LSE Festival Skills Session: AI in Higher Education

hi everyone welcome to this lse Festival online session how should we use AI in higher education part of our skills for a fast changing World Series hosted by lse online in this series we invite lse experts to discuss research Trends in their field about professional skills we need for Success my name is Jeff and I'm a program manager in the LSC online team lse online makes lsc's World leading teaching and research accessible to a global audience we provide a comprehensive portfolio of online programs to equip you and your organization with the Knowledge and Skills to advance in an ever-changing world was part of the lse festival people and change which is taking place all week and until Saturday the 17th of June exploring how change affects people and how people affect change the event is being recorded and will hopefully be made available as a video or podcast subject to no technical difficulties whilst we wait for everyone we would love to hear where in the world you're joining us from today please use the chat function uh below for those who are joining us on Zoom if you also please answer this short poll but let us know whether you're a student an academic or working for the public private sector today we are welcoming Dr Jonathan Cardozo Silva for another LSC Festival Online skills session Jonathan Carlos Silva holds a PhD in computer science from Kings College London specializing in the analysis of social political and biological networks with a background in computer engineering John brings a unique blend of academic and Industry experience he has worked as a software developer and led teams of data scientists at data science Brigade Brazilian data science consultancy startup currently John serves as an assistant professor lecturer in the data Science Institute where he is also the de facto deputy director for teaching his primary focus is on educational aspects of data Science teaching introductory level courses for social scientists if you have any questions during the event you'd like to ask Jonathan please post your question in the Q a box or post them in the LinkedIn chat function we will get as many as we can towards the end of the event I'm now going to hand over to uh Jonathan to begin his talk thank you very much for this introduction Jeffrey let me share my screen so then we can get started um so there we go okay I get I hope you can see my screen already yeah raped awesome uh so let me remove that out of the way Zoom the strange crew so I got this on I don't think I can I do okay great so you can see my slides right perfect yeah uh so again traffic takes thank you very much thank you all for coming and attending this session I hope uh you know we get to to chat a little bit towards the end of the session and as you know the the title is how should we use AI in higher education and mostly here I'm going to talk about my current practice um especially about the last academic term which is when people started really using it it wasn't that much of a big thing uh a year ago um even though the algorithms already existed but it became a huge thing at the beginning of this year so that's probably um that will be my focus fortunately so I would like to start by uh talking about you so do I get to see the poll results I know you popped up for me a little bit but I don't know if I can't see it right now uh disappeared for me as well but it did appear somewhere I'm sure the team will be able to get you oh here we go okay so let's look at who you are uh just so you know uh I like that perhaps the majority of you don't quite fit in the Box I suppose that's great um so we have a mix of universal students we have a bit a few academics here at LSC great right so we have a um a wide range of you know people in studying and work in a different fields so feel free to ask questions and then I can address them in the Q a afterwards in case you know I'm talking something too specific about higher education that you don't get what I'm talking about do let me know um so a little bit about me um Jeffrey already done has done a great job so I'll just mention uh you know what I do here at LSC and I said I teach data science courses and primarily I teach these three courses right now during the regular academic term um and I show this to you just so to to pave the way for what I'm going to talk about so we have these three courses they are all undergraduate courses some of you might be a former student of mine uh ds101 is focused on understanding the fundamentals of data science so you will learn we teach students about the foundations theoretical Concepts we don't go too crazy on code there we introduce some code but not so much but on the other two courses we have your own code we do python we do R and ds105 specifically is about collecting and handling data and dealing with the struggles that comes with lagged and there are many destroy is more about the fundamental algorithms in machine learning and it's not AI crazy it's really like the basics of of this world right so this is what I teach and I also have a summer school that is also about data engineering collecting data so you can see that all of them kind of goes into this world of the computer science corner of data science um as I said I mentioned all of this because I want you to understand what I'm going to talk about today and these courses they have two things in common and which I'd like to draw your attention to so most of the these courses they involve writing lots of code either r or Python and some of them we even let students um you know we allows students to choose freely which programming language they prefer um and that's that's perhaps the majority of the the day-to-day activities in this course is your writing code you're getting something out of it and then you're interpreting the results of it or you're writing a big data pipeline but some of these courses and some of the assignments in the courses that are more code related they involve writing essays and involve writing things if you have been paying attention to the discourse in higher education you will know that SNS chat GPT came out and became a huge thing especially at the beginning of this year at the start of this year many people started to worry about the future of assay writing so many said that assays are dead the essay assignment cannot be used anymore but I think this is misguided I and I have a different view of that and I don't think this is the essays are dead I just think that we need to think about them uh differently which is something that we should have already been doing anyways but first let me talk about how I am using and I'm embracing generative Ai and all of these tools in uh in our courses for code right so I'm going to talk about code related stuff first um specifically about one of those courses that I showed you in the last iteration of this course we had this final assignment one of these assignments of this course was to look up all of this data sources so you know you probably are aware of Wikipedia but maybe you don't know that there's a whole range of other data sources around the same branch of company you have Wiki Wikimedia which is the foundation that manages all of this data you have data about books quotes species news there's a whole set of things and then we let our students free to choose how they're going to combine data there were a set of instructions like you have to show us that you've done this and this and that that relates to the learning objectives of the courses but we left them free to choose what they wanted to focus on right and this is not a final project it's like a real a regular assignment in the course but it allows for some flexibility and creativity as part of that that would be key to what I'm gonna talk about when it comes to generative AI because if we want to embrace and we I think we should embrace generative AI in our teaching and our learning creativity originality is going to be the main thing so we can't uh assign students to work on things that are true formulaic or two uh close ended so that's why we're going for this kind of assignment that is like very a lot more open um and if you are a coder you will know that when you're doing that and on top of writing code you have to think creatively about what you want to get out of that you know that there is a huge deal of trial and error right you have to write a piece of code first and then you get the results and then you think uh no that's not what I want or that doesn't tell a good story or the data that I collected really doesn't make much sense so then you have to go back there's that part of it and there's the part where you're actually um facing the challenges of programming itself so you're learning to program you're learning or you're and you know expanding your horizons in in programming and then you face you'll be faced with different packages different challenges in specifics of of the syntactics of a particular the semantics or syntax syntax of a particular programming language so trial and error is a key part of the learning process for all of this whether the creativity whether it is for learning how to code um and as a core thing is in in discourses and in in the things we teach is that students and instructors alike we all need to be willing to embrace failure sometimes I'll be demonstrating a piece of code I'll be doing a live demo to our students and that live demo will fail and I would not recall exactly what it is I know which programming Library you I know exactly I know where to find information but I don't really recall um the right words to to type it in there and I would argue that this is okay this is fine and like this is not exactly a failure but even when it is a failure when you type something and then you get the wrong results you have to go back to a documentation you have to go back to Google that is part of the process that's how people who write code uh professionally as a software engineering as a data scientist as a data engineer uh this is the kind of things we're teaching here people in this career path they do this all the time that's all we do so we need to embrace that trial and error um and one thing that we're trying to teach our students and we're trying to get better with this uh over you know after term and term and term is that we need to teach students how to Google stuff and to understand how to read for example things on stack Overflow uh which is the one of the main websites for finding out code related uh questions so that was already a thing even before chat Deputy existed um and we wanted to have students to experience that so it's part of the real world to Gap stuck with something and then you have to go and find the information uh you won't just be able to fill in the blanks um and as you can imagine this requires a lot of contact time so instructors and students we would have to be in content uh in constant conversation to understand where the students like as instructor where the students is getting are getting stuck where can we change and how can we better you know guide them towards this process it really takes time to learn how to do this so I mean if only there was a way to automate part of this process and that's what I think generative AI plays a big role here so there are this is this is what you see here on the screen like on the slides like Googling and then going on stack Overflow is kind of how we always did these things for the past few years uh in this field but now we have chat Bots that we can pose certain questions and then it will kind of fill um you know you will perform the same role for us but in a more conversational way you know easier way to access there's I've seen anecdotally I see a lot of people sharing that they asked chat GPT for code related questions a lot more than they go on Google if you know how to write the question if you know how to do The Prompt engineering correctly that will help you a lot so I mean that is like Paving the way for what the philosophy of our courses is and I think there's something here to learn uh for the whole higher education system in general be more specific about what we do in the coding assignments at this part last term we let our students free to use AI help even in their coding assignments it's not considered cheating they could use chat gbt to generate code Snippets and rerun their code and then copy and paste things here and there and I'm going to give you an example of why that is a good thing and in line with the with the philosophy I just shared but I actually encourage them to use another tool a lot more than chat jpg I encourage them to use GitHub co-pilot with which I think is better for this and in and it integrates nicely with uh tools that people will use in professionally so let me give you an example of how it looks like she used these AI tools to write code and how that would work uh and I came up with this example I really came up with this example like at random so I I know that there's this thing in the programming language are called leaflet which is a package that lets you create Maps like you Google you have on Google Maps where you can zoom in zoom out but I knew nothing about this package or I myself didn't know anything about this package but let's say I wanted to create this for this presentation and I wanted to create a leaflet map using R this programming language and Center that at the LSC how would I do that so my normal approach would be to Google and find tutorials for leaflet so how do I run leaflet and then after using that I will learn how to create a pin in the map and place it on where lse is and I'll have to locate lsc's latitude and longitude coordinates that that is what I would do if I weren't using any AI tools let me show you how I did this using copilot the GitHub co-pilot tool so essentially you open this is like you're writing the code and the part in green is what I wrote so produce a leaflet map centered at the lse and everything you're seeing here now was generated by the AI tool so I didn't do anything after that so I just typed the things you see here now the green and I hit enter and then and GitHub co-pilot will fill in the blanks for me so it imported the right library and it created something so if I'm blind about like you know completely unaware of this package I would just say sure it looks like it works it looks like a actual code so what would I do next I would just simply run this code and see what happens so when I do that I see that this happens so if you zoom out you I mean it worked in a way so I get the map and I can see that it's placed you place the marker in London which is great but if you are if you are an LLC person or if you've been trying to see you will know that this is not where we are even though the marker says you know if you click on it it says LSC great amazing LS is not a total corporate LLC is around here uh is in a different place but it's serious impressive that this code produced the map it uses the right package as instructed and he created a market for me it's just not exactly there so what would I need to do here now is just change the coordinates live student longitude perfect now let's see how charged PT the most famous perhaps AI tool would do the same thing so chat apt would uh you know like charging likes it's it's built as a chat so it's very chatty so it will write a lot of stuff and not just give you what you want that is one of the reasons why I I usually suggest using copilot because with copilot you're being more productive and we stop anthropomorphizing these tools so I I think that's a better thing to do so the charger PTR OS kind of approach it as if you're talking to a person and that is not right for me like that's not the right way we should be using these tools but anyways it works it produces a lot of text he understood I gave it kind of the same prompt that I gave to GitHub copilot and you can see that it says rather than the text like it's centered at the London School of Economics so there's already something interesting it understood that lse stands for London School of economics and it gave me some code nicely formatted and explains what it's doing the problem is when I run this code and do the same thing as I did before I get this so I get a marker I get a map thing that I can click if I click on the marker I marker I see that London School of Economics great but it does not work um straight away so something here something is missing which for me is the most important part the math uh if you take a closer look at the code that It produced and then I just copy and paste it and clean it up a little bit you will notice that you know it created this plot but then you forgot charger but you forgot to add this particular line here so it's just this so this thing that I just highlighted here add tiles that's all we forgot so it added the last digits and longitudes correctly uh it added the markers similar to what copilot gave it to me I just forgot to you know surround the map with the with the fill up command if I just add that add tiles I would see that charger picture did a much better job so it placed LSC exactly where lse is so he places correctly so that means that these coordinates however no one understands how these tools actually work uh by the way so even the creators of charge apt they don't fully understand how Chad Deputy is so good but for some reason it found that these are the latches and longitudes coordinates for lse uh and it works and the map is there and it works as I want to so what do what can we learn from interacting with these two uh straight away um my takeaway from this and this is just one illustration of how you would incorporate that in teaching is that these AI tools they can help you kick start the coding process so if a student learns how to use how to write the right prompts to to interact with these tools they get rid of that fear of the blank page and like I need to create this leaflet map and I don't even remember I don't know where to go to so this can give you a boilerplate you can give you a template for what you do um but students will still need to understand the code they're writing if they simply copy and paste the code they will not get the response they wanted but it is it is key here that our uh our assignments as instructors are um conducive to this process if we give just filling the blanks kind of assignments so like this is a code add figure out what is missing and add something here so that would not be as stimulating and it would not be gaining much from this process of using the AI tools but if you give a more open-ended questions like I did like I want the leaflet map or centered at the LSC then you get to experience this process of the trial and error but it's a guided trial and error and you along the way you also teach students how to interact with these tools how to write more efficient uh questions and when the AI gets the code wrong you teach students where to go to for information so that's here where we go to the traditional approach of go and check the documentation of that particular package you're looking for understand a little bit how it works so it it perhaps even I would say that it gives a little bit of a curiosity in there like I can I almost see the thing working right when we have the especially with the co-pilot example it's kind of working but not really right so then it's it's one minute where like it's just one uh thought away uh for you to understand that okay what the thing that I got wrong was just the coordinates that's fine uh and then I would say that this helps me understand this code a bit better so I know where to add adjust and change so uh we still need to understand the code we're writing and I would say that this serves as a teaching opportunity to explain why these tools sometimes fail so why doesn't charge EBT give me the write a solution why did he forget and this is where as a structure you would say well we don't dischargivity was not created for providing the right solution the correct solution it was created to give you the statistically um convincing sound solution um convincingly Sound Solutions so you you there's there's a bit of you know a teaching here and there about the tools themselves and as I would say towards the end here I think this is an important skill because we're seeing this those um gaining popularity in being part of many professionals now and perhaps it would be embedded in many of the tools we use every day like Google and Microsoft tools so maybe that's a good way of going and approaching the source like let's incorporate that um and also one thing that is important for for the things I teach is that AI won't always Write Clean code we need to teach that like clean code involves writing code that I can understand in a year's time or that others can read understand and replicate and run alone AI would not be that good for that even co-pilot like it will sometimes it would just cop duplicate duplicate duplicate things so you might not as always get the best response so there's a lot of learning opportunities and I think this helps us focus on the right things rather than focusing on skin tags rather than focuses on oh you forgot um you know curly brake brackets here that's not the main important thing we're learning when we're coding we're we want to achieve some things and I can share that our students I had some informal conversations with students in the last academic term and they really enjoyed the productivity boost that these tools gave them some prefer charge EBT some preferred copilot interestingly you always have a group of students that prefer not to use any of these AI tools and that's also okay I don't think we should force people to use AI tools I just think that we should you know give that as another set of two so if you want you can use that but that is not that is not for everyone that is not the way everyone learns but we should embrace it okay so that's how we do with code and as I mentioned at the beginning students they learned that in the teaching when we're teaching but they also can use it for their assignments and that's not considered cheating what that um implies is that our assignments have to be to a standard where we judge originality and creativity a lot more than we judge you know we have it can't be formulaic fill in the blanks that's something that I would argue it's not good for learning anyways but what about then essay writing right as I said that's what people fear the most so if you're writing an essay with rgbt and then you know judgment you will confabulate things will come up with random facts that are not true so what happens then like should we not just simply abolish that like an abandon chargeability for this practice well um first we have to think about how well chapter BT can write an essay many people will feel that students will write their AI to use this AI tools to write in essays and you know they will not learn anything from it and that that will be the end of the learning experience I think that's misguided uh but if you play with charger PG you will know that it can write a pretty convincing essay it can give you you know if you give it a prompt and send it to charity you give you some um some essay um but I would say that most often than not they are very boring and unoriginal so the essays are like convincing they are okay but remember these tools they are trained they were created to generate um likely uh you know the most likely set of words that you would encounter so this is kind of like getting all the essays that has ever been written in higher education and taking the average of that and writing it so like that would not be super original that would be very boring like and you try it yourself and try to write it yourself if you've done it you see that the writing is good sometimes it sounds some people describe it as uh white male um speak like uh it's just like someone super convincing and authoritative that's kind of how chargability sounds sometimes but the in terms of content it's rather boring uh I'll say So to avoid this again we have to rethink our marking criteria we haven't really is that we still don't have the best solution for that um but we're moving towards something that I think is a great goal towards like originality in critical thinking very similar to the goals that we had for uh for coding um and because the assays will be kind of boring and unoriginal and from Lake that's why I would say that as instructors we have nothing to fear about that so if a student is getting straight A's with just using Chacha bichi in your essays it just means that your assay prompt is just not original as well so like your essay prompt is just too boring we have just to to improve that and I don't think that is a problem with AI tools themselves like it's a problem of how we think about these assignments um the key thing here is what are we asking our students to write about um the way we've done and if you remember the three courses that I showed you at the beginning ds101 is the course where is more assay based and in those course in in that course we asked students in the beginning like they have to judge a particular academic paper and reflect on that and we have like specific criteria about they should link to things they we show them in the classroom this is perhaps the most important thing so that way they show that even if they're using chat apt or whatever they're still linking to things that we expressly uh taught um and that would be part more unique in that way but also um they're asked to think more broadly and think okay so if this academic paper is about an application of of AI of an application of a new technology that is disrupting some some markets and some labor forces what would be the future like think about other stuff outside of the context of The Limited contest of the academic article um but in the final essay for those students we asked them to write about something about an application of data science and AI for their own degree program so if you are doing anthropology if you're doing sociology you will try to find an application or if there's any there isn't any you you try to see how Anthropologist sociologists could use these tools in the future or how would that reshape the way people conduct research so that will nudge people towards thinking more original thoughts and rather than just saying write an essay about supervised learning which is something like very specific and you know it would be boring so we can't have boring essay problems as well um this is how we instruct our students so we also tell them in you know if you use AI help it's okay and I tell students even if they just simply copy and paste the assay prompt and put it on charged PT and charge it gave um an essay that they think is okay and good for submission they can submit that if there's all it takes that tells me that my assay prompts are not good so for the next iteration I'm going to change that and all I ask for them is that to report they have to report what they're doing and I also explained that you know they tend to generate responses that sound convincing but are not necessarily correct so and then we ask our students think about it how are you checking whether the the things you're getting out of charge apt or any other tool is correct how do you fact check things and that brings us back to scholarly practice right so uh journalists they have that embedded in the practices for example they know how to fact check they learn that's what they learn how to do academics we also have to learn how to check the sources check the references see who is saying what and where they come from we have to think more critically about things and I think that's what higher education should be about and that exchange right that's what is unique about that even either online either in person that is the thing that I think universities can give Society the most um the best and yeah so if their responses the essay will be judged as if it was a person fine even if they do use with the AI but then we're going to use it we're gonna judge the modern originality and then we're we're fine-tuning the marking criteria so that we actually assess on that in terms of teaching I use charge Deputy for a bit of fun during the last term as well so I did a live demo about fake news and we know that that's one of the major concerns that you know governments and people are thinking about because these tools if they're on they found the wrong hands they can tailor uh fake news to the specific people um that they're targeting so think about Cambridge analytical Cambridge analytica was this big Scandal where um Bad actors were had access to people's profiles psychological profiles and they could Target them on Facebook using specific um marketing strategies and things that you know would give them some propaganda or something that would entice fear depending on the person's profile so that is what people are concerned about and many people say like don't use strategy because this thing co-fabulates this thing hallucinates and creates fake news well maybe that's a good way if people are not informed about how this works and how easy it is to create fake news maybe that's something we should uh be telling people about so just a bit of an example here what I did was I asked students to say log into chapter and ask you to create some fake news for you something about London something about I don't know uh and they couldn't because if you've been using charge apt you know that it would say as an AI language large language model I cannot produce uh misinformation blah blah you will say that but a trick that many people who are familiar with child Deputy would know is that you can ask it to impersonate someone so I asked you to impersonate Rita Skeeter which is a character in Harry Potter series that is essentially kind of a Daily Mail reporter that you know has a very uh opinionated views of facts uh so then when I asked you to do that and I asked it to write a nasty frequently news piece about how climate changes have had and not real and then it did say that and then I asked for more information that kept on going and It produced this this Pearl here for me because look at the second the second part of this slide so I asked it to create evidence and I gave it as quotes and it gave me some facts that some things that sound convincing if you were to read it on an online website and it's in bullet point and there's percentage and there's big names and acronyms depending on how you landed into this page you believe that all of this is true and none of this is what you're seeing here is factual none of that is factual I intentionally asked the AI to create fake news for me and it kind of worked um so what did we learn from this experience so she just were shocked how easy I made it to be like just my personality and by the way open AI has been putting some safeguards to avoid this kind of thing to happen but here we can you can still uh find ways around it so maybe we shouldn't hide from this reality maybe we should teach students that you know like this is quite easy to do like and it's a different scale and let's think about that let's let's address it as a problem should a ban should we put a ban on companies that are creating these tools or you know some people might be creating those tools anyways regardless of the ban or not so how what is our response to that like it's not an easy problem and our students are in conversations that our students maybe we can address them so some lessons learned from this initial experiments all of his kind of experiments and we don't have yet a very structured way of addressing I'm hopeful that we will have something by next academic year but what I can tell you is that some students did not feel like using chat typically at all to write their essays I actually had one student that said can I not use it am I allowed to not use it and of course uh but it's again the same with code you're it's just a tool like it's there it's a tool for you to use and we're thinking about it critically but it's not mandatory but those students who did use it they find it very very useful for outlining what they're going to to talk about they find a user for brainstorming you know that kind of thing where you're discussing with someone and then you don't know where to go and you you interact with the tool and then there's an idea there that doesn't replace the instructor that doesn't replace the classroom discussions is just like an aid it's just something that it can help with understanding where you're going um many people said that it helps and I do find that myself kind of it helps connecting sentences and ideas and I believe that even the process of iteratively editing and something written by an AI to can give me uh an understanding of you know how to write better myself without the ER too I find that by myself my personal uh opinion is that yeah I can I feel like I can write a bit better now because I have this kind of uh tool so final thoughts how should we use AI in higher education or should we use Ai and higher education my current answer is yes I think we should embrace it I would actually say it's an imperative to engage somehow with generative Ai and I post it here like this this photo I took out of tempo station here in London where there's this ad for a particular AI tool that can help uh you know solicitors like lawyers to find cases and judgments if you check the news you know that some guy some person in Texas used something similar to that they actually use chat apt to help them build a case for an argument he was making in court and he didn't know that charger BT fabricated information and he used them and now he's having to respond to you know I use arguments based on things that do not exist so maybe we should be teaching students how to you know use this tools but be careful uh and how should instructors do that I would say experiment so the first thing you can do is experiment I don't yet have a full clear path for you but I would say that the best thing I can do myself right now is experiment again as evident as is evident on this ad we see these tools um dominating much of our careers and markets and stuff so let's experiment with that even if you have a critical eye give it a go and see how it works and how it doesn't work uh I would say also experiment with texture image generators you've probably seen this photo uh of the Pope with puffer jackets uh that was generated by a text to image he's not that cool yet um and also try using try using these tools like really try to explore the different try creating something I like to do is coming up with the weirdest prompt of all uh something that is inconceivable you wouldn't see happening and then you you paste it there you get an image and then you're like hmm how did it assemble all of that and what are the hidden biases in there so why is this image generated particularly the way it is even though I created something that is weird and nonsense ago there's a lot to learn from this there's a lot to learn from the knowledge of the web of the internet so I'd say my my main takeaway here is like give it a go and see uh and criticize it but give it a go thank you all very much so I'll hand it back to Joffrey and I'll hear from you lovely thank you very much Jonathan fantastic uh for Ralph had a really interested in myself we've got a couple of questions and I'm gonna read them out to you um the first one is from Wallace they've asked how do students cite information for their essay that comes from Chit Chat GPT right so um it's not a good practice to cite charge a BT as uh as an information uh you know repository so what we ask students is to disclose that they've used it but if they let's say in the references they put reference to charge a PT they would be penalized for that because that is not a reliable source of reference so it's a tool but it's not a tool for finding the reference there are tools that are helpful in that so you can use Bing AI for example and being AI will sometimes give you and there's something called um I think it's illicit illicit.org that is for academic references and it uses the same technology charger PT but then you can write questions and then you give you academic references so you would have to check whether those academic references exist and then you cite those references but you would inside charge Equity as a source you what we ask them is to Simply disclose that they've used it not as a source not use it as a reference source right uh question from Dr Lowe Peach they said I agree that encouraging students to think more critically and add references can go some way to manage plagiarism but can you say something more about Ai and copyright breaching right yeah there's a big issue I actually don't know how to solve it and honestly I also don't know how I feel about it because uh one of the major criticism that I would have um about open AI for example is that you know we're using charge EPG we're using GPT they gave us some idea of how they created GPT 3.5 which is the one that most people are using but we have no idea what GPT for the one that you have to pay for was trained absolutely no idea we don't know which data sets they use nothing um and you're right in saying that maybe some of the content that um openai used is you know they shouldn't be scraping it they shouldn't they wouldn't be allowed to scrape that that information um my dad said I don't know I don't I don't honestly have a good answer to that I would say that I could even take the approach a more moral approach and say well if they don't disclose it I won't use it and I think that's that's adequate some people have that approach but I would say that as with many Technologies say Google Google also uh crawls the web for many things and they probably shouldn't have access like it was a legal battle to get Google to allow someone to ask for removal right now you can go and request the Google delete some pages about you or some things that you think is wrong and is um representing you but it took some time regulation is always slow to catch up with this kind of Technologies but what I'm saying is if I feel weird about the copywriting issues of child deputy and you know it's weird it's in there maybe I should also feel weird about search engines because they also crawl the web uh so I guess my um my answers are known answer because I I don't know exactly how I feel about that but there's a danger like there's a there's a real danger here that depending on how you query charge a picture will generate code or regenerate uh text that is um private and that shouldn't be used as a reference that is one of the reasons why some companies don't allow their employees to use GitHub co-pilot because they don't have Clarity and whether Microsoft has used only good licenses on when training those algorithms so I think that's a good discussion to have uh but I also don't think that it bars us from testing and experimenting with these tools just yet because frankly we're kind of powerless again that like we can we can pressure we can put pressure um but yeah I would say that I'm ambivalent about all of this thanks um another question um did those students that chose not to use chat GPT perform better or worse than those did and how do you assess those that do or don't use it right so that is why I was saying that it's all experimental at this stage we don't have the great guidelines what I can tell is that um we're just finishing uh the marketing process of the final essays on this uh on this course the one where we allowed more explicitly to use the charger PT there were two Assets in the first round not a lot of students picked up on charge apt the ones we did we noticed that some of the writing was the sentences were better connected than those some of those students who did not but I would say it wasn't anything like super amazing uh honestly as I was saying like sometimes they generate boring repetitive and commonplace stuff um but I haven't yet read the final essays that our students wrote but I can say anecdotally that there's one student that's had that you know she said like I was proofreading it all the time and she felt herself that her writing was better so the way she was expressing herself compared to herself was better when she was using the AI too um but yeah I would you don't have clear guidelines from you know how um we're going to judge the text right we're not going to assess them based on whether they use their Notch igbt we're going to judge them on originality creativity and how they connected the ideas if they were able to do that themselves without rgbt that's also great but we need we need clearer guidance on that yeah a question from Jin shui um they've asked how do you guarantee the equality in the classroom if you're rich and can afford GPT plus you can get so that you can get access to even better pt4 model While others out um students that are that writing prompts may get not only better writing paraphrasing but also ideas or Inspirations must get higher score uh will you officially make Trump engineering a more formal module just like courses about python or programming that's very good thank you um I'd say that the the question about whether you can afford a better AI tool or not um frankly myself I don't find gpt4 that much better than gpg3 so I I wouldn't worry so much about that but one aside here is that perhaps all of these tools we're talking about today if we do the same talk next year they'll probably outdated they'll probably be other tools I think these tools will become embedded in things we use like Microsoft Word uh and all of that things like grammarly has been there for for quite some time so how do we guarantee equality so I think the one of the best ways is teaching prompt engineering for writing in the classroom because you can get good very good uh you know recommendations and ideas and things from the free and perhaps even the open source solutions that will come out in the next months so I'd say teaching that's why I think like we should incorporate teaching of prompt engineering how to get what we want from tools uh that I I believe and that is a belief right I believe that this will flatten out the the the the skills but we have to teach that we have two encouraging students and that's the same and addressing the same question the second question about uh you know should we make prompt engineering a formal module or something that we should be teaching I don't think just yet but I think it's definitely something I would show and I will demonstrate during the course and if we show them you know this is how you get co-pilot to give you a leaflet map uh that solves some of the imbalance that we have so then all students they are aware of all the practices and if they're using a more advanced tool a paid version I wouldn't worry so much because I don't think they're that much greater than than the free versions that are out there that's great and that that brings the end of all the questions uh thank you for for hosting today Jonathan um for a fascinating quote which I hope everyone enjoyed and thank you for the audience for joining and asking questions there are a lot more exciting events coming up this week at the festival so do check out the program at lsc.ac.uk forward slash Festival that xlsc online skill session will be the same time tomorrow where Dr Karen King and Dr orally not Nielsen will explore how to negotiate the essentials you need to know to manage people and change in business today you can find out more about our lse online courses via the link on the slide and in the chat box and uh thank you everyone see you all soon thank you

Channel: LSE

Views: 348

Keywords: LSE, London School of Economics and Political Science, London School of Economics, University, College

Id: q9qA7pxwj0g

Length: 53min 27sec (3207 seconds)

Published: Wed Jun 21 2023