Brainstorm AI Singapore 2024: AI in education—fighting cheating and preparing smarter students
Khairul ANWAR, CEO and Founder, Pandai Tim BALDWIN, Provost, Mohamed bin Zayed University of Artificial Intelligence Sonita JEYAPATHY, Co-Director, Center for Pro Bono & Clinical Legal Education, National University of Singapore Moderator: Jeremy Kahn, FORTUNE
Category
🤖
TechTranscript
00:00So, Nita, let's start with you.
00:03At the National University of Singapore, your law school, how are you finding students are
00:06using AI and how are you preventing them from misusing the technology?
00:10Why you need to do certain things, why you need to learn certain critical skills with
00:14or without the assistance, that all helps to solidify for the student whether or not
00:20to use this particular tool.
00:22So yes, there will be students who, if they are not better aware, they will try to use
00:29it but sometimes that's not really their fault as well.
00:32They need to be made to understand why certain tools should be used perhaps at a later time
00:39after they've acquired certain critical skills without the tools.
00:42And do you have a specific policy about use of chatbots?
00:46So there is anti-plagiarism, obviously, in any tertiary institution and the knee-jerk
00:54reaction when ChatGPT came out was, as educators, what are we going to do?
01:00Are all our students going to be plagiarizing things from ChatGPT?
01:04But then after a while, things settled down a little bit more and we realized that we
01:09could leverage it as well.
01:11I mean, there was a lot of experimentation but it could be a tool that could assist students.
01:15It could be a tool that could assist educators as well.
01:18So we're still in the throes of experimentation but students basically know they can use it,
01:25they need to be aware of plagiarism policies, and they need to also take personal responsibility
01:30for how they use it.
01:32And when in doubt, talk to the educators, think about what the institution allows you
01:36to do.
01:37Absolutely.
01:38Cairo, at Pandai, you've created a kind of AI tutor.
01:40Can you talk a little bit about how this works and how do you make sure the tutor is providing
01:45information in a way that the student is actually learning as opposed to sort of enabling
01:49them to sort of shortcut through the learning process and maybe not actually take on board
01:53the information that they should be acquiring?
01:55All right.
01:56All right.
01:57So just a little bit of context.
01:58Pandai is an education app that helps school students outside of school, after school,
02:03to do revision, to study, to assess educational materials, and Pandai started in 2020.
02:09Now we have more than half a million students using it.
02:11So we've incorporated AI, and one of the features is a chatbot that helps students
02:17with their homework, with their assignments, and all that.
02:20But what we've done different is that, unlike ChatGPT, the chatbot in Pandai actually accepts
02:28students' lingo, it's trained on the curriculum, and what's important is that it's designed
02:33not to give the answers outright, but instead to give step-by-step solutions.
02:37So to ask the students themselves, what's next, and what do you understand now, and
02:42where do you think the next step is?
02:47And it's as if a personal tutor is sitting next to the student, asking the student's
02:51question rather than giving the answer outright, as if it's a calculator, right?
02:55So the danger is that, with tools like GenAI and ChatGPT, like Sonita mentioned, is that
03:02students would outsource everything to the AI, right?
03:06And we've heard many times that we can treat GenAI as another educational tool, like calculators,
03:12or when 2C18 came out, or Excel, or Sheet, we can move on to more complex stuff.
03:16But the danger is that, if we allow that to happen, the student will outsource not only
03:21their writing, but also the thinking process behind it, right?
03:25So with calculator, the student has to still think about the question, what numbers to
03:29input, what operation to input, similarly with Excel, Sheet, and Google, et cetera.
03:34But with GenAI, you can just copy everything into GenAI without even reading it, and then
03:38you get the answers, which is quite presentable.
03:41So that's the danger.
03:42You outsource the thinking as well.
03:43Right.
03:44And then you use a sort of Socratic method to prevent students from leaping ahead to
03:48just having to do everything for them.
03:50Tim, I know at MBZ, you've been using LLMs to help.
03:54You've created an LLM that can help people improve their modern standard Arabic.
03:56I don't know if you can talk a little bit about that.
03:58And again, how do you make sure that it actually functions in this kind of tutoring role?
04:04So as context, MBZ UAI, the Mohammed Bin Zayed University of Artificial Intelligence.
04:09We're based in Abu Dhabi.
04:11Greetings from UAE.
04:13We're watching very carefully what's happening in Singapore, learning a lot of lessons.
04:18So we're the world's first AI university, or a graduate university, masters and PhD
04:23students, about 70 faculty.
04:25We're, unlike most universities, we're actually in the game of commercially relevant LLM model
04:32training.
04:33So if you've done anything involving Arabic, if you've done anything involving Indonesian,
04:39if you've done anything involving AI safety, chances are you've used our LLMs, or you've
04:44used resources that have been developed at the university.
04:47So we're very much straddling the commercial academic divide as a university, a new paradigm
04:53university, if you like.
04:54So coming back to the actual question, the use cases for LLMs and AI, more generally,
05:03in education, are very real.
05:07In terms of research, I think there's a bit of a tip of the iceberg effect that we're
05:13seeing in terms of GPT, and it's important to realize there's a lot more that's happening
05:19in AI beyond just LLMs.
05:24So we're also working in the space of personalized learning, trying to increase both the access
05:31and the scalability of education, in part using LLMs, yes, but very much trying to learn
05:40how a student is learning, try to learn what are the points of confusion in terms of their
05:47learning process, and cater the presentation of the curriculum and the generation of questions
05:53to what their weaknesses are, and to try to complement their strengths, to both increase
06:02the personalized tutoring access that, if you come from a more wealthy background, you
06:07perhaps assume to students throughout the world, not just in English, in languages beyond
06:13English as well.
06:14And I think that it's very important that we don't fixate too much, yes, as educators,
06:20we think about plagiarism and cheating, and it's a very real issue as an educator that
06:27I do think about.
06:29But there are so many benefits, so many good things that can come out of the AI models
06:34well beyond the simplistic chat GPT use cases that I think we need to be aware of.
06:40Yeah, absolutely.
06:41That's interesting.
06:42I want to come to questions from the audience in a little bit, so please think of your questions,
06:47and I'll come to you in a minute.
06:49Cairo, I know that with Pandai, it's not just about supporting the student, but you've also
06:54built aspects or features of the product that support teachers.
06:58Can you talk about that?
06:59Because I think that's sometimes overlooked, is that there's so much focus on what is the
07:02student doing, and less on, well, how can this technology actually support teachers?
07:06All right.
07:07So when we started in 2020, our product was designed to help students directly.
07:12So students will go to school like they normally do, but after school, outside of school, instead
07:16of using paper-based stuff, they can use Pandai instead.
07:20And then, around a year and a half ago, teachers actually came up to us saying, hey, is there
07:26something we can do for them?
07:28So now, we have a new model called Pandai Teacher, where teachers can actually register,
07:34have their own ID, and have specific functions just for them.
07:38So I don't want to get into so much detail, but it covers teaching planner, it covers
07:43monitoring the students, analysis of the students, as well as self-development content to develop
07:48themselves, the teachers themselves, rather than just to develop the students.
07:52So what we've seen is that there's a valid concern on the over-reliance on AI interaction
07:59that diminish the vital interaction between students and their teachers.
08:04So it's important that any tool that we develop actually enhances the interaction with teachers
08:13as well.
08:14So that's why, for example, in Pandai, we have the parent model, the teacher's model,
08:18try to complete the ecosystem somewhat.
08:22And we've seen, even with our chatbot, when students are interacting, initially, yes,
08:30they were asking academic questions, but after months of using the product, students actually
08:36started to share about their personal lives, sharing about their parents, sharing about
08:40their boyfriend, girlfriends, about the problems in school with their friends, right?
08:44And confiding into the chatbot, right?
08:46So that's also a danger where young students, especially, because we're dealing with K-12
08:51primary school students as well, treating the bot as if like a friend, right?
08:57So replacing the vital interaction between students and their friends, the social interaction
09:01with teachers, et cetera.
09:02So that's the danger also, the ethics behind it as well.
09:06Interesting.
09:07Maybe we'll talk a little bit more about that, the way these, yeah, maybe this morse from
09:12being a tutor to kind of being a companion chatbot, and maybe there's a sort of tendency
09:15to go that way.
09:16So Nita, I wanted to ask you, do you feel like either the way teachers have to teach
09:21needs to change or what they have to teach needs to change because of generative AI and
09:26because of the popularity of these chatbots?
09:30So I'll share the perspective, the legal perspective, since I'm with the law faculty.
09:35So I guess projecting into the future, the now maybe, we know that human and machine
09:42will need to work together, right?
09:43There needs to be some interact, there needs to be interaction there, and how the unit
09:47operates, leveraging the strengths of each, that will make it a positive combination.
09:58We also know that things will change extremely quickly, and people will need to learn and
10:03unlearn skills and knowledge extremely quickly.
10:06So taking that in mind and coming back to the curriculum that we teach at the law faculty,
10:13the question then is, what do we need to change and how might we need to change how students
10:18learn things?
10:19Obviously, there will need to be a law and tech element, and by this exposure to technology
10:25can come in various different forms.
10:27Students need to be aware of, obviously, the latest digital products out there so that
10:34they can think about legal issues that can arise from that.
10:38There can be exposure in terms of how educational tools personalize learning because then they
10:43can see to what extent it is useful or not, and the strengths and weaknesses of the tool.
10:48So playing around with that, taking personal responsibility around that.
10:54Exposure to what's happening in the industry because that changes at a breakneck speed
10:57as well, and learning how to deal with or do the digital speak, I suppose, in the foreseeable
11:07future where students or lawyers need to speak to technologists, right, to build products
11:14that could help the industry.
11:17But apart from that, I think in terms of just thinking skills, you want to have higher order
11:24thinking skills like analytical skills where students can look at complicated things, break
11:29it down into its component parts, understand the relationship between the two.
11:33You need synthesis skills where they can look at different threads of thought and create
11:37new patterns.
11:39You need also the values that students are imbibed with because they need to understand
11:45why they are doing things.
11:46They need to understand why they are serving the community as lawyers.
11:50Those things aren't changing.
11:51Those are things that have sort of always been there and remain consistent, right?
11:54But I think now they're more important.
11:56More important than ever.
11:57Yeah.
11:58And if you don't kind of celebrate the humanness of it, I mean, that's what your combination
12:02in the human in the loop is meant to be.
12:03Right.
12:04I want to get questions.
12:05I know there's at least one question from the audience.
12:06So if we can go to the gentleman up here in front, he had his hand up, and then I also
12:10see there's someone in the back.
12:11But the gentleman up here, I saw him first.
12:13So I don't know if we can get a mic to him quickly.
12:16If not, just stand up and say your name and where you're from and what your question is.
12:20Hi, my name is Steve.
12:21I'm from Spark.
12:22So to your question, exactly the same thing that a student should know how to use AI,
12:28whether it's the early predictive AI or gen AI, to be ready for the future.
12:34And the university has to adapt as and when, because the things are changing so fast.
12:38So whatever new comes, the student has to learn.
12:40Oh, there's a mic for you.
12:41Yeah.
12:42Student has to learn that.
12:43Otherwise, there could be universities which are doing it better than you, and your students
12:46get left behind.
12:47So rather than saying, let them use gen AI.
12:50Google came in and people started using Google.
12:53So the whole curriculum has to change in terms of asking questions which are about critical
13:00thinking, where they just can't copy paste from chat GPT.
13:03Yeah.
13:04Right.
13:05So a student has to be well-versed with all the latest stuff and change the curriculum,
13:11change the question.
13:12And you're saying very rightly that students should be able to work with technology.
13:16A law student should be able to work with technologists to come up with new AI models
13:20which help law.
13:21That's a good.
13:22Do you have a question there, or is it just sort of a comment on that?
13:27I'm saying that the questions have to change, and the curriculum has to change.
13:32So I think that's echoing what Sunita was saying, which is that maybe you change a little
13:36bit to emphasize these skills even more now, because you have technology that can do a
13:42lot of the simpler things.
13:43That's what you were saying.
13:44Tim, do you agree with that?
13:45I mean, I do, and I think this is something that all universities are grappling with at
13:50the moment.
13:51I think even in the AI space, for us as a university, our focus has been very much on
13:55education of the AI innovators, the people who are going to be driving the next wave
13:59of AI innovation, breakthroughs in research.
14:03And absolutely, we need those sorts of people.
14:05But one question we've been asking ourselves is, what would an undergraduate curriculum
14:10look like if we had a blank slate based on the existing technologies that are already
14:17here, as well as what we know is coming?
14:19What should that look like?
14:21And the answer isn't necessarily that they need to understand linear algebra, and they
14:25need to be able to train models in PyTorch.
14:29It's possibly there's an even greater need for AI super users, people who understand
14:34how these different models work.
14:35They understand the weaknesses of the models.
14:37They understand why there are going to be shortcomings in terms of whatever.
14:41There was a lot of talk about culture, values in Sunita's answer.
14:46So why is it that you get biases in models?
14:49How is it I can use them most successfully?
14:52How can I apply them in different settings, whether it's legal settings or business settings?
14:57But not necessarily the core skills we had to train LLMs.
15:03We don't need the world to be awash with LLM trainers.
15:06You get a couple of thousand of those in the world, and you're set, really.
15:10There's a lot more innovation to be had.
15:12But really, it's the flow through effects and the middleware, if you like.
15:17And I think for a modern AI curriculum, even in computer science, I think these are the
15:22things we should be thinking about, rather than the sort of classic teach them algorithms
15:27and data structures, and hopefully taught the way I was taught as an undergraduate.
15:31No, I don't think that's the case at all.
15:33Interesting.
15:34I know there's one other question.
15:35The woman has her hand up very high in the middle back there.
15:38Yes, if you can stand up.
15:40We'll get your mic to you very quickly.
15:42And if you could just say who you are and what your question is.
15:46Thank you so much.
15:47This is Wan Wei from Singapore.
15:49And really happy to be here to hear your perspective.
15:52Because the definition of education has widened with technological cycles getting smaller
15:58and smaller.
15:59So my one question for everyone here is that, what is one practical tip?
16:04That means not theory, a practical tip that you have for everyone here.
16:08Because we are all here to learn, to decide what to outsource and what not to outsource.
16:14Yeah, keen to learn from you all.
16:16One practical tip.
16:17Kyro, do you have one practical tip for what students can maybe rely on an AI tutor for
16:22and what they shouldn't rely on an AI tutor for?
16:24I think on the student side, they definitely will try to cheat.
16:29The easiest, the fastest, the most efficient way to cheat.
16:32But like this gentleman has said, teachers have tried to change the way they do evaluation.
16:38For example, like now, exams are being done on paper.
16:43Essays are submitted with editing history now.
16:46And then assignments are increasingly more personalized.
16:49So to force the students to answer and include their personal stories as well,
16:54rather than something they can copy off the internet only.
16:57So I just want to highlight that it's important to highlight to the students
17:02what is education in the first place.
17:05If the message is, we say, you go to school, you go to college,
17:11you buy a big house, you buy an expensive car,
17:13then if that's the case, the message is that this is just a transaction.
17:17And obviously, students are going to cheat.
17:19But on the other hand, if we say it is actually a way for us,
17:23for students to become an informed citizen, a better decision maker,
17:26a better thinker, then maybe that's the way it is supposed to be.
17:32So we can ask philosophical questions to our kids,
17:35but it invites us to rethink how we think about education as well.
17:39So the fundamental of education.
17:41So Anita, do you have one practical tip or thing about output?
17:45What you should outline?
17:47I second Hyrule with what it means to be human.
17:50Like I said, the human and machine interaction.
17:52So you need to leverage on your strengths.
17:54And what are those strengths that will be important?
17:56Your critical thinking skills, your interpersonal skills,
18:00your emotional intelligence skills.
18:02I think you make sure you're not outsourcing those skills,
18:05because you're just cheating your future self, essentially.
18:08And Tim, do you have one tip for people?
18:10One tip, I think agency and accountability are the two things
18:13that we have to hold on to.
18:15So whatever you're submitting as a student, it is your work.
18:18If you're using AI tools as part of that,
18:21you need to own it.
18:23So if it's code, say, you have to be able to tell me
18:25what every line of code does that you're submitting,
18:28even if you've used copilots to generate it.
18:31And you need to own it.
18:33And I think it's important to realize that a lot of these issues
18:36we're talking about aren't actually new issues.
18:38It's a new guise, because we're talking about AI tools,
18:41but contract essay writing has been around,
18:44whether you like it or not, for a number of years.
18:46We've been grappling with it as universities.
18:48The big difference here, as I say, is agency.
18:50I can use these tools to enhance the work that I'm doing
18:55and embed them into my workflows,
18:57but I am ultimately the owner of what comes out of it,
19:00and I have to be able to defend everything that's there.
19:03If I'm getting someone else to write my essay,
19:05obviously I can't do that.
19:07This is me just with whatever embellishments using AI tools.
19:14Great. Well, that's a great note to end it on.
19:16We're out of time, I'm afraid. We could go on a lot longer.
19:19But I want to thank my panelists, and thank you all for listening.