the that's that's shocking, isn't it? And the the other interesting one is there is retention issues between school systems as well, especially in Australia where you've got people that move children between school systems. So they might start in the Catholic system, move into an independent system or start in the public system.
Yeah. So there's a systemwide definition of retention and then there's individual institutions. So if I do it at a university level, a university losing a student that moves to another university,
y
the system says that's still retained, but that university has just lost a paying seat in a classroom.
So, so why why is it okay? So, when you use mean a pay and seat, let's really get to the numbers. So, so the ultimately the problem is in a university context depending on the course.
Oh, yeah.
Or is it depending on
uh the numbers vary across different courses.
Yeah.
And they vary between types of students. So, for example, you get higher attrition in online courses. You get higher attrition in domestic students versus international students. Strangely enough, the international students drop out at a lower rate. But ultimately, I'm going to go back to dollar and cents.
So when you think it about as a dollars and cents problem, it's about $3 billion worth of students leaving each year from universities in Australia.
So that's a big dollar number.
Yeah. So let's just stick with the dollar number for a minute. So it's about a student's lifetime value is at least $50,000 because they're paying about $30 to $40,000 in fees. They bring in almost as much money from the feds as they contribute in their own fees. And then There's a whole load of campus revenue, the gift shop. Somebody's got to buy the sweatshirts with the university name on it. Somebody's got to live in the student accommodation that the universities are building. So, probably the real worth of student economy around there.
Yeah. It's probably 50 to 100,000. But if you just stick with $50,000 as a lifetime value of a student and then you say we lose 20% of them, that is $3 billion.
And obviously it's important as well, not just on well on the financial side because obviously as David Kellerman mentioned in one of our previous podcasts that that money goes back into research. So the universities, the more money they've got, the more they can research, the more that they can do for society.
That's exactly right. That money pays for teaching staff, but it is important it pays for research because 40% of the funds that students provide to a university is used to cross-f fund research. So if you're a researcher in a university, you don't mind so much about losing $3 billion worth of student revenue, but you probably do mind about the fact that it means you lose $1.2 billion worth of research funding.
Absolutely. So if you retained 100% of your students, you'd be doing over a billion dollars more worth of research in universities in Australia.
And and on the side, we're going to look at retention and recruitment there today, but also I suppose on the flip side, if you using that data to improve that, you're also thinking about recruitment as well at the same time. It's the same it's it's also attached to the same context.
It is. But let's just think of the life chances part of the story as well
because some of the biggest entrepreneurs in it are university dropouts. Bill Gates Steve Jobs,
Richard Branson's. Yeah.
The uh I found out recently the founder of Boeing was a university dropout. There are these exceptional stories of people that have done really well from being a university dropout, but the data says on the most part, people have less life outcomes as a result of not going to university or not graduating high school. So, forget the money side. Think about the individual. What is the life chance that people are being denied if they're not successful at going through school or university?
Okay, so we know why. Now, the interesting part then with this problem is we know why we need to do it but putting yourself in a position of a student and in our own life experiences as well. Why does attrition happen? Cuz you need to know that to be able to solve it, right? So if you're a student in a university, let's take the university context for now. Why would you drop out of uni?
There's a whole host of reasons. So if you fail part of your course, you maybe involuntarily drop out because you you can't carry on. And and that happens both domestic and international students. So your first assignment becomes a really important thing. In fact, your first assignment might be the most important assignment you do at university because if you come from a low socioeconomic group and you fail your first assignment at university, the internal thing that's going on in your head is you're go is you're saying, I wasn't really cut out for university in the first place because you have that feeling that you shouldn't be there because you failed your first assignment. And so things like that become really important.
But it's also other things. It's better career options. There's a job offer or you're a carer in a family or you simply cannot afford to be studying 3 years at university without out making an income. There are a whole host of reasons and in fact if you look at the data most of those reasons are actually related to things that you know before you arrive at university,
right?
And the same is true when you think about high schools. So why do students not complete year 12? It's a whole range of societal factors. So part of it will be you don't feel successful. You don't feel ready for it. That the nerves that you see in every student before the year 12 exams, well, for some students, that tips them over the edge to go, I I'm not going to do it. But also, you've got people that get a job before they get to the end of year 12. So, they drift off into, say, a manual labor job and and don't finish school or they come from an underserved group where they have society pressures or they have challenges around getting to school. There's a whole load of reasons, but again, there are very specific groups of students that tend to be the groups that drop. out in high school as well.
Okay, so we know why that is important and we now know some of the reasons all be it quite complex of why they all connect together, what data we actually could start to think about. But when we do start to unpick that, what are the data sources that we have already in our institutions that actually can provide insights into this and why do we need AI?
I was going to say this is coming into the AI bit of the conversation now, which is AI is about that ability to be able to learn from the past to predict future. Yeah.
So machine learning, for example, looking at the data from the past in order to be able to predict the future.
You bought these things or you looked at this product, this is the next product that we think you're going to buy and put showing that to you on the web page, but doing that at a much more societal level when we think about students dropping out. And so it's looking at all of those data sources and being able to predict the students that are likely to drop out and maybe providing better support for them.
And so that's where the AI bit comes in is how do you use AI in order to help reduce the attrition rate so that more students graduate.
So we need that data then. Yeah. Yeah.
Yeah. And so some of the data is very basic and is easy to find.
So let's so let's talk through that then. So some of the data from a university context would obviously be you mentioned some earlier on but the things that jump out to me would be you version of a school information system or a CRM system with students teachers. What details these days to you apply for university, do they collect
they actually collect a lot of the data that it could be used to decide whether somebody is likely to drop out, predict whether somebody is likely to be dropping out.
And so some of the key bits of data are if you are the first in your family to have gone to university. So if your parents didn't go to university, you're more likely to drop out. If you are from a low socioeconomic group, you are more likely to drop out. The degree that you study is going to have more of a factor.
I I still think like I just trying to think of courses and things that I've signed up for recently. I think unis are in a bit of a balancing act on I'm sure they want to make it as easy as possible to sign up for a degree or a course. So I reckon I'm thinking this is totally hypothetical but I think they're going to if I go on to a university in Australia's website today I think they're going to really care more about my name uh my citizenship and my payment. I know it sounds a bit harsh. Do they really ask me for a lot of those questions or is that once I've enrolled or because
yeah as part of the enrollment process that's when they do it. So they often don't collect it earlier on in the process. They'll collect it as part of the enrollment
and the reason they collect it is because the federal government says they have to
because the federal government know about the implications of this data
and they might have funding which they can release for certain areas.
So so for example they would have funding that is around getting more students of Aboriginal tourist rate islander descent
coming to university. So they collect that data in order to be able to do the reporting but also be to be able to access the funding and that turns out to be an indicator as well. I've read a number of reports about student dropout and the same key factors come out every time. So things like first in family and and this isn't just within Australia, this is internationally, income of parents, you know, carers, things like that. So they actually collect a lot of the really important data already.
But I suppose like just to pick on it again cuz This is critical really if the data they're really collecting at the minute is based on a couple of things. A the funding that comes from the government and then also data that would allow them to be added to the IT systems in school uh in university and actually some general information but not everything that might they won't collect every factor I wouldn't have thought.
No they don't collect every factor but for data they don't collect they can use proxies for it. So for example if you want to know know somebody's socioeconomic status and you haven't collected that data. Then often people will use a postcode as a substitute for it. So they know that certain postcodes have a higher proportion of parents that went to university, they know that certain postcodes have higher average income or a lower average income. So even in the case where they don't collect the specific data for that specific student, they still have data that they can use on a cohort to start predicting outcomes. So just skip across to an example I did in learning which was using postcode and income I could predict the NAP plan score for a school within 85% of accuracy across quite a range of data because there are these other factors that you can use and so I guess the same for student attrition there's some very direct data that you can have on those students and you collect and then there are other things that are maybe slightly grayer but help you to improve your accuracy about being able to predict whether this student is likely to drop out or not.
Yeah. And that's the AI element in the back end. So you can start to kind of highlight some of those things and using predictive analytics understand based on the variables you have which you've collected and which you've got at the start of the year then what percentage it would be or how you'd flag that.
Yeah. So let's think like a retailer for a second.
Okay.
If you are an online retailer what you want to do is to sell as much as you possibly can. So you look at the way that people browse your website. You look at the habits of other people so that when I put the red dress into the shopping basket. I'm going to get a suggestion that I might want to look at the matching shoes because they know that people's habits are they tend to buy a complete outfit. For example,
I'm just thinking of you in a red dress. You're amazing.
I know. As I was starting the example, I was thinking I'm putting a horrible picture into people's minds. But that habit of collecting that data and being able to use it to anticipate the next thing.
That's where retailers, they're driven by a very short-term how do I sell more products? But they're also driven around how do we retain customers. So you think of Spotify. Spotify get you to pay a subscription for music. Now they don't want anybody to drop their subscription. So what do you think they look at in order to understand whether people are likely to end their subscription?
Oh, that's a tricky one.
So if I'm likely, because I did end my subscription the other day with Spotify, they're probably looking at the the amount of uh music that I listen to over a You do a day and a month and a trend.
Yeah.
To see if I'm listening to less and less and less. Quite straightforward, really.
If you've got family accounts, how many of your family are using
Oh, that's true.
the accounts.
Yeah. Because I'm less likely to drop it even if my behavior is going down if everybody else is is continuing growing. Yeah.
Right. So, switch that across to a school or a university. What data are you going to use to help you to predict whether a student's likely to drop out?
Attendance.
Yes.
Because attendance is a key indicator. Yeah.
In fact, I know from a project that was done in TA that you could predict with 85% accuracy which students were going to drop out from things that you knew before they even started coming to TA.
But the one that increased the accuracy from 85 to 90% was attendance. And so if a student's attendance is starting to drop off,
then you know you're likely to have an attrition problem with that student.
But so we've we've got historical data then which is the stuff when you're going in, you know, the stuff well it's not necessarily historical but point of contact data, you know, where where you live, you know, the social economic background. Then you've got the data that's collected real time. So you've got your attendance data, which is vital.
Yep.
And the kinds of data.
Yeah. So remember last podcast we talked about the different systems and people's views about the data in the different systems. So what have you got in your lecture recording app? Are people viewing the lectures that they're missing in person? Can you link your attendance data for a lecture to the lecture recording and go, okay, so we've got 10% of our students that are neither attending the lecture nor watching the lecture recording. Put those two bits together that gives you an indicator that you can use predict. Are they watching the whole of a lecture recording? Are they watching it within a few days of the lecture or are they coming back the day before the exams are due and cramming it all? So that's data that you've got in your lecture recording system. What about in your learning management system?
There's a lot.
There's a huge amount of data. Now for a long time, people have fixated on lots and lots of different data in there. It's like how often are they logging on? Are they logging on every day? Are they logging on?
That's the one that bugs me. The logging on one, you know, have they logged in? So that's almost like an student.
Yeah. But then the challenge is you might then start to oh we got to find ways to make them log on. Well, that isn't necessarily about the behavior of what's in the mind of the student.
Motivation is an external thing which is you're forced to do it. So then you start to look behind the behaviors and there was some research published by Blackboard a couple of years ago. I can't remember if it was done by Blackboard or by one of their customers, but it identified the strongest indicator of attrition in the LMS was whether the student had had logged into look at their markbook.
So, did they care about the grades they were getting? Because if they weren't logging in to look at the grades, that was a big indicator of attrition. The other one would be, are they logging in to download their assignments? And when are they logging in to download load their assignments? And then probably go going over other things. It would be special appeals. So, are they always late with their assignments? Is there always a reason why they're asking for an extension on an assignment? Now, that's
and and the other the other thing from there as well where I've seen a lot of interest in correlations in K12 and I'm sure this in uni as well but maybe not done as effectively. I'm not sure but where it's almost more siloed than a uni. So I'm an engineering lecturer and I'm watching my engineering results looking at my markbooks there but it may be that I'm doing really well in engineering and my students doing really well in engineering but not in mathematics or in physics they may drop out and it's nothing to do with the data and the stuff that you're looking at but in a in a in a school context. I know they try to correlate the data more and we've seen quite a lot of that correlation of data between subjects recently, but what about in a university context? Do they correlate data or are they still quite siloed? How do they highlight things there?
Well, you've actually got data from a lot of different systems and places. So, a university is a bit more like a high school in that you've got a series of faculty areas, but the difference is the students don't cross those faculty areas. But then there's a lot of core data underlying what happens in the university, you know, are people coming on campus? Are they using the campus Wi-Fi? Are they using the coffee shops? All of those things all build together to paint a picture of a student that can help you to predict whether they're truly engaged with the university or not. So, we know the data that we've got and so the key pieces. So, if we were going to simplify this down and we wanted to really make sure that we were looking trying to solve this problem, we've got our attendance data with the socioeconomic background data. We've got the other indicators from other systems inside the school or university settings. So the LMS data and things like that, the markbook data, whether how you actually doing is a good indicator, right?
So I'm going to stop you right now, okay?
Cuz we could go on with a list of 200 different bits of data we should be looking at in order to improve the accuracy of our prediction of students dropping out.
Fair. Yeah.
But I think that is where the problem sits because the data scientists want to build the most perfectest algorithm. The one that can predict with 100% accuracy who's going to drop out.
And that's when everybody gets bored because the data scientists are going really precise with their algorithms and trying to avoid an incorrect prediction. But actually I say that if you can predict with 85% accuracy people are going to drop out. What that means is out of every 100 students 20 are going to drop out. So you might predict 16 accurately are going to drop out. You might also say a couple of students that are going to stay are also going to drop out. Well, it doesn't matter. Live with some inaccuracy in order that you do something about it because it isn't about great forecasting a dropout. It's about doing something. So, what do you do when you've got that data? What does a lecturer do differently when they know that these are the 10 students that are likely to drop out in their course? What does the system do differently about the first assignment for a student? If you predict they're in an atrisisk group, will you provide better support for them? Maybe you give them student services. support. Maybe you run special assignment help sessions.
It's about the what you do once you've got the prediction. And sometimes I think we've overfixated on making the prediction right and underfixated on what do you do with it.
Yeah. And I suppose that's where like AI can can step in if you do have a lot of data. AI can do a lot of that tuning of that data for you to allow you to make a better prediction I suppose. And that's where you're going from your AI now to your BI. Mhm.
the pit that I fell in a couple of episodes ago. But then going into your BI, so you're getting timely information for your uh lecturers, for your student support workers from a different angle from the student to the students themselves. I suppose not so much parents in a uni context, but from a K12 point of view, parental involvement and information as quickly as possible. As we know, the reports these days in schools, you know, are quite untimely. And then having the action of what to do, which is critical. I suppose once you've got that information back. So the actual causation, the effect, the effect that's happened and then what you can do to fix that.
Yeah. And traditionally we've had student information systems that are focused on recording the data on the student, not managing the life cycle of the student and intervening.
So then that's where you need, you know, like a CRM system because in a in a business, you'd have a CRM system that says, "Oh, these are the customers that are likely to drop their subscriptions."
Yeah.
Great. Send them a campaign. Send them an offer. Make them feel better. about their products and services. Having that same set of tools in place in a university to proactively go to students or in a school having the thing that says here are the students that get get the special support now how do we give it to them? Having a life cycle approach to a student in order that you can prevent those dropouts and you can support the students that are most in need. But being able to identify them is the air bit and then it's the relationship bit. And it's not just about people because If you've got 10,000 students in a school system and you identify the 2,000 that are likely to drop out, you can't just go to the teachers of those 2,000 and say you have to give them special attention because you're adding to workload, not reducing it. So, what can you do with your systems and your processes in order to be able to support students? What do you do to help them achieve at their assignments? What do you do to help them feel that they're part of the learning community as oosed to maybe feeling outside of it. What are the differential things you do?
Cool. So that's really interesting. I think you know I suppose what is what's alluded to me is that we can start to in 15 minutes get to the nub of the problem and look at the data we've got there. But it's about us in our own context looking at that and coming up with a solution based on not all the data but there's data we do have. There's data that we might not have that we've collected for a different purpose in the first place or something we there might be small piece of information that we could add to the enrollment form which could make a big difference for example.
Yeah. And it's interesting that there isn't that thinking about collecting that data. So my eldest daughter when she started at university actually started as an international student. So they didn't ask you to fill out any of that information even though it would have been useful for the university to know whether she was first in family or whatever because they were fixated on the business process side rather than the learning journey.
Yeah.
So We've talked about in the context of student retention
which is a problem for the high school system as well as for the university sector. What about personalizing learning? What about that journey? Where where does the data come in for personalizing learning? Because I see it being used by online retailers and online websites to personalize the information they show to me. So how would you use it in that context?
Well, well again it's it's siloed, isn't it? In school systems. So for personalization for me as a and it's very similar whether you're in K12 school or in um a university context. It's all about delivering the content and the curriculum that I want and there's a big debate currently in education especially in the kindergarten to 12 years where people are thinking about project- based learning a lot and how you can actually change that because at the minute it's siloed and we don't want to get into that in this debate now but the discussion around personalization essentially is exactly like you said it's like retail is I'm going in and I want something delivered to me as a service and it's learning as a service almost it's what what am I studying what subjects have I got and I suppose the complexity in a school situation would be the fact that you studying maybe 15 subjects so how do you correlate those together so when I'm thinking about personalization for me as a person to a certain extent as a student going into say a high school if you pick the middle ground here a student going into a high school point of view would have a different personalization experience because they they're limited to what they can really do the school systems will say well you need to do maths you need to do science and to personalize it to me it's not really personalized the only way I'm personalizing it is actually by getting delivered individual support for subjects I might be struggling with so that the idea I suppose when we have to think about education generally personalization it isn't as simple as one answer for a for a student in a primary school like my kids personalization is very much around the holistic element of the child and learning and how they move in forward and how we can support them with well-being and their studies and general key indicators like literacy and numeracy. In a high school situation and a university situation, personalization is much more granular and in it's definitely in a high school situation, it kind of sort of implies more of a support for individual subjects in terms of personalization. So how you where you might be struggling at Pythagoras in maths, how can I come and pick you up there early on and how can I make sure Like the earlier example, we you don't drop out the maths or whatever it may be.
I notice it's always me that's being picked on as the one that's struggling. The But you got it wrong, Dan. With Pythagoras, my problem was always spelling it, not doing the maths.
Okay.
Okay. So, what data points do you need to look at? So, that this bit about AI, AI learns from looking at data in the past to predict the future. So, let's say you want to think about a student and their score when they graduate year 12. What kind of data points do you need in order to be able to predict where they're going to go? in order that you can start to build your interventions.
Yeah. Well, some some of the key ones are literacy and numeracy because the the literacy is a key indicator because of the access to the curriculum from the start. If if they can read and write, then they can access a lot of the other subjects. So, literacy is a key indicator.
I remember somebody telling me that they could predict somebody's year 9 maths result in the SAT/NAT plan/national tests
based on their year3 reading. Because if their year three reading was weak, it meant by the time they got to year know and they couldn't understand the question to do the maths test.
Yeah, you were saying attendance.
No, absolutely. Attendance is obviously a key one. So, I think sometimes you do get bogged down with that and sometimes you oversimplify it. Uh, which is a good thing in some cases. I remember when I was teaching when I had a year seven cohorts coming in to the school for the first time when I was in high school. I got given a sheet of A4 paper for every kid and it basically gave me box and whisker diagram. So, like almost like a percentage of where those students would end up. at the end of high school. Yeah.
A right at the based on their exit from primary school. So it was about their uh numerousy and the literacy that they left primary school with was giving us an indicator of how many GCES which is the year uh year 11 exam you get in the UK at the the high school certificate say for example in Australia it's given you a really good indicator of well a really good indicator it was given us the best indicator we could to work out where where that student was. So it was using prior learning from primary school. There was no other data overlaid over that.
But that isn't personalization. That's just about predicting your future. So if you think about personalization, and I'm not going to allow you to use learning styles here.
Okay.
But what other data would you like? What other data could you use about a student to be able to personalize things for them?
So when we looking at data generally, like I said earlier on, we've got our attendance data. We've got the uh interaction with the lesson and the content, the assignments they've done. their progress in the assignments, the progress they're doing in between, you know, formal assignments and informal assignments, assessments by the teacher. You know, there's there's a lot of data points there which we don't collect and also things that we do. So, if you were going to look at things that we do have, it's about their interaction with some of the tools and technologies they might be using in terms of their learning management system in a secondary school point of view. But also, you know, the main D, if you go into any school and let's really unpick it. If you go into if we went now, we left the studio I went into a school and we landed in a lesson anywhere in the world and you go in there somewhere on that teacher's desk will probably be a piece of paper or a book or a planner or something electronic which will have progress over the last 6 months or the year where and they can have a good conversation with you about that. So when we look into personalization you'll be able to look at data in there to do with their attendance to do with the work they've completed to do with their homework and to do with any other elements that they flag which might be their socioeconomic background um learning difficulties English as a first language and second you know there's there's about seven or eight key indicators there as a teacher if we just landed in a classroom and they could tell us about that child
and I think if I jump to what David Cullerman was doing and what he talked about in his project at UNSW they're looking at all the assessments at question level to be able to understand what are the concepts that you don't understand so that they can provide a personalized revision guide for every student before the final exams. So I guess it's that level as well. It isn't just about the overall score, it's about what are the component parts and that's what we don't do very well in in in education generally in in high schools. You know, you'd look at the overall score and you go Ray scored 25% in no sorry 95% in his Pythagoras uh maths examination at the end of the topic. But actually, you know, it doesn't give me an indication of the 5% you had wrong and I might have got that and I might have fed back to you on a piece of paper directly but that's not recorded anywhere. So there's a lot of unrecorded data and support that without adding to the workload teacher that isn't done.
There's also a lot of times that we don't collect the data. I I think about Netflix every time I watch a video on Netflix I'm giving it a thumbs up or a thumbs down at the end of it
in order that it can personalize the recommendations for me going forward which is great as a consumer but I don't think in any online learning course where I've had to watch a video anybody's ever asked me to say did it work for me
and yet I'm guessing for students they all have different reading ages they all have different styles you know collecting that kind of data would help to both personalize for a student but also provide valuable feedback into the system about which learning resources are working and which ones aren't
yeah and there's a lot of lost data there you're right you know just listening to you there and just this conversation you know the lost data that that happens during the manual marking process and I'm not advocating online marking here by any stretch of the imagination but if you are teachers spend a lot of time when you ask what they're doing what they're spending their time on is marking and all of that feedback which is valuable because we know that good quality feedback will improve Ray's results in mathematics and Pythagoras getting from 95 to 96% but that feedback ends up in the bottom of your bag with a squished banana at the end of the year that's just a reminder to anybody on the podcast. So, just if it's a Friday, don't forget to get those uh uh lunch boxes out of the bags. But, um yeah, make sure you know that that data that we've collected or haven't collected is is captured somehow. Possibly.
People that have been following this podcast for a while will have heard us talking about different AI concepts and they're probably starting to think, well, this is interesting and I might have some data. How can I go and experiment?
Because we have the ideas, but we don't have the data. Some of the people listening to this podcast will have the data and they have access to it, but they maybe don't have the skills. So, what do they go and do?
So, well, one of the things is, and I think you exemplify this in Microsoft, to be honest with you, Ray, it's about giving it a go. You know, I think there's a lot of tools out there that you can just give things a go. I did a session recently with a load of schools in Sydney and I just got them to bring their own data and we went into one of our tools called PowerBI and then they put the data in there and came up with insights and looked at our data and just use the data they've already got and giving it a go and doing some learning around AI and some of the tools and technologies that are there and bringing the data data in and just giving it a go like you've done this. What would your tips be on doing that?
Uh so I love giving it a go. So from an AI perspective my most recent give it a go was with something called AutoML in Azure which basically means I can point the system at the data and say go and work out a prediction for me. So the what what I've been trying to do is I want to be able to predict the NAP plan score for a school. I'd love to be able to do it for an individual student, but I just don't have access to the source data. But I do have the NAP plan data for Queensland schools because they publish it in a big spreadsheet
and the sky hasn't fallen down.
So they publish a big spreadsheet for all their schools. And so I take that data
and then I look at other things that I think could influence that and I just use AutoML to go and do an experiment to find out how accurately I can predict the outcome of schools that I don't know the data for.
And uh what's really interesting is that I found that there's a lot of society data around that comes from the ABS census that is a better indicator of nap plan results for a school
than some of the education data. I don't need much data from education to understand it. It's factors around the society that affect that school's intake.
Yeah. And and then so I suppose that's the that's from a technical point of view then looking at some courses giving it a go, opening up AutoML in Excel or using some of the tools around like PowerBI and things like that. So, actually giving it a go, putting your data in. And then in terms of the general kind of thoughts around AI and learning about that because obviously we talked a lot about governance and things. What about the AI business school? I know you mentioned that.
I think AI business school is probably number one go-to and the reason I say that is because it links the technology to the business problem. That's exactly what we've been talking about today. We've been talking about the business problem first, which was student attrition or personalizing learning and then we talked about the data and then we talked about the technology. So AI business school is I think the only resource I know that links those things together because you start with the problem and you don't have to follow the education track in AI business school. You could actually follow the retail track and see how retailers are using data or the local government or the healthare one.
That's great.
So I I would say take any of those because you learn how to relate the business problem
to the AI potential to then see how you solve the problem. And so I know we'll put it in the show notes, but the URL for AI business school is aka.ms/ AIBS.
Okay. Okay.
BS is for business school.
Okay. Okay. I suppose finally one of the things you know we are working with lots of schools and universities to bring in specialists in this area as well. So there's a partner ecosystem around these things in whatever technologies they're using. So bringing experts like we had the cloud collective previously to come in and look at that data and do that analysis. But I do think just listening to you talking there and having that discussion around the entire problem. I think once we understand the problem and start to articulate it and spend time locally in our school or our university setting thinking about what the actual big problem is we can understand a lot of that. We didn't need consultants in today. We could kind of talk about well roughly what we thought the issues were and you can get quite a lot of insights from Well, you think we didn't need consultants in today? Maybe we were just talking twaddle then.
Oh, yeah. True. Yeah, that's a good point.
Okay, so just how we avoid that going forward, maybe we should have some people that really know what they're talking about come and join us on the podcast.
That'll be good.
Let's have some of those partners and people that are doing interesting projects in education or have got solutions building AI for education. Let's see if we can track some of those down and get them to come and talk with us.
Sounds fantastic, Ray. Thanks, Dan.
Cheers.