Rise of AI Conference 2019 · Berlin

How AI Adaptive Technology Can Upgrade Education

KP Thai · Squirrel AI Learning

At Rise of AI 2019 in Berlin, KP Thai, senior learning and data scientist at Squirrel AI Learning, argues that AI can redefine education rather than just make it more efficient. Using the SAMR model, he distinguishes substitution and augmentation, which merely enhance existing teaching, from modification and redefinition, where AI changes the paradigm itself by moving students from passive to active learning and shifting teachers from the 'sage on the stage' to a coaching role supported by an AI 'assistant coach'. He describes Squirrel AI's after-school, business-to-consumer model in China, built on granular knowledge components, a probabilistic model of what each student knows, and a virtual personal assistant that uses natural language dialogue to diagnose misconceptions behind wrong answers. He reports that external randomized controlled trials across more than 18 studies show students using Squirrel AI outperform peers in traditional instruction, including some taught at favorable teacher-to-student ratios, learning the same content faster. Founded in 2014, the company operates over 1,700 learning centers in 200 Chinese cities and is expanding internationally through research partnerships.

Key takeaways

  • KP Thai uses the SAMR model to argue AI belongs at the 'redefinition' level, changing the education paradigm rather than just making existing teaching more efficient.
  • Squirrel AI reframes teachers as 'head coaches' supported by AI 'assistant coaches' rather than replacing them.
  • The system diagnoses individual knowledge gaps using granular 'knowledge components' mapped as a prerequisite network, since mastery time varies by student and by topic, not by a fixed learner speed.
  • A virtual personal assistant uses natural-language dialogue after wrong answers to diagnose the underlying misconception, not just that the answer was wrong.
  • Across more than 18 external randomized controlled trials, students using Squirrel AI outperformed peers in traditional instruction, including some taught at favorable teacher-to-student ratios.

Detailed notes

From substitution to redefinition: the SAMR model

Thai uses the SAMR model to frame four levels at which technology can affect education. At the bottom, substitution merely replaces a teacher's activity, such as MOOCs like Coursera or edX offering the same lecture via video. Augmentation adds efficiency, such as automatic grading giving instant feedback. Modification changes the task itself, as in the flipped classroom, where students watch lectures as homework and do hands-on problem-solving in class with the teacher's guidance. Only the top level, redefinition, is where he places AI: it does not just change tasks but changes the paradigm, moving education from passive learning, where all students receive the same content at the same pace, to active learning, where students generate and search for answers themselves.

From sage on the stage to guide on the side

Thai argues group-based instruction wastes time because every student learns differently, at different speeds, with different prior knowledge and goals. He proposes reframing the teacher's role using a sports-team analogy: the human teacher becomes a head coach who strategizes what students need to learn, while AI-based tutors act as assistant coaches providing one-on-one attention. Freed from delivering content, teachers can focus on social-emotional support and collaborative group work rather than memorization drills. He stresses this does not remove teachers but changes their function, combining a human head coach with an AI assistant, plus continuous, contextual, actionable analytics rather than data collected only via a final exam.

What implementing AI in education requires

Thai lists four requirements: AI expertise, from elite researchers to engineers who can turn models into products; mature AI models and algorithms able to solve real-world problems; sufficient quantity, quality and ease of experimentation with data; and adequate computing infrastructure. He says a business-to-business model selling directly to schools does not work well, so Squirrel AI sells business-to-consumer through after-school learning centers in China. He warns against a one-size-fits-all algorithm: a model tuned for math does not transfer to reading, so implementation must be built from the specific content students need to learn, using comprehensive, contextual, historical data about each learner rather than single-course, one-dimensional data.

How the adaptive system works

Squirrel AI breaks subject content into very granular 'knowledge components' mapped as a prerequisite network, so the system can diagnose exactly what a given student does and does not know, for example a student needing to master only another 6 percent of a topic. Time required to master a component varies by student and by component, not by a fixed notion of a 'fast' or 'slow' learner, which Thai says justifies personalized content sequencing. A virtual personal assistant uses natural language processing to start a dialogue when a student answers incorrectly, asking them to verbalize their reasoning so the system can identify underlying misconceptions rather than only knowing an answer was wrong. The company also experiments with facial-expression capture, transdermal imaging and EEG headsets to gauge stress, attention and focus, though Thai says usefulness of these signals is still being researched.

Q&A highlights

Asked how AI can adapt to a student quickly enough, Thai says the system does not retrain in real time but iterates on a roughly weekly cycle, collecting performance data on how students interact with content and feeding it back into model updates. A neurotechnology company representative working on stroke rehabilitation notes parallels with their own EEG-based work and suggests possible collaboration. Asked how the system avoids giving students a one-sided experience that omits things like personal anecdotes a human teacher could offer, Thai says content still includes videos and varied question types, and that human teachers remain involved to supplement, sometimes replace, and inform the AI.

Quotes

Group based instruction is a waste of time, so instead what we should be doing is use AI based technologies to continuously adapt one on one.

KP Thai

Questions answered in this talk

How does Squirrel AI personalize learning for each student?

According to KP Thai at Rise of AI 2019, Squirrel AI breaks content into granular knowledge components mapped as a prerequisite network, diagnoses exactly what each student knows or does not know, and recommends personalized content and sequencing rather than group-based instruction.

Does AI-adaptive learning replace teachers?

According to KP Thai at Rise of AI 2019, no: AI acts as an 'assistant coach' handling one-on-one content delivery, while the human teacher becomes a 'head coach' focused on social-emotional support and collaborative work.

Is there evidence that AI-adaptive learning improves outcomes?

According to KP Thai at Rise of AI 2019, over 18 external randomized controlled trials found students using Squirrel AI's adaptive learning outperformed peers receiving traditional instruction, learning the same content faster and with better engagement.

Mentioned

SAMR model, adaptive learning, knowledge components, probabilistic knowledge state model, virtual personal assistant, outer loop adaptivity, inner loop adaptivity, EEG-based affective computing, randomized controlled trial

Keep exploring

More from Rise of AI Conference 2019

More on Enterprise AI & Adoption

Transcript

Auto-generated transcript, not edited. · 5,721 words

Show the full transcript

so swirly learning is a a I based adaptive learning platform so we're currently only in China at the moment but we have a number of collaborations throughout the world so I'm actually part of an R&D team based in the US so today I'd like to talk a little bit about how AI can transform the education industry specifically a little bit about what we're doing and what we're is for the future in terms of how do we use AI for education purposes so my role is I'm a senior learning and data scientist as gorel AI learning basically what that means is I don't have an AI background I have a cognitive psychology background so my PhD was in in on how people learn how they think how we perceive remember and process information right and use that with data science and more recently AI so really looking at the intersection of learning science data science and AI to try to find ways to enhance education using AI to specifically personalize education so in the next 20

minutes I like to cover these four topics four points let's start out with first why AI for education so let's talk a little bit about some of the ways in which we can think about how technology can impact education so we find that this model to be really helpful this is what's called the same AR model we didn't build this but we find it to be really helpful so that at the very bottom level what you can see is technology can substitute right so technology can just substitute a teacher performing a common classroom activity and make it a little bit easier so for example Coursera EDX those MOOCs those massive open online courses they're really good at providing essentially a replacement of what the teacher would do anyway standing in front of the classroom lecturing at you via a video based platform so will you have you know much greater accessibility to that kind of content but it's exactly the same functionality it's really just a teacher lecturing at you the next level up we

can do Ockman tation where the technology can actually augment that by offering an efficient tool to perform common tasks for example a technology can provide automatic grading so you can get feedback to students immediately so they know what they did wrong or right and then you know adjust frame so that's augmentation another so at this point substituting and augmenting are essentially enhancing education not actually changing education or transforming education just yet at the next level up you get a little closer so what you're having here is that the technology can now modify the tasks that the teacher would normally do for example flipped classroom right have you heard about flipped classroom before a few of you awesome so what the what happens is that as homework the student could be watching a video or essentially you know getting a lecture beforehand as homework and then when they go into the classroom they doing practice based group work that actually get at problem-solving hands-on activities with the teacher's guide so at this point you're beginning to

get a little bit more effective so you're moving beyond just the efficiency of augmentation you're moving into the effectiveness getting better more effective and actually educating the students now at the highest level redefinition this is where we think AI comes in we can essentially redefine education we don't just change the task we change the paradigm so we essentially flip education on its head basically but we do that in two ways the first one is essentially really pointing out what is the existing paradigm which is passive learning and what do we want to do is more active so let me explain that a little bit passive learning is essentially ironically is what's happening right now I'm a teacher staying in front of a classroom essentially broadcasting the same information I'm expecting as students you're processing the exact same information in the same way at the same speed and learn you know and master the content by the time the bell rings let's just not it's just not natural that's not how humans learn right so instead

of just passively receiving the information what we actually should be doing is by having students actively generating information instead of receiving the answers students should actually be producing the answer is trying to find the answers for themselves that's how we learn and really that's how we prepare for the 21st century the next another way in which we change the paradigm is we move away from this teacher or content centric kind of view where let's take at the attention away from the teachers okay the sage on the stage kind of perspective that just doesn't work in education anymore instead what we should do is focus on the learner so move the teacher from the sage on the stage and move them turn them into a guide on the side right so the focus and it's now on the total students what is it that students need to learn everyone learns differently everyone learns at a different speed everyone has different interests goals prior knowledge right so how do we adapt and personalize to to students and

moving away from just bombarding them with content and really have them interact with the content okay so we believe in this revolution of the education method essentially we're saying every students is different so frankly group based instruction is a waste of time so instead what we should be doing is use AI based technologies to continuously adapt one on one to give students a personalized tutor at the giving them the content that they actually need at the right time in the right way right and we're not removing the teachers this is something that whenever I go to conferences I have to really truly clarify we we're not saying that the teachers is out of the picture who is simply changing their role so now the teacher is no longer the the one that's providing the providing the primary source of information instead they are now a coach think of it as like a sports team where you have a head coach that kind of you know strategize what is it that students need to learn and

then the assistant coaches are really the one that are providing the one-on-one attention so essentially what we're saying is that the AI base tutor are those assistant coaches and the human coaches are the head coach so the human coach is now are really the ones to provide targeted social emotional support and so they're the ones to engage students in the content engage them in collaborative group work so that they can actually focus on challenges rather than just memorization okay so own goal is in three three ways I realize the text is quite small for you all the way back there I hope you have a TV to see it so we're moving from the traditional lecture center to the more personalized adaptive teaching okay and at the bottom here instead of just a human instructor we're now combining a human head coach with the AI assistant teacher assistant teacher and there's the one in the middle I haven't mentioned yet is that what AI base technology we can actually provide contextual and actionable analytics we

can continuously monitor what students know and don't know and change the parent I change the way they learn continuously rather than the traditional classroom where often when you want to collect data about students who give them a test and often it's at the end like a final exam that's often too late to actually make any changes to your instruction okay so our goal then is to enhance at first to enhance the efficiency to give the time back to students so that they can master what they need to do in order to solve problems real-world problems okay and then once we solve that we're moving into how do we actually teach them to to learn how to learn to be more metacognitive least sophisticated to know how they think and monitor their own learning right so they can control their own learning those 21st century skills okay so how do we do this what is our approach one way in which we we have to kind of think about how do we implement AI how do

we make sure that the AI is actually producing value the kind of educational equity is really what we're aiming for and so the business model is truly important here where the AI is truly the the source of the disruption we start with a problem the educational problem right and it's really helpful to to support the business model if you have the social economical and foolish and political environment support and in China that's a little easier to come by so we start there of course we also need AI expertise so we need to know who and you know we need to have the elite researchers those who can actually move the field forward we need to have AI engineers and product teams who can actually turn those models those vision into products and we need those AI models and algorithms to actually be mature enough to solve a real-world problems oftentimes that's actually something that we you know we struggle with the third thing is data this is kind of obvious quantity and quality of the

data but also the ease of experimentation and this is where a lot of our research partners interested in us in that they want to be able to use our data to actually inform the models use the models to then apply collect more data right so continuous improvement but also feed forward to standardization and kind of contributing back to the field to promote more research more models the last one is some computing power we have to have the technology to support what we do so a lot of the problems that we're trying to solve we have an inkling of how to do it but sometimes the technology isn't that infrastructure may not be there the scalability that's the stack is not quite ready so that's another area that we're working on but in general this is the four areas in which we absolutely need to implement our AI okay so we're not alone and we're surely not the first I'm sure many of you have heard of at least one of these major players of the

adaptive learning fields in fact many of our employees and core leaders are actually many of them came from these companies so from them we learn quite a lot and we also had to adapt to the the China system so we learned that to b2b model the business to business model doesn't work in China frankly I don't know if it works anywhere if you've had experience in the u.s. you would know this is true it's really difficult to sell directly to schools so instead what we do is beat so directly to the consumers we are an after-school learning market and the fundamental models an algorithm still work really well but specialization is not easy by that what I mean is that sometimes a single algorithm may work for math really really well but learning how to read it's a completely different beast okay you can't expect to just slap on the same AI and expect it to work so instead we actually have to build our implementation from scratch at the core is actually the content

what is it that students need to learn how they learn it comes after okay and where AI comes in is entirely dependent on what is it that students need to learn okay often in in you know in other cases we often find that data is one-dimensional so what that means is that you often just get data from a single course where students is taking a class and you get data from how the students is doing in that class you know sometimes really really great detailed data but that's not enough because now at this point we actually want more comprehensive more contextual more historical data you need to know what did the students learn before right what did they what did they do prior to this class what are their goals what is it that they actually want to get out of this class what what are of support do the teachers the parents provide them it's really the full picture and sometimes we even get at the effective experience the motivational engagement factors that students

have as well so really in order to do this you kind of have to go all in you can't just have the AI you can't just have the content you can't just have you know like the videos or kind of the setting you actually have to provide all the things and all the things means you have to have the AI content and service and pedagogy so the way that it works for us sorry is we have students come in in an after-school setting where students are sitting situated in a small group setting okay so in front of them is a computer and this computer is the one that's providing the adaptive one-on-one instruction and the human coach is kind of always on standby and the human coach also have their own supervising system so that they can monitor what is it that students are doing and sometimes we so we're currently doing some experimentations where we can during students faces to get a sense of their facial expressions and we have a number of research

partnerships where we could look at things like transdermal imaging blood flow under the skin to see how students might be stressed what the heart rate might be mm-hmm and sometimes the expressions can tell us if the students are frustrated angry or sad or happy right not saying that those are all meaningful but that's an area of research what we collecting data we can see if how informative is that data and actually helping us to improve our learning experience sometimes we also have students wear a headset an EEG headset to measure their brainwave to get at attention and focus that's what EEG is probably the most meaningful for at this point again we don't know how useful it is yet but we're trying things out so let me give you another sense of how AI based adaptivity could work and what does it require so the first thing that is required is very very granular content to identify exactly what students know and don't know your content have to be granular enough right to be able

to actually diagnostically measure what is it that they know and don't know so this is just a you know a small very very small subset of the knowledge components that students would need to know in middle school and rational numbers okay so we're using a number of algorithms to determine where students is on this on this map so I'm laying them out here as kind of boxes in a grid form but it's actually a prerequisite network a very complex network that we're talking about here so the point that I'm trying to make is that let's say these are the things that students need to learn some students already know a lot so this particular student here only needs to know another six percent like that this is the areas in which they're weak on another student could need a little bit more and another students would need a lot more so every student is different so aside from diagnosing what students know and don't know another thing that we really need to pay attention to

is the individual differences and how they learn right and so I'm going to show you an example of on this axis right here it's just the same a random subset of students these are random subset of the knowledge components that they need to learn and on the y-axis is the number of seconds that they need to master each one of these knowledge components so take a single student student number seven the same student may require different amounts of time on different knowledge points it's not like the student is a fast learner or a slow learner it's it's entirely based on their prior knowledge they interest write their goals and any other like other relevant factors on the other hand for a single learning component different students may require different amounts of time so it's not that this particular the cube root is easy or hard but when and for whom right so those kinds of individual differences we are taking into account beautiful grid just to show you how how varied the students are but

you know we need to be able taking these into account to provide personalized content recommendation what content do we need to provide and in what way in what sequence do we provide them that is where the power of AI comes in this kind of thing could only be done with a one-on-one tutor in real life right but nobody could really afford that or at least a very few of us in the world could actually afford one-on-one tutor with a very very good instructor so what we're saying is that AI could provide the solution in a most in a much more scalable way and in a very effective way okay I want to clarify one thing is that one way to when I talk about the AI engine I don't just mean the AI algorithm in order to actually implement this solution you need the AI algorithms you need the content the the things that students need to learn how are they gonna learn it the data of course means you learn a lot from the

data and the pedagogy right so the the instructional framework in which you're going to carry out the learning and the teaching and some of the key resource accessories are just kind of laying out here we want to be able to really just do three things with AI diagnose to identifies what students know and don't know to provide recommendations of both content and sequence of content you provide them with the kinds of interaction that truly engaged them with the content rather than just you know giving it to them they need to be able to think about it in a in a productive way and productively struggle sometimes in order to learn something you have to fail and struggle and how do you do it in an engaging way so that students can kind of keep going learn from the mistakes right and then another thing is optimal learning goals we want to take into account students actually a pretty good at controlling their own learning and you give them a chance to do it right if

you tell them you know these are things you need to learn usually they're a little resistant but if you actually get them right you actually mobilize what is it that they want to learn they become a much more successful learners and in China specifically we have a lot of parental involvement so we are capitalizing on that to you know provides students with adequate parental and environmental support so as I said AI could enable those three things and these are the things that the teachers could get actionable insights in real time optimized to tint because if they get exactly what they need exactly when they need it they often stay very well engaged continuous formative assessment because we continuously collect data and inform student learning using that data and we also empower teachers to focus on developing 21st century skills in a number of ways and by 21st century skills I mean those skills that get at how students learn right how do how do they learn how to learn to control their own learning and

so I want to give you another more concrete sense of how AI can inform adaptivity so they're actually two levels the first one is something that I've already mentioned before where the a I could actually inform what the task is so what is it that students are doing now what should they do next and in what sequence so it's a task based kind of outer loop is what we call it outer loop adaptivity now within each task let's say within a single question you could actually use AI to figure out what is it that students actually need help with one way in which we're doing that right now is using something called the virtual personal assistant really and what it does is that uses natural language processing semantic analysis to engage students in a dialogue so when students get something wrong so this is just a single example of physics students are dealing with a problem a reflective light so they got an answer wrong they they picked a wrong answer and so what happens

is the VP a would now trigger a dialogue with the student trying to engage them in a conversation where we will be asking the students to kind of like can you verbalize your thought what was your thought process going into it and we can actually what we get out of this is the reason underlying the mistake okay are there possible misconceptions that students may have that we although I just could not get at if we simply giving them multiple choice questions they could get something wrong you can know that they didn't know it but you have no idea how why they didn't know it okay so we're using this kind of technology to try to get at it it still is a research area but we're making some good headway okay so these are some other research area that we're focusing on we are primarily started out with knowledge graph driven outer loop adaptivity so the task based outer loop adaptivity is what I talked about don't have time to go through all of these

but that's one area of research another one is something called contextual multimodal integrated behavior analysis that's the video captures I was talking about and viral neuro input analysis that's the brain wave measures that I was talking about so we're doing affective computing we're doing different kinds of supervised and semi-supervised learning a little bit of deep learning with those we're also doing inner loop adaptivity so I gave you a small snippet of an NLP kind of research domain but we also doing a number of other things what we have we're trying to cognitively simulate so imagine if you can simulate a student okay that actually saved quite a lot of time and being able to just kind of predict how students would behave and while at the same time really you know and inform your model of how students learn collaborative design how you group students that seems to be a pretty simple question but it's actually quite a complicated kind of thing to do should you group students based on common interest based on common

level of knowledge or very different so how well does it work a no amount of time but I can wrap this up really quickly yeah okay so how well does it work well the one way to figure that out is to do external testing so these are just a sample of four different studies that we did using third Valley third-party research collaborations so what that means is that someone's independent you know a researcher comes and do a randomized control trial which is the gold standard of research right of treatment versus control blind like random assignment and they use an external assessment so pre and post-test so everyone takes a pretest split up the groups of students into two groups treatment versus control treatments the one that gets squirrelly I learning and control is anything else usually in this case its regular traditional instruction or traditional tutoring and everyone gets a post-test so you have to do is compare pre to post and you evaluate the learning gangs so what we see here is really across

the board I'm giving a sample of four studies here but we've done over eighteen and those results are just very consistent across so students who learn what's correlating outperformed those who learned with traditional instruction so they actually have better scores and they learn the exact same content faster so it's effectiveness and efficiency and another thing is that a lot of the times they actually report better engagement as you would expect right and I want to point out one more thing is that some of these studies were done with a human teachers in a 1 to 15 ratio one teacher 15 students in some of these cases is one teacher and three students like what you would expect in a pretty decent small size you know classroom or small group tutoring environment and yet we managed to outperform those situations as well okay so this is our current distribution in China we actually only four years old we were founded in 2014 late 2014 but we're currently an over 1,700 learning centers across 200 cities in

China and growing we're looking to open 4,000 learning centers by the end of this year are we getting there so how do we make this better one way to do it is to establish collaborations with top international institutions many of them are actually in the US but you know we're open to European partnerships we are leveraging advances in learning science data science AI and and others and all those include you know content development and user research etc take advantage of standards data sharing open platform we're very open about thinking about how do we later on open source our product we're contributing very heavily towards standardization efforts of how to build an adaptive instructional system how to use federated machine learning to share data privately things like that and of course we are continuously improving on data how do we get more data higher quality data more comprehensive contextual data which is probably the most important thing continuous data driven improvement there there so this is where my research comes in where how do we think

about improvement using data so it's not just about improving the I because I know it's an AI conference we tend to think about using data to make the AI models better but it's actually that's only one area of it another part is how do we improve the content how do we improve the pedagogy we know a lot about how students learn but in practice it's not as easy to just implement you know what's often driven from research done in like an academic setting to actual real-life okay so these are just a subset of our research partners we have research labs joint lab set up at Carnegie Mellon University and SRA international at Stanford and we actually have so most of these as I said are in the US except for of course Toronto which is Canada close enough I'm gonna Sydney Australia China of course and we also are collaborating with an institute in Spain on AI okay just a quick thought on a I and M L in general so kind of very much

along with Pascal what he was saying earlier where small data is actually going to be the next big thing for us we don't we don't need big data we don't actually want very big data if we can get more contextual well labeled small data we actually learn a lot more from that deep learning we think needs to go beyond perception beyond pattern recognition towards reasoning and understanding because for us causality is what matters the most actionable AI so we we often think about AI is being able to predict or eye pattern recognizer and predict and that that's the what kind of question what is next wait we're asking so what like what insight is that get but even beyond so what now what does that mean for us how do we actually use that to actually act upon that cut those kinds of insights to actually help our students learn so thank you we generally just look for partnerships and research development standardization and philanthropy endeavors and that's my email in case you want to

reach out thank you so Teresa is coming to get you the microphone we have questions over here we have two questions which are very important let's take both of them huh sure yeah thank you very much for this very inspiring talk I have a question how can a I be quick enough to do this job be quick enough yes what do you mean by that you have to train AI to recognize maybe some patterns and you say your AI has to recognize the level of the student yes yeah we have to pick up the student where he wants to go right he's gone how can this be done in a quite reasonable amount of time it's actually being done right now and a quite reasonable amount of time because we're dealing with us kind of a smaller subset of data and we have enough students to just kind of rotate right so that we can actually apply existing models do some quick modification collect data tell us what's going on how well is it working

and then just go back and change the model and kind of reiterate so that kind of iterative cycle is what we're good at and that's where a lot of the research partnerships are interested in yes okay not that fast I'm talking about well let's say if you have a new model for addict tivity like let's say for a small subset of mathematics 7th grade algebra I don't know right you could implement something collect data over a period of days and then look at the students performance because that's what we care about the most or students progressing how are they interacting with the items are they interacting with items means questions in our case how are they interacting with those questions and are they doing so in a productive way so we have metrics set aside to kind of evaluate those things all we have to do is look at those metrics those key performance metrics right and then feet so we're not we're not doing like model making changes to the model at the millisecond

this kind of way that you're thinking about but it's it's fast enough where you know within a week we can get some data back with some some performance data back that we can inform the next iteration okay if you have two more questions one over there and then the one in the back the young guy was a bit faster but she's still there open for for Christians afterwards I'm sure okay your question please hi I'm Martin - fine presenting at the end of today's conference and there's a lot of work that you've done which is how can I say shadowed in some way the work that we're doing I work for my maze which is a neuro technology company that's working in healthcare markets and particularly with stroke we're kind of reeducate inhale people to relearn through therapy kinematic actions that they've lost in the lesion so it's a kind of education application in the therapy and some of the things that you're doing in effective computing we're doing we're working on the EEG side

so it's really interesting to see what you're doing I can see a lot of parallels which is the better than shadowing in what you're doing with what we're doing so perhaps there's an opportunity for collaboration and our absolutely yeah let's talk to you yes thank you okay no question so and now there's a little bit Christian in the back the last one unfortunately but I'm sure you have time later and please when you raise your hand ask a question how do you prevent students from having an one-sided experience because I think some some learning tasks can't be put in some she and some genetic can you say it again because we didn't hear it right I was wondering if how we do things to prevent students from having a one-sided experience because I think some tasks some skills can't be put into into a simple somatic because for example some anecdote can help you improve you understand someone experience by evening when you have a personal teacher did you understand the question how do you

give this personal kind of note to it you have only one side of experience right right so it's not it's so we're telling the content we identify what is it that you need to learn right and let's say you have the content it's not all that straightforward either we have a series of questions kind of interactions that students could be working on to interact with the content we have videos right and of course we have that human teacher to engage them whenever needed and that human teacher is actually really great not just to supplement the AI but a lot of the cases sometimes replace the AI or inform the AI that's another area of research it's when we incorporate how do we use the teachers and their expertise to actually inform better teaching a is teaching

Rise of AI 2027 · Berlin · May 2027

Be in the room for the next one.

Ten years of talks started as conversations in one curated room in Berlin. The next edition brings 300 selected leaders together in May 2027.