GEF Madrid 2024: AI and Personalised Learning

Summary

In this AI and personalized learning session, a panel of experts discusses the potential and challenges of tailoring education to individual needs. They explore the role of generative AI in creating adaptive learning experiences, the importance of understanding learners' profiles, and the balance between technology and human interaction in education. The discussion highlights the possibility of AI-driven feedback mechanisms, the need for a holistic understanding of learners, and the importance of human elements in teaching that AI cannot replace.

Takeaways

• 🧠 Personalized learning through technology aims to provide each learner with a tailored experience that considers their needs, pace, and learning style.
• ⏱ The moderator emphasizes the time constraint, highlighting that AI, despite its capabilities, must adhere to the laws of mathematics and time management during the panel discussion.
• 📚 George discusses the historical aspect of personalized learning and how generative AI presents a new approach to achieving the 'Holy Grail' of education, which is personalized learning at scale.
• 🕵️‍♂️ Alvin envisions a future where AI and wearable devices offer personalized learning experiences, potentially creating an 'apprenticeship' model that adapts to the individual learner in real-time.
• 🛠 Rafael suggests that AI can serve as the 'perfect 50% teacher', providing knowledge and answering questions but acknowledging that human teachers still play an irreplaceable role, especially in emotional and social aspects of learning.
• 🤖 Maria Isabel expresses both excitement and concern about AI's role in education, emphasizing the importance of maintaining human connection and the holistic development of students.
• 🔧 John describes a practical application of AI in providing feedback for hard-to-evaluate skills, suggesting that AI can offload cognitive load from teachers, allowing them to focus more on individual student needs.
• 🛡 Unice discusses the development of tools that work in conjunction with AI to enhance personalized learning, such as visual aids and multiple-choice question generation, catering to different learning styles.
• 🔄 George mentions the importance of student profile data in personalizing the learning experience, suggesting that integrating this data with AI can help in creating a more nuanced and effective learning environment.
• 🤝 The panelists agree that while AI has the potential to revolutionize personalized learning, it's crucial to balance technological advancements with a deep understanding of human needs and educational goals.
• 🚫 The discussion concludes with a reminder of the limitations of AI, particularly the importance of not losing sight of the human element in learning and the need for further exploration and professional development in AI for educators.

Q & A

What is the main topic of the session described in the transcript?

The main topic of the session is AI and personalized learning, discussing its current state, future potential, and the role of generative AI in education.

What is personalized learning as defined by George in the transcript?

Personalized learning, as defined by George, is the intention of providing each learner with a learning experience that reflects their needs, learning pace, and best approaches to be an effective learner, often through technology.

What is the 'Bloom's 2 Sigma difference' mentioned by George, and why is it significant in education?

Bloom's 2 Sigma difference refers to the significant educational benefit achieved through one-on-one tutoring. It signifies the goal in education to replicate the personalized attention and effectiveness of a tutor in a scalable way using technology.

How does Alvin envision the future of personalized learning with AI and wearable devices?

Alvin envisions a future where AI and wearable devices provide personalized learning experiences tailored to each individual's needs, potentially including multi-dimensional, multimodal presentations of information and real-time feedback based on the learner's responses and understanding.

What is the role of a teacher according to Maria Isabel, and how does AI impact this role?

Maria Isabel believes the role of a teacher is to build connections with students, develop curiosity, agency, and a sense of belonging. AI impacts this role by potentially taking over some aspects of personalized learning and feedback, prompting reflection on what makes a good teacher and the human element in education.

What is the potential issue with students relying on AI for learning, as highlighted by Rafael?

Rafael highlights the potential issue of conflict between what an AI tutor says and what a human professor teaches. It raises concerns about the credibility and consistency of information and the challenges of managing such conflicts in educational settings.

What is the current limitation of AI in providing personalized learning, as discussed by John?

John points out that while AI can provide initial feedback and offload cognitive load from teachers, it currently lacks the ability to nurture creativity and respond to the individual student's needs beyond knowledge dissemination.

What is the importance of understanding the human element in AI and personalized learning, as emphasized by Rafael?

Rafael emphasizes the importance of understanding what makes us human to ensure that AI does not get out of control and to identify the real benefits and problems. He suggests that we need to study humans holistically to better integrate AI into learning.

How does the panel view the potential of AI to scale personalized learning, as discussed by Unice?

Unice discusses the potential of AI to scale personalized learning by creating tools that can cater to a large number of learners, providing immediate feedback, and using visual aids to cater to different learning styles, thus overcoming the limitations of traditional classroom settings.

What is the significance of student profile data in personalized learning, and how can it be integrated with AI, as mentioned by George?

Student profile data is significant in personalized learning as it allows the system to understand the learner's needs, strengths, and weaknesses. George mentions that integrating this data with AI can help tailor responses and interventions to better suit the individual learner, enhancing the effectiveness of personalized learning.

What is the challenge of implementing AI in education, and how can it be addressed, as suggested by the panel?

The panel suggests that challenges include technical limitations, resistance to change, and the need for professional development. Addressing these challenges involves continuous exploration and adaptation, fostering a mindset open to change, and providing educators with the necessary training to effectively use AI tools.

Outlines

00:00

😀 Introduction to AI and Personalized Learning

The session begins with an introduction to the topic of AI and personalized learning, emphasizing the importance of managing time given the number of panelists. The moderator highlights the potential of AI in education but also reminds participants of its limitations, particularly in relation to the laws of mathematics. The session aims to define personalized learning and explore how generative AI presents new opportunities in this field. George is invited to provide his perspective on personalized learning, its history, and the current impact of AI on its practice.

05:02

🔮 The Future of Personalized Learning with AI

Alvin discusses the future vision of personalized learning, likening it to an apprenticeship model that has been refined over time with the help of AI and wearable technology. He envisions a future where devices can perceive and adapt to an individual's learning needs, potentially using neural signals to gauge understanding and adjust the learning experience accordingly. The conversation touches on the idea of a 'perfect teacher' that is always available, leveraging immersive technology and AI to provide lifelong learning opportunities.

10:04

🤔 The Role of Teachers in an AI-Enhanced Classroom

The panelists explore the evolving role of teachers in the context of AI integration in classrooms. Rafael suggests that AI could serve as a '50% teacher,' handling knowledge dissemination while human teachers focus on emotional and social aspects of learning. Maria Isabel reflects on the importance of human connection in teaching and the potential loss of that with AI, highlighting the need for a balanced approach that utilizes AI as a tool without replacing the human element in education.

15:07

🛠 Practical Implementations of Personalized Learning

John shares his experience with implementing AI in education, specifically mentioning a pilot project for the World Bank where AI was used to provide personalized feedback on student assignments. He discusses the potential of AI to offload the cognitive load of teachers, particularly in evaluating complex skills, and to provide initial feedback that can later be refined by human teachers. The conversation also touches on the challenges of integrating AI into existing educational structures and the need for professional development in this area.

20:07

📚 Enhancing AI with Student Profile Data

Unice discusses the development of tools that work in conjunction with large language models to enhance personalized learning. These tools can cater to different learning styles, such as visual learners, by generating graphs or diagrams. They also support the learning process by providing a chain of thought reasoning and adapting to the learner's profile, which is maintained and updated separately from the AI's context window. The goal is to create a more nuanced and personalized learning experience that can scale to large numbers of students.

25:08

🧠 The Importance of Understanding Human Beings in AI Education

Rafael raises a philosophical question about the understanding of human beings in the context of AI and education. He emphasizes the need to study humans holistically to ensure that the benefits of AI in education are real and not just assumed. The conversation suggests that there is a risk of losing control of the AI revolution if we do not understand the human element it is intended to serve. The panelists agree on the importance of balancing technological advancements with a deep understanding of human nature.

30:10

🏫 Personalized Learning in Online vs. In-Person Settings

Maria Isabel and John reflect on their experiences with personalized learning in both online and in-person settings. They discuss the benefits of online learning in reaching a large number of students with limited resources and the challenges of maintaining personal connections in a physical classroom setting. The conversation highlights the potential of AI to enhance the learning experience in both environments and the importance of exploring its use in education.

35:13

🤖 The Potential of AI in Enhancing Classroom Interaction

The moderator proposes a hypothetical scenario where AI-enhanced glasses could provide real-time insights into students' learning states, allowing for more personalized classroom interactions. Maria Isabel expresses openness to exploring such technology, while acknowledging the resistance that might come from traditional teaching methods. The conversation suggests that while technology can offer new possibilities, changing mindsets and embracing these changes is also crucial in education.

40:20

📝 The Difference Between Information and Learning

In the final moments of the session, the panelists address the distinction between access to information and the process of learning. John emphasizes that information is just data until it is combined with personal experiences and reflection to form knowledge. Ricardo adds that much of the valuable knowledge in organizations is tacit and resides in people's minds, suggesting that collaborative intelligence is key to effective decision-making and learning.

Keywords

💡Personalized Learning

Personalized learning refers to the educational approach that tailors the learning experience to meet the individual needs, pace, and learning style of each student. In the video, it is highlighted as the 'Holy Grail' in education, aiming to achieve the significant learning benefits of a one-to-one tutor relationship through technology. The script discusses how AI can potentially scale personalized learning through adaptive and responsive educational models.

💡AI (Artificial Intelligence)

AI in the context of the video is the use of technology, such as generative AI models, to enhance or enable personalized learning experiences. The script mentions that while AI is powerful, it must work within the constraints of established educational laws and practices. It also suggests that AI can help in creating a new way of personalized learning by not having to specify all learning paths, but rather adapting to the learner's needs dynamically.

💡Generative AI

Generative AI is a subset of AI that can generate new content based on learned patterns, which is relevant to the video's theme as it presents a new method for delivering personalized learning. The script suggests that generative AI, through the right prompting and data architecture, can offer a more adaptive and scalable approach to personalized learning compared to traditional educational models.

💡Learning Experience

The learning experience in the video is described in relation to how content is presented to learners, including the speed, format, and depth of the material. It is a central theme in the discussion on personalized learning, as the script explores how technology can be used to create a more impactful and effective learning environment tailored to the individual.

💡Educational Technology

Educational technology encompasses the tools and systems that support learning and teaching processes. In the video, it is discussed in the context of how technology can facilitate personalized learning, such as through intelligent tutoring systems and AI-driven feedback mechanisms. The script also touches on the historical development of educational technology in the field of computer science.

💡Immersive Technology

Immersive technology in the video refers to advanced devices like wearable glasses that can perceive and present information in a multi-dimensional, multimodal way. The script suggests a future where such technology could provide real-time, personalized feedback and adjust the learning experience based on the learner's reactions and understanding.

💡Knowledge

In the video, knowledge is discussed as a fundamental component of learning that AI can help disseminate. The script mentions the potential of AI to serve as an 'unlimited teacher,' providing information on various subjects and facilitating a more interactive learning process through questioning and feedback.

💡Learning Analytics

Learning analytics is the measurement, collection, analysis, and reporting of data about learners and their contexts, which is used to understand and improve learning experiences. The video script discusses how learning analytics can identify key variables in a learner's profile, such as cognitive and metacognitive attributes, to inform and enhance personalized learning approaches.

💡Curriculum

The curriculum in the video is the planned content and learning experiences of an educational course. The script suggests that the development of a curriculum should reflect on the possibilities offered by AI and personalized learning, and how it can be adapted to prepare students for future job requirements.

💡Teaching Assistant

A teaching assistant in the video is a person who supports teaching activities, such as answering student questions. The script contrasts the traditional role of a teaching assistant with the potential of AI to provide personalized feedback and support to a large number of learners, highlighting the scalability of AI in education.

💡Context Window

The context window in the video refers to the amount of prior conversation or data that an AI model uses to inform its responses. The script discusses the limitations and potential of context windows in personalized learning, suggesting that while they may not fully capture individual learner profiles, they are part of the ongoing development in AI's ability to personalize learning experiences.

Highlights

The session emphasizes the importance of time management for panelists, highlighting the constraint of 45 minutes for nine people, equating to 5 minutes per person.

Personalized learning is defined as providing a learning experience tailored to an individual's needs, pace, and learning style, aiming to replicate the effectiveness of one-on-one tutoring.

Generative AI is presented as a new approach to personalized learning, with the potential to overcome scalability issues faced by previous technologies.

The discussion acknowledges the challenge of creating data models and interventions for personalized learning, given the difficulty in assessing student capabilities.

Alvin envisions a future where wearable devices and AI provide personalized learning experiences, adjusting content delivery based on real-time learner feedback.

Rafael suggests that AI could serve as a 'perfect 50% teacher', providing knowledge but acknowledging the limitations in understanding student emotions and reinforcement needs.

Maria Isabel raises concerns about the emotional connection in teaching, questioning how AI can replicate the human element of building relationships with students.

John shares his experience using AI to provide personalized feedback on student assignments, demonstrating a practical application of AI in education.

Unice discusses the development of tools that augment AI with additional software to cater to different learning styles, such as visual learners.

George emphasizes the importance of student profile data in personalizing the learning experience and the challenges of maintaining stateless interactions with AI.

The debate on the effectiveness of context windows in AI for personalization is highlighted, with differing opinions on their utility in learning.

Rafael stresses the importance of understanding human beings as the raw material for AI, advocating for a holistic study of humanity to ensure ethical AI development.

The potential of AI to democratize education by impacting millions of learners with minimal resources is underscored, especially in areas with limited educational infrastructure.

Maria Isabel discusses the ease of personalizing learning in a physical classroom setting due to the ability to build relationships and understand student interests over time.

The session concludes with a call to action for educational institutions to prepare for and adapt to the future of AI in personalized learning.

A provocative question is raised about the difference between access to information and actual learning, prompting a discussion on the role of AI in knowledge creation.

The importance of 'know-how' as a key source of knowledge that is difficult to codify in AI is mentioned, suggesting limitations in AI's ability to capture tacit knowledge.

Transcripts

Ai and personalized learning welcome

everyone to this session of AI and

personalized learning uh so we have a

distinguished set of panelists uh here

uh but it is also true that we have many

panelists so if they introduce

themselves we will eat into uh the time

that we have to learn from their wisdom

so uh I'm going to get right into it uh

I do want to remind uh the panelist that

AI is extraordinary but it hasn't defied

and eliminated the laws of mathematics

so 45 minutes nine people means 5

minutes per person so please keep that

in mind as we uh proceed so I want to

start with defining terms uh so I want

to uh ask uh what is uh personalized uh

learning uh and maybe George since you

have studied this for many many years uh

we know that it isn't new uh but maybe

the vehicle or at least the way in which

uh generative AI Works uh presents a new

uh way of doing it so if you could start

don't give us a whole history but if you

could start telling us what personalized

learning is and then uh why why is it

different now sure um so I'd say first

of all uh personalized learning through

technology is the intention of providing

each learner with a learning experience

that reflects her needs her pace of

learning uh her

best approaches to being an impactful

and effective learner and that typically

is a function of how quickly content is

presented the format in which content is

presented the depth at which content is

presented and so the intent and this has

long been a Holy Grail in the education

sector is personalized learning that

allows the a technology system to

achieve what's often referenced as

blooms 2 Sigma difference which is the

benefit of onetoone relationship between

a qualified capable tutor and between

the individual learner themselves so

that's a quick overview the definition

it's had a long history in computer

science as a domain CMU has been one of

the more active places everything from

intelligent tutoring systems to a range

of different models around how we might

guide and support students unfortunately

it typically doesn't scale it's very

intensive in terms of time and effort to

create the right First Data models and

then the right approach to uh trigger

interventions that the skill set of an

individual student and very hard to

assess the capability of a student so I

think that's where generative AI starts

to come in with some promise because

with the right uh prompting context

window and I'd argue data architecture

of a learner profile fed into those llm

interactions uh there's a prospect that

we can maybe get closer than we have

been in the past all right and just so

that I understand well the previous

Technologies were trying to personalize

learning by mapping a whole tree of if

you do this you do that and the promise

of generative AI is that you don't have

to specify all the paths that the

technology can help you with that it

roughly that's a so it's it's in some

ways a bit similar to uh if you're

familiar with the uh argument that Gary

Marcus has made for a while you know the

idea of uh symbolic AI which requires

heavy processing heavy articulation and

definition in advance versus neural

networks which or more closely mimic a

non-rule based human neural art

great thank you thank you so much George

all right so U now let's try to picture

what personalized learning looks like

I'm looking at you Alvin you you have a

you might have a vision of how this

might look like in five or 10 years uh

can you tell us um what what does that

experience actually looks like yeah so I

think beyond what Bloom did which was

essentially a one toone tutor uh I think

that concept isn't even you know that 10

20 years ago what we're talking about is

actually apprenticeship is something

that's been around for thousands of

years that's actually how we've been

learning long before we had universities

and that's what essentially one toone

learning is that that's what

personalized learning is you spend not a

class but years with somebody and to

understand the the the learning model

the habits the the the the skill sets of

that individual and then to change the

way you teach with it and I think that's

what we're coming to today is that with

with AI with with wearable devices we

will have an ability for the for there

not just to be neural network training

but for it to be personalized to every

individual because like the glasses I'm

wearing today it has microphone it has

CL uh cameras it can actually perceive

everything that I perceive and take that

information to help change the way uh

that the information presented to me in

fact um if you know in another year or

so this kind of a glasses there an

immersive uh device can can get to this

type of glasses it will also be able to

present information to me in a

multi-dimensional multimodal way and

that way it can actually change the way

that the materials presented to give me

the the most uh immersive way to get

that point across and it in fact if if

at some point we may even get some

neural signals from it to know am I

confused am I learning am I actually

understanding it and to adjust the

difficulty uh of that content so you

know all of this is coming so the the

the combination of immersive technology

with AI with sensing information is

going to allow us to have the perfect

teacher that is with you 24/7 not just

for kids but actually for all ages so

you have lifelong learning and and I

think that's what we really need to get

to is to allow every single individual

in the world to have a perfect tutor um

that is all knowing and takes into that

perspective that what do you need to

know at the time I don't know how many

people have read the book um Diamond age

U young ladies primer Neil Stevenson

anyway and there there the idea is that

Nell the the protagonist has a a tablet

that goes with her as she grows up from

a child to an adult and it it it keeps

teaching her lessons as she goes through

the story and I think that's what we are

going to have is because you know these

kind of glasses you're wearing

essentially all your waking moments and

it will be able to sense all the

information you have and it will be able

to give you realtime feedback in terms

of teaching you based on your

experiences and and that's what I think

the the the kind of ultimate potential

of personalized learning is all right

I'm going to do a followup in 10 seconds

you need to answer yes does that mean no

teacher human teacher in the

classroom um I I think that changes the

role of the teacher the role of the

teacher today is to go and give you a

bunch of facts put it on to board and

say okay 10 seconds are done all right

we're moving all right Rafael what's

your vision for how personalized

learning looks like in an actual setting

building on what you said for me is

going to be first of all for let's think

about what AI can do in the next two or

four months because in six months we

don't know okay so for me what they can

provide us of today is the perfect 50%

teacher okay I think there are plenty of

sin that never are going to be able to

do it Ai No AI is never going to be able

to understand if he needs a a positive

or enthusiastic reinforcement feedback

we're very far of that but in terms of

knowledge okay what for me is incredible

and it's in a fantastic opportunity is

to have an unlimited teacher for you

take economy no because you move to

depends of the of the subject the the

topic no but you take economies you can

put as of today everything that we know

about economy and have unlimited teacher

for each one of the students okay that

will create a lot of problems because

imagine when you have your professor and

when the and the student says no no my

my my p is saying another thing you know

so can create a lot of conflict but the

potential for 50% is immediate and is

there and this opportunity to have

unlimited interaction is fantastic

people from my experience learn when

they are searching not what they are

listening okay so and knowledge calls

for knowledge yeah so you can ask

question and questions and question and

you get the people to enter on that

Dynamic I think this is unmatchable but

again we going to have to manage the

confli and the yeah that's a powerful

Place CU When you're learning with an AI

tool you're asking questions which is

kind of much better than the alternative

and you have to be I mean from what we

have learned in the history in science

everything that can be done is done okay

yes here I think that the challenge for

educational institution is at what level

they able to control you know the

implementation or adoption of these type

of Technologies no I mean if you provide

for example a CH specialized on economy

you must be very confident that very

well curated information of course you

can even trust him on top of the

professor and you say that uh when you

say 50% you you're you're essentially

describing that chpt or similar

Technologies nowaday is equivalent to

the median teacher in terms of uh skill

and knowledge is that what you're

describing you can create the super

teacher because you can have I mean

there is a limit in the amount

understanding that you can have okay and

also you can create a teacher that has

no

ideological beliefs biases which is very

important but this is for 50% there is

another 50% that has to do with

emotional oh so 50% of the job of the

job of the teacher I'm don't it so I

won't lose my job immediately is what

you're saying okay as you as you have

noticed I'm worried about that a little

bit all right Maria Isabel I want to ask

you so you're an expert in education

and I want to ask you um I mean George

described personalized learning as a

Holy Grail what why is this so important

what why do you think we should we it it

seems like every time we talk about the

benefits of AI personalized learning

comes up uh even George's presentation

this morning towards the top of the list

why is this so important well first of

all I would like

to compliment you know your words I mean

this is is super exciting I mean I think

about the possibilities and and and it

makes me happy right as an educator I

see the possibilities in the classroom

on the other hand it makes me a little

bit sad right because I wonder like why

do I wake up every morning right why do

I show up in the classroom right why do

I build that connection with my students

if you know artificial intelligence can

do that for me right so we can't forget

that piece right like of course you know

like it's all about a balance right like

we can use it as a tool but we can't

forget you know like what makes us like

good teachers and my mission right like

I get up every day in the you know every

day I go to school um and I just want

to share the possibilities with my

students at the University I mean I want

for them to be to be the best teachers

ever right I want for them to be in the

classroom to listen to the students to

develop curiosity to develop agency to

develop a sense of belonging um to

develop joy in in the students so I

wonder like are those Mach machines

going to do that for for us right so so

yeah so it's exciting but at the same

time it has to make us reflect about the

kind of curriculum that we build right

and how can we develop the best skills

for these you know kind of jobs that the

future requires great great so Rafael is

saying don't worry because for that 50%

of the job you will still wake up every

day to do it okay okay great thank you

all right so I want to maybe move to how

examples of specific ways in which

teachers can or professors can use

personalized learning I think Alvin went

to the this is coming we're all going to

wear glasses and um in a classroom of

the future I think Raphael approximated

a little bit more like what we could do

today but I want to I want to ask

specific ways in which Educators can can

do this so uh let me start with John

John you you want to describe uh so is

this L on yeah so I I think that for me

I think when we talk about personalized

learning um it's it's s of this false

promise you know this unicorn that's

always out there it's because our

schools and universities just aren't

built for it these are these are

factories and it's kind of like walking

into a Starbucks You' be like oh my gosh

I anything I want in a Starbucks right

you have how many thousands of

combinations of stuff you can get out

there but at the end of the day you

still have Starbucks right and that's

the same thing that we have within

school so we've got fixed we've got

fixed curricula fixed bodies of

knowledge fixed expertise and now I

think George was touching on this that

we're about halfway there with uh

personalized learning because we can

provide feedback and this is where I

think that that things get really

interesting because we can offload much

of the cognitive load that teachers have

in providing feedback especially for

hard measure skills and competencies

like uh are you a creative uh Visionary

can you express um can you express

entrepreneurship can you can you express

some computational thinking now for a

teacher to evaluate these things these

These are really hard to evaluate and

take a whole lot of thinking a whole lot

of cognitive load with AI we can offload

a lot of that and provides at least some

initial feedback and then ease the role

of the teacher to actually focus on more

of the individual students needs um

based on that let me push a little bit

more can you do that today we can do it

today so I did uh when you say we you

mean like you and two other people or we

like everyone in this room okay so

anybody else can do this so I did a a

quick pilot for the World Bank uh for

their evoke project uh where I I don't

know really how to program with this

stuff I'm using PHP that's not my

language but it's used in mood right uh

so I asked chat GPT I said well here's a

problem I want to connect API to API

immediate between them and all has to be

in PHP which I really don't know so

using using a chat GPT I actually built

you know 50 50 lines of code that pulled

student assignments out of Moodle

portfolio posts or written essays and

then fed them through chat GPT uh

through a query to provide uh person

personalized feedback and then it spits

it back to mood as as a is an automated

response and the results were actually

really good this s this first take um

does it does it answer everything does

it s thing no but at least students get

that that initial feedback so if you're

doing a first draft or something they're

able to work on this stuff and bring to

a teacher later on for for Fuller

evaluation and I was really impressed

but I think that's just half that's half

the way because it doesn't really

respond to students creativity so what

do you want to learn where do you want

to go right it can provide feedback on

how you're doing but in in terms of

nurturing your curiosity it's not quite

there yet we have a ways to go I'm going

to push you a little bit more all right

so um you describe a process and I'm

sure people here are wondering why

didn't the students submit their thing

to chbt and get that feedback why do you

need to write 50 lines of code and do

all of that give us the answer to that

you know they absolutely could okay but

um this is I think this is very new for

students and I think that access to

technology is not quite equally

distributed uh but I think that it's

going to get there and so it's going to

be common place uh heard many times

already today people are concerned about

students uh writing through this stuff

but I think it's be common place to use

this stuff to augment students work uh

to maybe I think we have to get away

from creating the correct response to

creating your response and I think

that's the important parts of this

especially we put in the context of

personalized learning perfect thank you

thank you so much John okay I'm going to

move move to Faron I'm going to ask the

question in Spanish but it's the same

question that I asked

John for

vide

m

on

PR

okay so now I'm going to move um to

Unice so Unice the way that uh George

described for us personalized learning

at the beginning there are many aspects

of it but uh some of the aspects have to

do with uh being adapted to the the pace

that a student wants to use in their

learning their particular interests uh

their background on the topic and so on

so nowadays with uh chat gbt or similar

models a student can learn by

personalizing to themselves by simply

typing into the chat I'm this kind of

learner I have this background on the

topic and so on you seem to have spent

some time

creating

tools that are on top I imagine of some

of these llms or maybe independent of

them um could you tell us about the

tools that you have developed and how

they play into this yeah definitely

thank you so I'll share with you a

little personal Journey how I came into

Ai and personalized learning so we

started I started teaching on campus and

when you have you know 20 students 30

students you can give them personalized

feedback and it's very manageable even

in a class of 100 students or 200

students you have you know Tas and uh

you can give them personalized feedback

So eventually we started building muks

massive open online courses and when we

had like 1,000 2,000 students it was

still you know difficult but you can

still reply you can still give each one

you know where they're stuck uh fast

forward we ended up having around 1.4

million Learners and then it becomes

extremely difficult to cater to every

individual and that's where we started

building these tools and these tools uh

they end up helping the learner first of

all uh these tools they're not simple

chat GPT they're they tend to be

augmented with tools like you can

augment them with wallframe alpha or

calculators or different software and

you can also help with Chain of Thought

reasoning so you walk them how uh the

problems are solved the other problem

we've seen that also comes uh into play

when talking about Ai and personalized

learning is 65% of Learners are visual

Learners and what that entails it means

when you use a large language model it's

very difficult to uh create graphs or

plots or diagrams that are relevant to

that question so again with these tools

you can actually cater to the other 65%

of personalized Learners doing multiple

choice generation for example uh when

you have four potential answers each

answer is usually a different track or

you know a different like wrong lead

that's identifies a different concept so

with one multiple choice uh question you

could Target you know several skill sets

and quickly Target what the next

question will be so there are a lot of

different tools and strategies when it

comes to uh implementing these AI in

personalized learning great great great

thank you thank you okay I want to come

back to George George you said uh maybe

in passing but I think I think you meant

it more than in passing that it's not

just adapting to the PACE and background

of the students and so on but you

mention um student profile data as being

an important component can you describe

for us and and uh we have reached the

stage that I was hoping we reached which

is every one of you has spoken at least

once so now I want more interaction uh

so if you want to respond to George just

raise your hand and I'll call on you but

could you describe um is that possible

today and uh if it is how to do it and

if it's not what other barriers to get

there sure so uh one of the challenges

with uh llms is they're largely

stateless and uh the difficulty is if

you want a persistent profile of a

learner that you want to attend to

specific needs you need some indication

of who that learner is because you know

if you go to chat GPT ask it a question

go back same browser two seconds later

you'll get it sometimes reasonable

variations in your responses so that

means it doesn't care about you and it

doesn't know you and the whole point of

it education is for us to be known by

the system and a teacher to personalize

or you know as Alan was saying for for a

a mentor or an apprentice to be guided

in effective way so the way we've

approached it is we've said look we're

we're going to at least short term I I

don't believe that larger context

Windows like Gemini has is the solution

in the long run so my background I've

been active in development of learning

analytics as an academic discipline and

so we spent a lot of time understanding

which sequences of learner generated

data are indicative of

attention um wandering are indicative of

learner confusion disengagement and so

we developed a model where we're looking

at six attributes of a profile so

cognitive metacognitive affect social

emotional well-being and skill sets and

so what we want to do is say each of

those levels we've identified sort of

key variables that we want to work with

and identify that says you know George

understands this concept or George

manages time well and if I don't then

we're going to through an integration of

we're largely just passing it in context

windows to the llm that says this is

George George's profile Georgia's

current state uh this gets fed here at

responds the outcome of that we're

pulling it back as a text file into an

S3 bucket it gets analyzed fed back to

the profile and so that's the model

we're saying look llms have a lot of

potential to appear to be intelligent

but when we're teaching there's much

more Nuance about a profile so that's

the mechanism we're utilizing to help

integrate and accelerate the

capabilities of the llm in that approach

great so your your platform passes on in

essence the profile information to the

llm so that the response is tailored to

that profile great thank you can I

respond a little bit um I I think it's

that essentially what you're doing is

extending the context window right

because even even with even with Gemini

totally different well yes and no I mean

if the context Windows 100,000 tokens

then you're right it's very different

but if you get to like's say long net

which is a billion tokens right then

then essentially you have a a long-term

memory of that conversation in fact if

you look what it's not the memory it's

the ability to create attributes of

profiles from data that are disconnected

from Context it's actually saying this

sequence of pattern says you are an

individual who learns in this way you've

shown cognitive deficiency in

understanding a key topic for this

reason yeah but if you have enough of a

context then you essentially it it

should be able to to grasp that type of

understanding if if it's if you're doing

uh you know interactive discussions with

the learner then it it should be able to

know whether are right or wrong or

you're doing you know quizzes in between

so I'll just make quick it should is a

very big future word I can tell you

right now I can build a profile of your

interaction data while you're engaging

in a learning process that indicates

your mind is wandering or you understood

a concept I don't need a large context

window in the big context window there's

a fair bit of controversy on whether a

context window solves that even with the

1.5 million you know size token context

window still the benefit of rag or graph

rag Technologies to make sense a local

Iz data makes a lot of sense so I think

a context window as a personalization

mechanism that's disconnected from a

profile I'm not convinced of if it

happens I'll happily change my mind but

I know what we can do today yeah all

right wait wait moderator has it right

here it's getting a little bit technical

you two go and have a cup of coffee uh

John you want to you want a rejoinder

here I was just going to say I just want

to break up that fight there say it's a

no no no no it's not a fight but they

the disagreement is at a level where 80%

of people in the room might not be

following so I just exactly but but

here's here's the thing though for all

us who don't follow right because we

talk about whether context windows work

or not what you need Etc all right

here's the thing if it doesn't work now

it will work in the near future and the

important thing for for schools

universities is that finally we have a

heads up on what the future is going to

be like we have we have a chance to

prepare we have a chance to act now

right so whether or not we can debate

whether or not what what Works what

doesn't work but we have a vision for

the future we have a vision for the

technologies that that we are developing

and we're going to get there that's it

you know every time you say we I'm like

am I part of this we or not who is we

yeah yeah you have many sleepless nights

now okay okay all right all right so uh

Rafael you want

to there's something that does it sounds

there I mean we're talking all the time

about the potential benefits and the

potential risk about artificial

intelligence in human no but I have a

something that distracts me like a major

question what do we know really about

the human beings okay because I found

interesting what the Anthropologist

start saying this morning

but where do we know where can we

find a deep

understanding of what as of today is

known not known what is debate and what

is fake about the human beings because

we're talking about the raw material

that's going to use it yeah and I my my

concern is to make sure that this

doesn't get out of control we are going

to have to come back to understanding

what human what what what makes us human

we had a talk about that this morning

and you have a startup to figure this

out I think this is a challenge we going

to have to face in the world okay I

think since and nit kill the human being

at the end of the you know 18th

century uh we have a stopped to study

human being as a whole in holistic way

okay so I cannot spend 60 years going

through the 40 degrees in Humanities and

still on top of that we'll have to study

biology Neuroscience we have to come

back a better understand the human

beings because if not we are going to

jump to the potential

benefits and we're not going to

understand what are the real benefits

and the real problems no so not

everything to make sure that this which

is absolute the revolution

doesn't get out of control going to have

to in parallel try to much better

understand human being and to be able to

convey the knowledge in a very

synthetized clear actionable way which

is something that we're not looking for

maybe because it's a huge task but I

don't think it's Huger that we have we

have done with the with the Homa no or

with other similar technology okay thank

you thank you right I want to come to

the two of you now so um you have done

both teaching and person in online but

I'm going to I'm going to assign you as

a online role and you care about this

little kids as human

beings uh and teach in person so I want

to understand the potential of

personalized learning online relative to

in person so can you can you reflect

since youve done both and then I want

you to come back marel with how do you

personalize learning with the help of AI

or maybe not uh thank you definitely so

online uh one of the you know Main

benefits uh of personalized learning is

that you don't need to have an army of T

who would go and help each indivual

teaching assistant yes a teaching

assistant thank you uh who are going to

go and you know answer each individual

learner's question and you can with one

person you can impact you know millions

of Learners and this is exactly what

we're seeing today in government that we

work with that from a very centralized

governments within the Ministry of

Education for example they could build a

simple software and then suddenly impact

7 million Learners especially where they

don't have you know uh capabilities to

improve the education system classes are

crowded uh you know student to teacher

ratios aren't as as good and this is one

of the most beautiful things about Ai

and personalized learning how you need

you know maybe a team of two or three

people at most and you can have a

massive impact and that's exactly what

we've seen online with massive open

online courses and completion rates

thank you thank you very much all right

yeah well can you can you do that in a

physical classroom yes I can I mean I

had to say that you know being in this

conversation is making me think about

how little I know about artificial

intelligence right I I haven't gotten

any kind of training and we need more

professional development around this

topic right and in person it's a little

bit easier right I have been teaching

online as well and I know the challenges

is super hard for me in person is easier

right because I have the the the the the

ability to I mean I have the possibility

to connect with my student right during

you know a whole semester I can get to

know them I can um I I learn from them

right I I know what they are interested

in right I know their passions so I can

build my curriculum around them right so

it's easier in that how many students do

you have well I have um a group of 30

people so e year exactly so we you know

we do like um large discussions you know

like a small group discussions um so

that give us the opportunity to know to

to get to know each other yeah um

artificial intelligence I have been

using it a little bit right I use it for

um the development of my own syllabus um

and also I let them mus it right right

we have to explore it um but I don't use

it in the sense that okay um use it yes

to get the result use it for the process

right of that investigation we are you

know doing an inquiry so yeah let's

explore it let's use it right let's

learn uh but um but yeah for for me you

know like I'm not focused on the result

because you know I don't care I'm in the

machine can do it for us right so I'm

focusing more on developing the skills

during that process of learning let me

ask you one question suppose yeah I'll

I'll go to you in a second so suppose

that we could transport his glasses to

you

in the glasses have more capabilities

can you imagine going from a classroom

of 30 to a classroom of 60 where maybe

it's a little bit harder to get to know

your students but when you are teaching

and you look at a student all of a

sudden there's something that tells you

something about that student that would

allow you to interact with them is that

like no way I can't handle too much

technology or how are we think about I'm

happy to explore it like everything

right like yeah the glasses you want to

give it to her okay all right yeah happy

to learn yeah for sure okay you would

explore it yeah I would explore it yeah

it's like you know would your colleagues

do it would your colleagues do it m

I don't think so I mean resistance is

there right like even to change people

how to teach right like some people

teach more in a traditional way right so

to change that is even hard right

because that's how they learned um but

uh yeah so then if you add technology it

might be a little bit more difficult but

uh who knows right like we have to

believe that people can change their

mindset right that's why we are in

education

so yeah great thank you techology is

available today that technology is

available today all right we you we're

going to we're going to match you up you

have coffee with him for your Technical

and then with her for for the classroom

person

okay um we have 90 seconds left if you

have something on your chest you want to

get

out all right so I have one more

question maybe you can so um this

session was about personalized learning

but we didn't personalize learning to

the people who were here

so what what could we have done with the

technology of today and with the

constraints that we had in this room to

personalize the session to the people in

this room them to ask a question maybe

uh allow them to ask a question is a

good idea but we only have a minute uh

so I'm I'm uh first person I ask a

question wins all right there's a

question there I'll be a little

provocative I touched it You' talked

generally about access to information is

access to information learning is could

be interpreted as like a super a super

elaborate Google right with a very

elaborate very great how is that

connected to learning okay John you seem

you seem ready for this one no just

simply having access to I'm going to try

to be really quick Simply Having access

to information isn't isn't learning

because it doesn't mean anything mean

information is built from data but to

create learning you have to build

knowledge which is incorporating you

know the tcid experiences you get from

day-to-day life plus this explicit stuff

that you get from from tapping into

information smartly you combine these

two things together they got personal

learning they got some learning but

otherwise information is just you know

it's it's chewed up

data I think you ra a very important

issue which is we are assuming all

information is written yeah so AI is

learn to access this information no and

have the problems how we going to De it

with it but we have analyzed 4,000

decision that were made by 60 large

organizations in the last 5 years we

found 80% of the knowledge was on the

minds people yeah and WID spread on

average in group of 20 to 25 people okay

so this is very important because we're

assuming you know that everything is

written and it's not okay when you come

to close decision a specific decision a

specific situation the knowledge is

still a lot of the knowledge in the

people's minds so we find new ways

to have collaborative intelligence which

is something we are very bad on it

because teams or this is impossible to

interact with people and they have the

knowledge thank you the Ricardo hman and

my colleague at the Harvard Kennedy

School uh speaks of knowhow as being

kind of the key um source of uh

knowledge that helps country uh

countries develop so to the extent that

know how is hard to codify in in AI uh

that certainly limitation all right uh

speaking of limitations it's 5:00 we

need to close this session but please H

give a big round of applause to our

marvelous panelist thank

you all

righty if you very thank you so thank

you thank you thank you