#281 - Risk Assessments for AI Learning Tools, a conversation, Part 1

Tech podcast. I'm Ron Wills, creative producer at Education. Professor Rose Lucky And I sit down with my colleagues Dave Turnbull, Deputy Head of Educator AI Training, and Ibrahim Bashir, Technical Projects Manager at Educate Ventures Research, to explore the critical need for such assessments.

their target audience, and how to make them more accessible and impactful for everyone. In the second half of the episode, we're joined by Professor Rose Luckin from her own Zoom studio in Sydney, Australia. At Educate Ventures Research, the team has produced a set of risk assessments and innovation maps for the most popular AI tools for schools and classrooms in the UK. And you can find these on our website under our resources at educateventures.com.

Our hope is that these assessments can reliably inform practitioners about the features of the AI tool, whilst highlighting any limitations, possibilities of misuse, and value to teachers and learners in their context. I am the Deputy Head of AICPD at Educate Ventures.

And I also have the kind of secondary role of managing the Safe AI in schools LinkedIn learning group that we have, a community of teachers who are um using AI in the classroom and who are responsible for the deployment of AI in their in in their schools and in their educational establishments. Um and that group itself is really focused on the safe implementation of AI. And it is these teachers who are kind of my ideal audience for these risk assessments that we are doing.

They're also completely valid for just any educators who are using AI as well, who don't necessarily have responsibility for the for what tool is getting chosen at their establishment. It's a kind of way to kick start the conversations around risk. And my aim for these is to almost be the to be the bandwidth for teachers who don't have the maybe the time or the uh research background or the mental energy to um look at the risk levels of all of the tools they're using.

it's a really clear way to kind of highlight some of the risks and to offer some tips on risk mitigation. So the way they look is um a very kind of clear, concise table with mapped across description of the tool, the innovation of the tool, any data risks. then ethical risks and then a little tip on risk mitigation, followed by our EVR staff notes, which are kind of our user experience of using that tool. I think one of the key things is that we mentioned it as risk mitigation. We

I we don't think it's possible to have uh, you know, eliminate a risk entirely. We always talk about what you could do to be mitigating the risks of these um AI tools. So even tools that have got kind of low to to middling risk, we still offer a kind of a note on how you can best uh kind of best practice in terms of keeping your your learners and yourself as as safe as possible.

So yeah, that's my my kind of take on what they are. We've we've done three or four of the kind of more popular tools at the moment and we're working through them all. So I've I've actually got the got one open at the moment and They... the way the kind of map between data risk and ethical risk works is that we've kind of got the data risk side of things is a kind of technical risk analysis and then the ethical risk kind of steps more into that education focused, learner focused safety as well.

We think it's important that these these risk and uh these risk assessments are education focused. That's kind of the core difference between what we're doing here and what we've seen. Kind of there are some really large AI risks. ymwneud â phobl sy'n ymwneud â phobl sy'n ymwneud â phobl sy'n ymwneud â phobl sy'n ymwneud â phobl.

And in terms of um like the the profile of the person that that is supposed to be in engaging with this, I mean, given your We know that you have sort of extensive experience, you know, going into schools, speaking to educators and and um, you know, ha having those those conversations about AI training, you know, continuing professional development. I mean, i i i is is every teacher able to sort of understand the

i the kind of the the extent uh of this risk assessment to understand the implications of it. You know, is it written in language that that you know they're going to that they're going to understand and embrace, you know, it W what's the sort of time commitment on them, things like that.

Yeah, I think that's a a really good question. I think there are kind of there's there's depth of use for these risk assessments. If you're if you're a more kind of if you're engaged in the world of AI and education at the moment because it uh is a responsibility you have at your school, you might engage with them on a different level from a teacher who has just been asked to use a certain tool or has just had a certain tool

introduced to them, but the way that they are written is in kind of short bullet points that get the kind of key messages across without too much technical jargon. And it is very accessible for anybody without a kind of data science or AI background.

it really spells out kind of what the risk is and what the impact of that might be in a kind of really concise um short piece of work. And that that's what we're talking about with the kind of being the bandwidth is is there is back end work that's gone into these these analyses. is is time consuming and is is important to do to make sure that they are accurate.

But we're able to then kind of consolidate that down into something that's a kind of accessible piece of information that that any educator can can look at and get an idea and just kickstart that thought process of oh there might be this sort of risk to this or I haven't thought of, I haven't considered

that, you know, we need to think about where data's being s saved or what sort of data we're giving to this um tool. Ibrahim, tell us uh what the risk assessments sort of look like to you in terms of, you know, your

one of the sort of main members of the team that's actually doing these things. And um, you know, from your perspective, when you're writing them, when you're researching the tool, you know, what what kind of background are you uh of of the person looking at the risk assessment, what kind of background are you kind of uh envisioning that they have, you know, is it Is it the same as as what Dave is saying or

Like as the person writing them, how do you feel about, you know, what you're putting out into the world? Thank you for having me. I'm I'm Ibrahim Bashir. I think Uh it's interesting because when we started creating these risk successful one, it was it was much much more limited than the form it eventually took. And when I was creating them, I was actually thinking about uh a while ago, a couple of years ago, I had my I my background is in computer science, uh software engineering.

Um, and a couple of years ago, I had my own startup. Uh, this was before LLM sort of thing. This was right after I done my master's in computer science, uh, with the focus on AI. And at that point, um, me and another colleague of mine, we were Uh we had we had developed this startup where we were making an AI tutor. And I realized at that point that, you know, coming from a tech background and not an education background, there were so many things, so many risks.

that I wasn't considering that I did not I wasn't aware of that I wasn't considering at all. M my focus was was completely on tech. And I think I think Dave's already explained really well uh you know like the the kind of the teacher audience that we're that we're targeting with this, but I think there's another audience and that is maybe perhaps the tech companies to sort of let them know of of

To be fair, a lot of them are very already well versed in this. Uh some of the tools that we have been reviewing. they usually have it it does seem like that they're very pedagogically grounded. They have uh you know actually spoken to teachers, teachers have used the tool. So it's not like they're, you know, like uh

They're going in dark like like I was. Uh, but at the same time, I I do realize that a lot of these risks are not necessarily a sometimes well understood and be sometimes not conveyed in the way

uh that makes them be well understood. So in one example, when we were doing it for Teachmate AI, I remember one of the one of the founders, they they got back to us when they saw our our assessment and they're like, oh actually, you know, like the It was a point about using student data using teacher data from the users and, you know, where it was going because a lot of these a lot of the tools that we've been reviewing, they use open AI models at the back end.

So the question was was the data going back to open AI? And that was essentially a data risk. And it wasn't clear on their website whether you know where that data was going. And the the one of the founders, they got back to us and they were like, uh, well actually no, you know, we we use a closed version of that model, all of that data is secure.

It relies on that server. But from a from a reading I did of their website of their priv privacy policy, I wasn't able to at least at the first glance, I wasn't able to find that out that information. It was only when we had released that were they able to get back to us and tell us, Oh, actually no, we do have much more stringent data security policies and controls in place.

And I think yeah, d one, A, that's great, but you know, like those might have been it might have been uh I I think the the the thing that we wanted to bring out was That these should be more accessible, they should be more well known. The teachers usually, you know, like are considering a look one of the number one risks uh when we talk to teachers. Uh one of the number one concerns is what happens to our student data.

you know, like concerns around safeguarding, uh around privacy. So I think these things, you know, developers especially, I think they should be it's it's definitely a lot of them are good about it, but I think there's there's definitely Uh a lot of gaps there still. I think Also in terms of for example, ethical bias, there's there's this understanding a lot of these tools again, since they're using open AI models in the back end.

And they might be really good in terms of protecting student data or using even a closed version of that model so data doesn't get shared. or data isn't trained, uh data isn't used to train these models, it still does come with its own inherent set of biases. And the question is, have those, you know, have the developers of these tools, have they considered what those biases are or how have they accounted for them?

And if, for example, they've they fine tune these models and usually a lot of them have, they've added uh a lot of safeguarding controls in place. But if they fine tune them using their own data or they're again like doing some kind of you know, they're using your their own data as some sort of context for these models. Uh are is there any bias in their data or is there any, you know, like gaps in the data that they should be considering? Have they done any kind of

you know, controls for or checks for uh efficacy or have they checked, you know, how uh different demographics of students uh, you know, like use their tool and whether there's a difference and you know, like whether there's any inherent bias of that. So I think a lot of these questions are Uh one, it's a lot of work. Um so b but two, you know, not all of these are something that are that are usually considered when developing a tech tools, or at least, you know, like from my experience.

That was one of the gaps. And that was one of the things that we wanted to bring out with us.

And coming back to the point, I think the tool, like initially I remember when we were developing this, the the risks themselves were very we had just one category of risks where we were like, no, we we need to dis we need to separate the data risk. Which is, you know, like where the data is going, how it's being stored. Uh under what, you know, regulation are there are the regulations followed uh regarding the data security and privacy.

We needed to separate that from the ethical risk. Uh and the ethical risk also by the way includes the whether, you know, like how pedagogically aligned or whether it considers you know, how it considers the students Socioeconomic or educational background, all those kinds of things. And then we also thought, you know, like

a lot of these tools should all you know, like they have some degree of innovation. We should also highlight that. The the point isn't to kind of bring down these tools, is to highlight that, you know, oh, these tools are They should be used, but these are some of the things that that you should be considering. In a lot of cases actually.

We've been using these tools for different parts of our work. We've used we've had clients who've used these tools. So we we definitely do uh want to encourage their use. Uh in a lot of cases there is very simple things that you can do to mitigate the risk, to h you know greatly mitigate the risk that you have with using these tools.

And I think uh also one of the things that I do enjoy doing when we review these tools is actually, you know, like use them as a teacher would use them or, you know, to test out all the new features and that's why we added the the staff notes kind of stuff, you know, like it's just What was actually, you know, uh useful to us besides all of the, you know, what it does, what it doesn't do, just things that we found interesting.

So yeah, you know, like just developing it, just using this it's it's it's always it's it's been an interesting exercise and always bring me back to the yeah tutor that we developed and you know, like how again the How could we could you know, like I I realize now that if we had if someone had done a risk assessment for us back then, there there were so many things that we could have done better, we could have done different.

that element of doing the work in order to kind of um, you know, not strong arm, but sort of uh to kind of challenge the developer and, you know, the their literature around the tool and the ways that the tool works. to make it uh sort of more accessible to, you know, a non technical audience, uh, an audience that isn't necessarily, you know, ha has no background in data science, I think is

is kind of really important. You know, it's it's the same as somebody, uh a a user of of one of our courses, you know, messaging me and saying, um, you know, the the user experience on the website is is not But like I can't find what I've just paid for. Um and so that challenge to to the developers, I think is is kind of crucial. And of course, you know, we don't want to be in a position where we're irritating

uh developers or kind of um you know smearing them. That's not the purpose of it, but it enhance that kind of mindset uh for people that they need to be i i in a sort of a a a risk and ethics and and a sort of a data mindset when they're using uh these tools. Um which kind of leads me really to the the second question, the the end goal for pushing this risk and this kind of safety mindset, this proactive. this idea that that you know educators you know naturally have about what are the sort of

you know, actionable things that they can do with this uh, you know, with a tool, you know, be be aware of all of the or at least, you know, a a majority of uh any of the risks uh involved. So I It seems like we're trying to raise the sophistication uh of the conversation around evidence and and risk.

many of the people who who might benefit from AI tools might not have the time to exhaustively, you know, hunt out all the little details of of a of a learning tool, of its policies, of the literature and interpret it themselves. It's it's just not something that they necessarily have the time for. So I I think it's easy for us to say that we're doing like a sort of a public service.

and for developers to kind of look at this at this sort of take and think, well, you're just kind of pulling apart our tools.

But the fact is, is that the audience that we're trying to reach does not necessarily have the capacity or the kind of the the the the sort of the data and AI literacy. I think one of the key points for me is that one of the issues is that that ultimately educators have got the responsibility for their that what's being used in their classrooms to be safe, but don't necessarily have the power to

you know, inact change in you know, they can't teachers can't set the regulations, teachers can't hold tools to account in the way that um perhaps they they may be in the future. Um and but innovation is running at a far greater pace than regulation is keeping up with. And then in education kind of There's a typical kind of slow uptake in technology to be contending with as well. So you're getting this kind of really spread out map of.

Teachers that are flying with it, teachers who have never heard of it, teachers who won't use it'cause they, you know, oppose the idea of AI, teachers who um haven't yet started their journey but are quite interested. Th but then when the tools are getting used, they they have to be safe and they have to be kind of that you have to look at those risks before using them in the classroom. And at the moment there's not any kind of

government line on on this is a safe tool, this is how we're going to do this. It is on the teachers themselves to, if they want to use these tools, to understand the risks that they're kind of putting out, putting into their classroom and mitigate against them. Um, it was one of the things that came out of the the recent DFE paper on genitive AI, that there's kind of there's a call out for teachers for guidance on what is safe practice, what is

best practice, how do we use these tools effectively? And one of the things I've noticed is that I think teachers are doing that. instinctively anyway, in the way they are engaging and using AI. So the like highest use case of of AI is kind of teacher workload reduction, uh generative AI is teacher workload reduction. And that I think is a way the teachers are already naturally mitigating the risks. They're kind of keeping the tool one step removed from the learner.

There was a recent um teacher tap poll that said, I'm just gonna make sure I get the actual numbers right on it. That 42% of teachers said that they had used generative AI in their work, but only 7% had used it in a lesson. like even amongst teachers who are using it, it's a really small percentage of teachers who are willing or happy to be putting these tools in front of their

learners. So I I think teachers understand, well they have to understand risk. It's it's part of our job. We we need to be constantly thinking about safeguarding concerns, worries in in using tech tools or any kind of interventions with with learners. And I think teachers are naturally very good at it. But then there's there's extra depth, as you say, for teachers who don't have

you know, data science backgrounds don't understand the way these things are working because it's not, you know, it's it's not it's not on them to learn that. It's not their their teachers shouldn't have to be data scientists to use these tools. But they should there should be reassurance for teachers that these tools have been checked for their their safety.

Uh it it sort of makes me um wonder there's there's that sort of component of the fact that, you know, E VR and Rose ha have been involved with um the ed tech sector for

uh many, many years. Um, you know, we i in uh you know our time at sort of University College London, we had uh this EdTech Accelerator, many of the startups that came onto the program uh that were in the cults had um educationalists you know as part of the sort of you know the advisory team or the board team some of the the development teams and it's just it's that sort of Crucial thing of trying to accommodate the audience, but also

Being in another sphere where they're sort of creating the technology and the dialogue uh between the two can. uh can be sort of full of tension because the development of the tool and the use in the classroom aren't actually as aligned as uh as they need to be, um because the the sort of the interests are slightly different Um which isn't to say that, you know that that startups and and um

you know, ed tech companies and sort of AI learning tool companies are working in opposition to them. They're not, but it's the the s the spheres uh that they come from um can just mean that sometimes they're speaking slightly different languages.

uh um, you know, e even though one would think that they that they wouldn't be. Ibrahim, from from your um perspective, you know, you you said earlier about the whole sort of um your time working uh with the um you know, with with your previous uh company that that was, you know, your own before the time of LLM's, you know, what would you what would you have felt like if somebody had done a a risk assessment on uh on the technology that you'd been, you know, working on?

As as working as a developer, you know, on that, would you have thought, well, you know, we need to sort of throw out the plans that we've that we've made for this technology? Or do you think you would have adapted to it. I mean, what what's your perspective from that angle? That's a very interesting question. I think maybe an eager reaction would have been like, Oh, these guys don't know anything what it'd like to develop a tool. But I think uh

I think it would have been really helpful. I think There are when we talk you know, in our experience with um as part of also the you know, like working with the UCL and the accelerator and everything, working with a lot of ed tech companies in the past, um And you know, like as as we as we've done research, working even with our clients

we realize that AI in education is still, you know, like it's a very nascent field. And there's a lot of especially with the introduction of L L Ms, there are definitely a lot of things that there are a lot of gaps, there are a lot of blind spots uh that are only, you know, like now in the hype that were initially lost and are only now being considered. And I don't think necessarily that all developers

you know, would have you know, would have thought about those those gaps too deeply or would have had the capacity to address these. And these are not necessarily to you know, again, we're not trying to pull these tools you know apart, pull them down, say that they aren't worth using. It's just that realizing that there are a lot of things that we need to consider uh you know, as we develop these tools, especially in terms of I think that we would like to see

evidence of efficacy, uh, you know, or any kinds of, you know, like this is something that more and more tools are now doing. To some extent it happens kind of in the wild as teachers use it. Uh but you know, teachers aren't necessarily test beds for for this kind of technology. You know, like what is what is the evidence that we have that in terms of both the technology and in terms of the pedagogy, these tools are actually

delivering the outcomes that we expect them to. And I think while it doesn't seem like outwardly that these the you know the risk assessments since it's it's since it's a risk assessment. It wouldn't deliver that. But that was one of the one of the guiding principles or but that was one of the things that we were definitely considering.

all these gaps around yeah, yeah, evidence of efficacy around what kind of decisions are you are you are you leaving to the L L M? Is it making some decisions that are that are gonna be you know, like really impactful, you know, the ethical framework that you're using. Uh, you know, like a lot of a lot of companies don't don't consider the, you know, like well or or they do, but they they don't necessarily have it

have it written out. So I think for us, you know, maybe after the initial knee jerk reaction of of of you know, maybe the negative reaction, it would have been I I I I would have liked to think that we would have taken it positively and it would have opened our eyes to all of the things uh the you know, the vast amount of things that need to be considered or that need to be taken that you need to take account for

uh when developing uh such an app because it is you know, even in the A EU AI Act, uh I think education or one of you know, like this these categories one of the high risk Or the extreme risk categories, I'll I'll have to double check. And there's a there's a good reason for it. That implies that there there's definitely much more controls, there's definitely a lot more considerations.

uh that we should have in place when developing AI tools for these kinds of use cases. And the goal eventually ultimately is is to bring to light all these use cases, all those considerations, all those gaps that might have been missed, really.

It brings the the conversation I I think sort of neatly round to to the third question. With the advent of of sort of emerging technologies Um, you know, we had sort of XR, you know, augmented reality, virtual reality. There is a danger that Many people outside of the learning environment consider schools, consider the learning environment a test bed. for uh tech interventions and uh IP uh at its core. Um and

Why do we allow it to be the case that that sort of it it it becomes the practitioner's responsibility to make these kinds of evaluations? You know, even with the aid of risk assessments, with the aid of sort of frameworks, why is it the case that you know, it it becomes the teacher's responsibility. I know that they're front-facing, you know, they're the ones um who are, you know, uh who are sort of dealing with um you know the learning environment on a daily basis.

Um but uh why is the tech and AI sector, you know, not capable of of entirely regulating their own practices, you know, like you know, c can't the market sort itself out? But the fact is that the market is is not necessarily it's not wet necessarily within the market's interests to uh to perform this way. So the question becomes It becomes the school's responsibility and the burden is on the teacher. Um, you know, is it going to help if?

uh organisations, you know, the the developers of the tools are um you know able to sort of uh make arrangements with the schools. I mean, uh Dave, what what do you sort of feel um is is a kind of a way forward? Um Yeah. Um I mean I I don't think I can speak to why um companies themselves can't or won't do this work internally, but I can kind of speak to the

the issues with using schools as test beds or or even any kind of kind of evaluative system. J if thinking about kind of ed tech more broadly and and drawing on my experience of um of teaching. I well again go back to you mentioned um augmented reality and VR headsets. I know that I I was part of the purchasing process of choosing and

and and getting hold of some VR headsets for my previous school. They got used by me extensively, and then I left the school. And I am 99% sure they are in a cupboard and have not been gotten out since. So and I was trying to evaluate the use of them and what the impact was, but I'm one person with one class.

that's not enough information to to get any kind of relevant data on whether these have done what I wanted them to do, whether they've been an investment that's worthwhile. And as far as the company are concerned, I've bought that already. The sale is done. I don't need to

they they don't need to engage with me any further than that really, other than maybe I'll buy some new ones if they make some more down the line. But they weren't in the business of checking that I was using them effectively. They weren't in the business of of making sure that their product was doing

what what they wanted it to do, and they were in the business of selling me the product. And that's that's their prerogative as a company. And it was it kind of then fell to me to try and do that myself. And you're teaching. You you haven't got the the space to do any in-depth research or evaluative tools. You can you can do your best. And the more teachers you can get on board with you doing it as well, you can generate your own kind of

internal kind of information. But then I was finding myself reaching out to further groups of teachers to try and get more people that had used them and getting their experiences. Um but that's so time consuming'cause you're making connections, you're getting all this information. So

Schools just they're naturally the way they work, they're not a good place to test products in because they're they're busy doing what they need to be doing, which is which is keeping kids safe and look and and teaching.

You know, that's their core core use. They're not bases that are they're built and designed to to evaluate tools that are being used. And there's another complexity as well, and that is that There's a huge well, there's a growing field of research called implementation research that demonstrates that a piece of technology that works well in one place won't work well in another place.

So the tech company you know, so that so and yeah, as you say, Dave, the schools don't have the time or the resources. So there has to be a better way forward. But it is It's not obvious. And I worry about when it comes to AI, I worry about the tech companies being allowed to regulate themselves because they won't. Unless it's heavily policed. They won't.

It's too risky. I actually think the education sector has to get much more involved in the regulation, but it can't be teachers in classrooms, they haven't got the time. You know, somehow that you know, there has to be whether it's bodies, I I don't know, but if we're not careful, what we end up with in terms of regulation.

won't really meet the needs of the education sector. Ibrahim, like it w with more of a c your kind of sort of technical hat on, you know, do doing the sort of the the science. I I assume that you feel that there is not resentment, but a little bit of tension in, you know, you're thinking, you know, why are practitioners

Why can't they take, you know, just a little bit of extra time to kind of work out the intri the intricacies of of this tool and and so on? I mean, how what what kind of advice I suppose would you s would you give to, you know, uh a a developer, you know, a company, maybe even know, big tech as as you know, we touch on the issue of regulation, to kind of manage their expectations as to how they sort of interact with, you know, schools that that they're you know trying to get their tool into.

I think that's definitely one of the false assumptions that we had was that you know, like uh, oh, we'll just develop this tool and we'll go to schools and teachers and kids will use it and you know, we'll fix it on the fly, all the issues will sort themselves out and we'll do it and you know, it'll be great, it'll be a learning experience for everyone and Uh but it you know, in the wild it often

More times than not, I mean, very rarely does it work out that way. Rarely ever is a tool so well developed. Uh there's you know, like there's always a lot of friction. Um, especially, you know, with teachers already being so overworked. and already juggling so many responsibilities, you can't really expect them to conduct an evaluation a whole evaluation of your tool and for you to, you know, like give you helpful feedback that helps you improve it. I think One part of

the the the lack of regulation is, you know, generally the whole field is is is so new that there's a lack of regulation across AI altogether. And I think with it being So new there was there's definitely this period of with LLMs at least. I mean, yeah, it's not new, but with LLMs, I think there was this. you know, gold rush with everyone just trying to get out the next tool as fast as possible to not lose the race.

Um and I think that in part is to at least in my opinion, I might be wrong, that in part is to blame for for the lack of regulation for this kind of approach. I think though, as um or at least I hope is as as the kind of the hype dies down and you know like they are already they've been, you know, like incidents we've seen with, you know, like uh

AI being used to to grade examination papers and the and the and the backlash room that I think eventually you will have, you will arrive, hopefully. Developers realize that you know, it's not really worth developing these tools until you have a very, you know, if you've conducted an evaluation, you've conducted, you know, you've really thought about uh how effective this tool is, how it's going to be used in the classroom, minimize that friction when it comes to using it.

co-design with teachers, with students. And again, you know, like just, you know, like have not it be their responsibility to evaluate the tool for you. Um, how that happens, I mean, I definitely you know, like with with with my experience and uh the experience that E VR has had with, you know, like working with ad tech companies and our clients, that's definitely something we've seen. And that's definitely something that we've helped a lot of companies do.

Um and I think it's it's something that is catching on, but but yeah, I think I think part of it is is just the the kind of the the the the hype, you know, and this rush to to get out the next you know, killer LLM app uh out there for education. And when you look at the vast amounts of money invested, you just yeah, there's so much pressure, I think, to to do exactly what you said, Ibrahim, to develop that killer art to to

you know, give investors some comfort that they might actually get a return. I think there's also um a There's a difference between products that are being or tools that are being developed specifically with education in mind and tools that are just tools that are being adopted because they are widely available and are um, you know for all purposes for um

that they there will be and should be different levels of regulation. But as an educator you then need to be aware of that and think, right, so this tool isn't actually hasn't been designed from the bottom, you know, hasn't been thinking about me. And that adds adds an extra layer of risk inherently. Um, and that might not be something that people are thinking about when they're engaging with using AI. Yeah. Yeah. I think that's true.

In the companion piece to today's episode, I'll be sitting down with Rajashwari Ayer, CEO and co-founder of Synaptic. Synaptic uses AI technology to deliver real-time personalized examiner quality feedback to free text answers. And I'll be in conversation with her, asking who needs a risk assessment for a learning tool and why?

What the goal of the risk assessment is? Why push for a safety mindset? And why is evidence of efficacy so important, especially in the age of AI tools? Catch that episode directly after this one. Dedicated to helping education organisations. Visit educateventures.com.