1 – Using Generative AI

[00:00:00] Lily: Hi, and welcome to the RAILect podcast. I’m Lily Clements.

[00:00:03] David: And I’m David Stern. And today we’re discussing “Using Generative AI”.

[00:00:07] Lily: Great. So what do we mean by generative AI?

[00:00:11] David: That’s a good starting point because this is where the world changed a few years ago with ChatGPT suddenly changing the way people interact with AI. Generative AI is AI which is not just doing a task in the background, but it’s generating content, and content which is realistic enough to sometimes be confused for human generated content.

[00:00:37] Lily: And I suppose it’s AI, which is quite accessible to us these days as you say, is where it was changed a few years ago. And I’m sure as we will discuss in later courses, we’ve had AI around for a while, later modules, sorry, we’ve had AI around for a while. But this is where we can interact with it.

[00:00:54] David: There was a point in time, you could say, where suddenly generative AI came into the public consciousness. Deep learning has been in the public consciousness for a while, ever since a computer beat a chess master, if you want. And that was a while back now. And so, you know, AI has been around, it’s been discussed for over 80 years at this point, but the point which has happened very recently is that suddenly with the advances in generative AI, it’s creating content in ways which are changing the way, let’s say, essays can be written.

You know, we’re now worried about “did the student write the essay or did they just throw it into ChatGPT and get ChatGPT to write the essay for them?” That’s a question which you can ask now, whereas you couldn’t do that a few years ago.

[00:01:49] Lily: Yeah, and this is a kind of question that we run throughout the course of how can we handle this or what about this?

[00:01:57] David: And this particular section is all about the fact we can’t really discuss that unless you can do it. And when I say you, I’m not meaning you personally, I’m meaning you as in the listener. We want to sort of encourage you to say generative AI is something which you can use now and try it. You need to try it because your students are going to have access to this. They’re going to use this throughout their lives.

So using it is the starting point to that debate. Otherwise, you’re just debating in the abstract because what it can do and can’t do now is rather precise and what it’s doing is changing over time. But just using the tools that are there and available now is an important starting point.

[00:02:41] Lily: Yeah. And I guess there’s three areas which we’ve wholly split it into in terms of using it. We’ve, me and you, David, have discussed about it being used for text. So for writing things and what not, for kind of code, and for imagery.

[00:02:56] David: Yeah.

[00:02:56] Lily: And I’m sure that this doesn’t encapsulate everything, but these are the three brackets.

[00:03:02] David: When you say brackets, we’ve done three and three in a sense, because we’ve got those three areas where you’re using it, which as lecturers, we know lecturers will be thinking about people writing their essays using it. Well, that’s generating text. We know that people will be worried about well, “I’m teaching people how to code here and they’re just coming back with AI generated code”. And also this element which you’ve found so important, which is about the fact that actually by using it for generating imagery you’re actually able to get creativity in interesting ways that were not possible and were not accessible to people before in the same way. And so imagery is the third one which you’ve really brought in, because you’ve done a lot on that, and I think that’s really important.

[00:03:47] Lily: Absolutely. And then from there, as you say, we’ve then dug into it into three different categories.

[00:03:53] David: Well, when you say three different categories, I would see those as the categories, and the three other things, I would argue, are sort of more like, I don’t know.

[00:04:02] Lily: Phases?

[00:04:03] David: This is a really good question, because in some sense, you could think of them as phrases, but you use it for this, for that, and then for the other. But I actually think they’re three different tasks because you might not use AI for all three of the tasks. They do have an order to them as we’re going to explain, but you might only use it for one and then not use it for the others. Or you might use it for all three. So do you want to tell us what the tasks are?

[00:04:30] Lily: Yes, absolutely. So we’ve got ideation.

[00:04:33] David: Yes. Having ideas, being creative, thinking about something.

[00:04:38] Lily: Absolutely, yeah. Sparking new ideas, directions, creating things .

[00:04:41] David: Exactly. There is actually research evidence already that generative AI is good for this. And it can enhance human productivity being used for ideation. There’s been a wonderful little study with I think it was business students in the US. We don’t need to dig into that now, but it’s a good thing to do with it, to use AI for.

[00:05:04] Lily: And then the next one is iteration, or iterative.

[00:05:08] David: Yes, and this is…

[00:05:09] Lily: Refining, improving, sorry.

[00:05:11] David: You’d often use it in all of these phases, but this phase in particular

[00:05:15] Lily: Task, not phase.

[00:05:16] David: Sorry, thank you for the correction. This task, because, so many times I heard you say “well, I did this and then I iterated on it using the robots”, as you call it, “to be able to optimise or to think about a different way of doing it” and so on.

[00:05:31] Lily: Yeah and I was a little bit hesitant there of you saying optimised because I suppose it’s optimised it, but that doesn’t mean that it’s complete and I guess that links to the third stage. Task, sorry.

[00:05:44] David: Now I can critique you, when you already did that, sorry, go ahead.

[00:05:48] Lily: Of this learning, perfecting, reviewing stage. Task, sorry.

[00:05:54] David: Yes, that polishing, putting a bit of polish on actually.

[00:05:58] Lily: Yeah, where we can then check it over, review it, critique it, and fact check it. This is a huge part of it, and we can dig into that, but fact checking it is incredibly important when it comes to generative AI.

[00:06:13] David: But also fact checking what? This is great. So yeah, I’m looking forward to discussing that third task. I guess the question which I have for you here is, you’re the person who uses generative AI regularly, you’re the one who’s actually going to talk us through what you would recommend. If our listeners never used any of these tools before, talk us through what you suggest they do.

[00:06:36] Lily: Sure. So let’s start with this ideation that we mentioned. So this is just a useful step in creative or problem solving of generating all these different ideas. You don’t have to worry at this stage about the feasibility, the practicality of it, you can just click to keep going, keep giving me more and more ideas.

[00:06:52] David: And I can think of a time when we’ve done this together where we didn’t know how to call something, and we then wanted to have ideas of a name or something like that, just give us 50 names for this.

[00:07:05] Lily: And we didn’t use any of them.

[00:07:07] David: No.

[00:07:07] Lily: But it helped us to refine, it helped us to then see, oh, we like that word, we like this idea.

[00:07:14] David: Exactly. It’s that creativity where people think of creativity as being a very human activity, but in practice, actually, it’s often based on what’s come before. And because it’s based on what’s come before, a data rich tool like generative AI is quite good at seeming to be creative by most measures. So it can certainly enhance our own creativity of building from what’s known and what other people know and having these ideas.

[00:07:44] Lily: And as well as that, I’d say it helps us see different perspectives as well. Of course, AI always has its issues such as bias, so you need to take this into account, but it can still help see it from a perspective that I wouldn’t have considered.

You can even ask it, okay, I want to generate this text, this something, this paper design, this whatever, using this sort of perspective. Well, actually one example is if I’ve generated an image and I’ve wanted it to be from this perspective of a farmer in West Africa. And so it can help to see different perspectives, and by just continually generating things…

[00:08:20] David: But you’re getting towards the iteration now with that, that continually generating is exactly where once you’ve got a perspective, then that iteration process comes in.

Let me just come back to the ideation phase for a second. We mentioned this study that happened where I believe there were 50 student generated business plans, 50 AI generated business plans, which were then judged. And the AI generated business plans actually won, on average. Now, this does not mean that AI is better at generating business plans than students. That’s not how you should read that.

This means that a student who uses AI to help them with their creativity at generating a business plan and actually has that ideation phase in conjunction with the ideas that come out of an AI system can probably be more creative and think more about these different things and get further than they can do on their own. And that’s a way to think about it.

 This is why we are saying that you as a lecturer should be using generative AI to see how it can be used to spark creativity and things, and then thinking about how you can help students to be able to achieve more by using it in that creative process for ideation.

[00:09:43] Lily: Absolutely. And that’s a great example of using it for text.

[00:09:47] David: Yes.

[00:09:48] Lily: Or just generating ideas for whatever kind of task you have. And so it might be that you want to use it for these ideas for like business ideas. Or it could be that you want to use it to give you ideas on how to structure something like a paper or…

[00:10:04] David: Whatever it is that you’re using it for, really that’ll depend on the domain of the lecturers who are doing this. But the point is almost any domain expert has elements where they’re wanting their students to be creative. And if you have ideas of how you could use AI to enhance your ideation, your creativity, then you can think of how you could create tasks that use these generative AI tools to enhance student creativity and the outcome students can come out with.

That’s, I think, our key point here. So lecturers should try this for themselves, using this in some way related to their domain, to their topic, to be able to try and say, can I think of something where I’d be normally wanting to have a bit of inspiration and can I find a way that actually AI can enhance my inspiration and I can learn how to do that?

And I think that’s when we should move on now to, okay, if you’ve used it for ideation, then you want to be careful because you certainly don’t want to get stuck and you don’t want your creativity reduced down to what is suggested by the AI tools. And maybe this is when we should really talk about this iterative stage.

[00:11:26] Lily: Absolutely. So using AI, you can use it to generate all these ideas. And then just to enhance or refine where you are.

[00:11:35] David: And where you are could be coming out of the ideas AI has generated, or it could be enhancing something you’ve created. You’ve written your own paragraph or text on something and you could ask AI can you liven this up a bit? And see what it gives you and have those iterations. Can you make this more, I don’t know, more personal? If you’re wanting to write a letter to someone, well, you may want to make it more personal. Those language skills, and this is particularly, you know, I’m thinking from a language skill perspective, this might be useful for people for whom English might not be their first language, but they’re studying in English. So this could really potentially help them to take their ideas, articulate them and refine them.

[00:12:20] Lily: Yeah, and even to tailor to different environments. I know someone that’s studying to become a PE teacher, they’re planning out a lesson plan, but they still need to think, okay, but what if we had a student who wasn’t able to participate that day due to a disability or an injury? And so even using it to iterate, in his case which is, in teaching at school level, which isn’t quite what we’re discussing here, I know, but in his case to refine his lesson plans to suit, to tailor, to fit different people.

[00:12:52] David: And different contexts.

But I think what is important is, if you’re thinking now of the lecturers training such a student, who will become a teacher, how do you encourage them to use AI responsibly, particularly in this iterative process? What is it that you can do if you’re comfortable understanding, and this is where we should be encouraging lecturers to try this for themselves, to take something, maybe you take something you’ve created or maybe it’s something you’ve shared with students and now say, okay I’ve got a student who I’m engaging with, who is, I don’t know, maybe rather than disability, a student who’s very interested in music or whatever it is. Are there ways that I could refine my messaging so they’re more likely to receive this? Depends on the discipline you’re interested in. I don’t know whether that’s something you could do, and certainly as a mathematician, it’s not something which comes naturally.

But it’s something where thinking about that I think is important. What certainly is the case is, I’ve taught people to code in different ways as a lecturer in the past, and I can certainly see how, taking some of the code I’ve written and saying, could you optimise this for speed, or could you optimise this for legibility? Actually asking the robots, as you call them, to help you to take the code you’ve written and then iterate on it for different reasons and in different ways could be very interesting.

[00:14:19] Lily: I’m not going to pretend I haven’t done that myself and it’s helped, it’s made it a lot faster. And that comes to actually then, the next stage of refining, that perfecting stage, the, as you called it, polishing.

[00:14:32] David: The polishing, yes. And again, you could be polishing something which you’ve iterated on, with AI, or you could be polishing something which you’ve just created and just said, actually, I really want this to be, I want to make sure that I’m capturing this. And there’s all sorts of ways you can give prompts related to something you’ve put in that you want polishing. And I like the fact that you brought out here fact checking.

[00:14:59] Lily: Yes.

[00:15:00] David: Now, if you’ve generated or iterated based on AI, then actually the fact checking you’re talking about is a very human process, it’s not an AI process because there’s wonderful cases, there’s law cases and all sorts which have come out, where fabrications from AI have come up. Do you want to say a little bit about that?

[00:15:24] Lily: Yeah, absolutely. For me anyway, the first case that kind of came to light was of a lawyer who inadvertently included these fictitious cases in a personal injury lawsuit against an airline. And they went to court to present it and it came to light, basically, we can’t find these case studies that you’re referring to. The lawyer came out and he said I actually, I had no idea that using generative AI would create stories. I was encouraged to use it by my daughters. They mentioned it to me. I thought this is a great tool.

And as a result, the federal judge in the case was actually considering sanctions on them over this kind of misunderstanding, as it were.

[00:16:02] David: Yeah.

[00:16:03] Lily: And this all just comes down to seeing that output and taking it as it is. And there’s so many other cases. There’s a professor in America, who’s been accused of abusing his students based on generative AI fabricating stories, or using him as an example of someone that has done that at a school that he never taught at, on a trip that he never went on. And obviously that’s devastating.

[00:16:25] David: Yes. And this is the thing, that it’s something where, to recognise that if you’re using AI tools, it is generating stuff, and by generating stuff, that doesn’t mean that what it’s generating is real. And so therefore, you need to be able to think about how you actually fact check what has been generated. And that’s part of the policy.

But it also works the other way on, doesn’t it? That you could actually do something where you’ve written something, and you could actually feed it in, and ask it to fact check it for you in some ways. Now that’s the sort of thing where, again, you need to be very careful about it in different ways because it may or may not do that accurately, but sometimes it can draw out some of these issues.

And it’s something where it can be used as a tool to polish and to finish. And this is the robots interacting with the robots, is the key point here. There are layers of this in terms of how you use this, and so one of the things that I would strongly recommend is not only do I recommend that, lecturers actually take something, put something in, get it out, and then fact check it for themselves to understand how to do this, what could be generated, really be on top of what it’s doing and what it’s not doing.

But more than that, to then think about what happens when you ask ChatGPT or some other generative AI process to be able to fact check something it or another AI process has created? And that’s also interesting, or something which you’ve written yourself.

[00:17:59] Lily: And I suppose we’ve focused a little bit there on the text side of fact checking, but you can still do it for the code or imagery side.

[00:18:05] David: Yes.

[00:18:05] Lily: On the code side, you could check it if you’ve asked it previously to optimise your code, you can check that it has, you could check it against your previous code to check that it gives the same output. You could also polish your code by asking generative AI to add in comments. Make it more readable, therefore, for someone else to come across.

[00:18:25] David: And this is really important because you’re right, because the fact checking is a small part of the polishing. The polishing could be documenting the code, and AI is great for that in many cases. It really makes it much easier to get shareable code, and a first draft of that. Now, of course, going through and checking that it is actually correctly documenting the code, that’s a human activity. You have to do that because you’ve got to be careful that it might not be getting it right. But it’s a first draft, it can save a lot of time.

[00:18:54] Lily: And it does. And it does. And for other things on code, helping you generate tests, unit testing for your code. I won’t go too specific into there, into those areas. And similarly with imagery. This comes back into that human aspect that we talked before about fact checking or polishing. But checking the images that come out are, you know, not legitimate? I’m thinking of the Google Gemini….

[00:19:18] David: I knew you were thinking about that. This was a big interesting scandal where the images that came out for Nazis , Second World War Nazis, were amazingly diverse. And so it was very politically correct in terms of making sure that there was a diverse range of ethnicities. Yeah, I think it was specifically German Nazis.

[00:19:42] Lily: And also on the founding fathers of the United States, they were also very diverse.

[00:19:47] David: Exactly, politically correct, including diversity, which is maybe historically slightly inaccurate. This is where you’ve got to be really careful with that imagery, and actually making sure, and this is the polishing phase, and it’s a very human element of this. Using it, identifying instances related to your domain area, where it is bringing out results where you need that human element, you need that human element of attention. That would be a fantastic thing to then actually get students as exercises to look at. This is a way you could be thinking, and this is coming to the next topics, which we’ll move on to.

But I think the first topic we wanted to get people to do is: Don’t be afraid to use it. Get down, use it in interesting ways, have fun with it. And recognise what it can do, what it can’t do, its limitations, with respect to what you care about, your domain.

[00:20:42] Lily: Absolutely. So do you have any final way to conclude, David?

[00:20:45] David: This is not just a podcast , this is part of a course. So, for the people taking this course, going through, your tasks for this week are to use Generative AI tools for ideation, brainstorming, to create interesting ideas, for iteration, to take something which exists and refine it, iterate on it, to be able to make it something new, and then polish, actually thinking about what it means to either use AI tools or need a human element in that final polishing, finishing something off, finalising something. That’s what we want you to do. Have fun. And next week we’ll be on to our second topic, which is demystifying data science, isn’t it?

[00:21:40] Lily: Yes, yeah, absolutely.

I guess with that we’ll be going through a bit more on fact checking, where people haven’t put in that human element at the end, and other case studies.

[00:21:50] David: And I guess we’re not going to preempt what’s happening next week, but the key thing in terms of demystifying it is, if you as a lecturer are going to use AI in your teaching, then we don’t want it to be just something which is totally mysterious. It shouldn’t be. It’s a scientific tool, and it’s a tool which you can use and it’s not a tool which is easy to use responsibly. But that’s what we’re aiming for.

That comes later in the course. But hopefully from next week, from this week, you’ll be comfortable using it. And from next week, you’ll be comfortable talking about it.

[00:22:30] Lily: Perfect.

[00:22:30] David: See you next week.