Episode 123 - Interview with Dr. Tullio Rossi about AI in scicomm

Transcript

Jen (00:00:13)

Hello hello hello everybody, and welcome to another episode of Let's Talk SciComm. I'm Jen and the reason this is one of my very favourite places to be is because I get to hang out with my friend Michael and of course our awesome guests. But let's talk to Michael first. How are you doing, Michael?

Michael (00:00:29)

I'm doing very well, Jen, and I can tell you're excited. We got three hellos there for this special episode.

Jen (00:00:34)

Yes.

Michael (00:00:35)

But I'm also very excited because we are speaking with Dr. Tullio Rossi. And Tullio is a science communicator, trainer, speaker, entrepreneur, and originally marine biologist. He's also the founding director of Animate Your Science.

And if Tullio's name sounds familiar, it might be because you've seen some of his great work online. But it might also be because you remember that we interviewed Tullio before. Way back in season five or six, one of the you know, one of the earlier ones, Tullio.

So thank you so much for coming back. And we want to talk today all things AI. Tullio, welcome back to the podcast.

Tullio (00:01:22)

Thank you so much for having me. It's a great pleasure to be here to chat with two of my favourite people in the SciComm world. So very keen for the next 30 minutes or hour, whatever.

Jen (00:01:37)

Well, the feeling is very mutual, Tullio. We have not invited many people back for a second episode.

Michael (00:01:43)

And for the listeners, I'll refer you back to that episode. It's really excellent. You know, all about, you know, thinking about visual communication of science. And we did a bit of a longer introduction for Tullio there. So we're keeping it brief today.

And Tullio, I believe that you have descended deep down into the rabbit hole of AI, exploring how it can transform science communication so... How did this happen Tullio? And what is it like down there?

Tullio (00:02:20)

All right. So yes, I very much descended into the rabbit hole, which at the beginning looked very scary. And I've been through a total rollercoaster about the way I feel towards AI. And it all started with fear of AI wiping out my business. You know, my business is for the most part a creative agency. We produce various kinds of creative services, such as animation videos and graphics.

And about two, two and a half years ago, you know, soon after ChatGPT came out, I had the realisation of what was happening and where things were going to go, especially when it came to image generation with AI. And so it scared me and I literally wanted to understand the enemy better. And so that led me down this rabbit hole and I read books about it, podcasts, audiobooks, newsletters. You know, I just started collecting a whole lot of information and opinions about it.

And in that process, I realised, Damn, this technology is amazing. This is the most incredible technology we will see in our lifetime. And if, I realised that if my business gets it right, we can surf a huge wave.

So if you get AI right in your life, it could be the biggest opportunity in our lifetime. But I think I choose to be a tech optimist and try to figure out how we can use it right, which is still a work in progress.

Jen (00:04:07)

So I feel like, Tullio, we should put you on the spot then, given that you are an excellent science communicator. Can you...? Let's take the positives. Obviously, we'll talk to some of the risks and why there's so much fear later. But let's talk about the opportunities first.

Can you pitch to us why you think AI is in a position to support, enhance, improve science communication? Like what are some of the things that most excite you about how we can use AI in our mission of making science more accessible and engaging audiences?

Tullio (00:04:42)

Well, certainly. I'll give you this example related to what we do. One of our core services is video abstracts. And there is, you know, it's not just me saying it, there is scientific evidence that shows that making a video abstract for your research paper is a really good idea. It helps spread to a broader audience and it's correlated to a higher number of citations down the track. So it's a good thing to do.

And at the moment, making video abstracts in the, let's call it traditional way costs quite a bit of money. We're talking thousands of dollars and it takes a month, if not more to make a nice one.

And this of course is limiting. Only some researchers can afford it. Only some funding schemes allow it. And so it's limited how many are getting done. And so the reality is that there's a lot of science that is not getting communicated in this way, but it would be beneficial if it was.

With AI, we're very rapidly approaching a point where making a video abstract for your new research paper will become much faster and much more affordable. So... And that's huge because it also ties in with the publication cycle of research papers. The moment the paper is accepted for publication and it actually goes public, it's about two weeks.

And that has always been a challenge for us to you know, getting researchers approaching us at the last minute and asking, "Oh, can we have a video in two weeks?" And like, "No. It's not possible". It happened many times. It's quite frustrating. Couldn't you think about it earlier?

With AI, it's going to get faster and way more affordable. So I think we'll be able to communicate more science in more interesting ways, more visual ways that before was just cost prohibitive and impossible to do at scale.

But essentially, the way I see things evolving in the future is that the role of the creative will shift. And rather than being on the tools, will be more of a director role. You know, directing these new tools to do the job right. And yes, and that the result is that instead of spending weeks and weeks on the tools, the whole thing can be done in a week or two.

Yeah, so that's where I see it going. We're not there yet. But we're moving towards that pretty quickly. I think we're just you know, a year or two away. It's bonkers how fast the technology is moving. Absolutely bonkers.

Michael (00:07:43)

It is, yeah. I mean, that's one of the challenges, isn't it? It's like, when do I try and get across all of this? Spend a lot of time you know, reading up about it and upskilling. How soon are those skills going to be redundant?

One of the things that you mentioned in the opening there was you know, you were initially fearful of AI. That you know, this is going to take my job essentially. And a lot of people I think share that fear in some way. That the fear of what do we need to be doing in order to remain valuable, remain productive in the workplace.

You know, what was it that kind of abated that fear in you? Because listening to you speaking, I'm imagining that lots of scientists might be able to create their own videos very easily. And so then they wouldn't have to rely on you know, a service like yours. But are you saying there that there's, that there will always be a role for that human touch and bringing the human expertise into it as well? I hope that makes sense.

Tullio (00:08:58)

Yeah. So I'm a strong believer in the upskilling, which I think is what everyone should go through now. And to help with that... Once I came out of the rabbit hole, I realised I know more about this technology than the average researcher. And so I started teaching researchers how to use AI.

And so I have an online course and workshops. How to use AI, how to use AI for grant writing is the last one I developed. And it's amazing how... what an awakening and eye opening process it is for people. Because the average researcher just didn't have the time. Like you said Michael, you don't have the time to upskill and try the millions of tools that exist out there, which one to start with et cetera.

So there is value in getting some training and getting it in a condensed, curated way so you don't need to waste a million hours like I did reading books and YouTube channels and whatnot. It's all condensed in a more accessible way. Yeah.

So it's also, to be entirely honest, it's a way for me to stay relevant, to just keep up with the world. The world is changing and I love teaching. For the past eight years I've been teaching science communication purely. Now, "Okay, I'm also teaching AI". And you know what? I quite like it. It's a lot of fun. Yeah, so...

Jen (00:10:40)

We will definitely link to your courses in our show notes because I'm guessing a lot of people are thinking like thank goodness Tullio has done all of that legwork for me because I don't have the time. So thank you on behalf of the world of STEMM for curating some of this for us Tullio, because I'm super keen to understand more. I just don't have the millions of hours available to do it, so...

Tullio (00:11:02)

I'm fully aware. But the gain is huge. I saw it with my wife, you know, I've basically been teaching her bit by bit what I learned. And at some point she realised, Wow, the time it takes me to prepare a grant went from weeks to days.

Jen (00:11:24)

Wow.

Tullio (00:11:25)

That if you think about it is life-changing. You know, she's still writing the grants. There's still a human in the loop. But, you know, all the grunt work, the time-consuming, reformatting, rewording, trimming, and all of that, it's all... Now, you have an assistant next to you. You're not alone in a way doing it anymore.

Michael (00:11:48)

Yeah. Okay. So, I mean, you're working with a lot of researchers and I suppose sharing possibilities or upskilling them in different possibilities for how to use AI. Where do you see some of the most kind of helpful use cases, like on a day-to-day basis that you know, where it has really made a big impact in say you know, the day-to-day job of being a researcher.

Tullio (00:12:22)

Look. I can give you... I'll start by giving you my favourite example which I'm sure is also applicable to the average researcher because we all spend a huge amount of time in our email, right? Every knowledge worker needs to write lots of emails every day.

Jen (00:12:40)

The bane of my life, email. My life would be so much better if it didn't exist.
So tell me, how do I solve all of my misery Tullio?

Tullio (00:12:49)

It's everyone's pain. So the way I handle my inbox completely changed thanks to AI. And so I basically use two different AIs in the process of replying to emails or writing them from scratch. I haven't typed an email manually in... At least six months.

Jen (00:13:09)

Wow.

Michael (00:13:11)

Wow, that's a... Can we just pause there for a second? That's a really big statement.

I'm still writing emails manually, but yeah.

Tullio (00:13:18)

But my emails still sound like me. So a better way is to dictate the email to AI and just ask for light editing. So I have this app which I absolutely love called Whisperflow, where I hit the function key on my keyboard and it starts recording my voice. So I am in my email. I open Gemini on, with the sidebar. So it's reading the email in front of me, just like I am.

And I dictate, "Oh, see this email? Please write a response that says this and this and that. Just apply light editing to my words". Done. That took me seconds instead of minutes. It takes, I think it's four times faster to talk than it is to type regardless, on average. And then hit go and the email is written.

I go insert, make small changes if necessary. But most of the time it's not even necessary. Send. And yeah, we're very much approaching a world where nobody will write anymore. It's weird, I know, but it's just the way things are going to go.

Jen (00:14:43)

God, that makes me so anxious when you say that Tullio.

I want my children to grow up knowing how to write and how to edit writing. I don't want them to rely on somebody else to write at all.

Tullio (00:14:56)

I know it is strange to think about it, but that's what's going to happen. Just like we don't write emails manually with a pen on a piece of paper and send it and post it. We're not going to write our emails and even our research papers to some degree.

Jen (00:15:19)

Okay, so we've got to start talking about risks then Tullio, because I'm all up for spending less time writing email. I'm going to immediately start learning how to do what you've just described.

And of course, we are all looking for efficiencies all the time, so I'm definitely not someone who is anti-AI. But you say no-one's going to be writing anything ever again kind of terrifies me.

So given your knowledge... Like how are the ways that you think AI can actually undermine?... And obviously we're talking specifically about science communication here so I won't get anxious about my children and writing. But you know, what are the big risks? What are the potential pitfalls?

Are people going to trust anything they read about science if they know that it was written by a tool that at least everyone today is aware of and hallucinates and makes stuff up? Like what are the risks for misinformation? Yeah, what could go badly wrong? I'm interested to know.

Tullio (00:16:18)

Oh, there are many risks. When I give the workshops on AI, I start there and I start by scaring the bejesus out of people. It's like, "Do this and you'll ruin your career". So yeah, it's all about understanding the technology so that you understand where the pitfalls are. If you just use it blindly and trust it blindly, you can easily get in trouble.

And we've seen examples. The most common examples we've seen is fabrication of references. It could be research paper references, but even lawyers fabricating court cases. And they brought them to the court and then the judge was like, "I can't find this one. Where did he get it from?" It turns out ChatGPT made it up.

Now, they're getting better and better. So most of the time they're right. But sometimes they still mess it up and they never tell you when they make it up. That's what fools people.

The lawyer I mentioned earlier did his due diligence, so to speak, by asking ChatGPT, "Are these court cases real"? And ChatGPT said, "Of course". But they were not. They were fabricated.

And so you cannot ask ChatGPT or any other similar LLMs for things like that. It's great at more generic things. But when you go in the very specific, when you go at the leading edge of a scientific discipline, it just improvises. It makes it up. That's how they function by design. But what fools us humans is that they do it very confidently. So they never tell you, this one I'm not so sure. No, they just give you a very confident answer. And so many people got fooled by that.

Jen (00:18:23)

You can experiment with that, though. I have asked ChatGPT, how confident are you of this? And it will respond and tell you, "Yeah, actually, I'm not super confident with this bit of it." You can play around with it. It's quite fun.

Tullio (00:18:35)

That's an interesting one. I haven't tried. Yeah, I think a good rule of thumb is to never assume truth from an answer you get from AI. But there are ways to improve this problem which is for example to provide better context to the machine. Because if you just ask the basic chatbot relying on its training data question, it will... there's a good chance it will make up an answer that is not factual. But if you instead provide context like attaching research papers, attaching books or you know, grounding basically the model into the content you upload, then it will look there, and the chance of made up results dramatically is reduced. So yeah, that's one issue.

Another one to be aware of is bias of all kinds. Racial bias, gender bias. Because it's not that the machines are racist or anything like that. They're just tools. The problem was in the training data, which we humans created. And yes, did contain all of our biases, and so now it got baked in. And the machines just amplify that.

And then I think a big one that we should recognise is that all of these AI models have IP theft baked in by design, at a scale we've never seen before. And basically what these companies did was really sneaky. They started pretending it was all for research purposes. And so they gathered copyright protected content of all kinds. Books, papers, images, videos, anything. They basically scraped the entirety of the internet. They made these giant datasets and they used those for training the models without asking for permission, without offering any compensation to the owners of that intellectual property.

And then they started playing with it and eventually they came up with a really useful technology that they realised, Oh, we can commercialise this and they did. And so they were the typical Silicon Valley... It's the you know, mindset of 'Go fast, break things'. They just made the products, started selling them. But they all have this like original sin if you like, in it.

And it's still very much in the legal system right now how this is all gonna play out. If they're going to have to pay fines, if they're going to have to be you know, limited, or go back to retrain their models with copyright. You know, for things that they own, or buy the things that they trained on.

It's still very much in the legal system right now. But the problem is that the legal system moves slowly while this technology is a rocket. It's like the Manhattan Project to develop a nuclear bomb times 100. You know, it's huge. Huge amount of investment going in it right now, so... Like it or not, it's not going to stop.

Michael (00:22:14)

Yeah, no, I think you're right. I really want to ask you about what I think is the biggest risk. I mean, so bias and IP theft, you know, of course concerns, but mainly concerns with the technology.

I'm mostly concerned with how the technology affects us. And you know, you mentioned there that you don't think people will be writing in the future. I think writing is a really important exercise for your brain, in order to have high quality ideas.

And I think that you know, there's a risk of undermining our ability to think. There's a difference between using it when you've already built up a level of knowledge on something, on a topic. I think then you can use it, you know, as a co-pilot.

But I think, you know, there's a whole generation of kids who are growing up with the technology and yeah. Like, what do you think about that, that risk of using AI tools and how do you think it could undermine our ability to think?

Tullio (00:23:18)

It's a great question. I do ponder about it a lot. I have a two-year-old son and I really struggle to picture how school will look like for him and what kids will go to school for and what the job market will look like in 20 years. It's almost impossible to think once you understand the technology revolution we're in, where things are going, and then give it another few years. We're also going to add robots to the mix. It's going to be absolutely crazy.

I don't know. What I've personally found very reassuring is learning what Socrates thought about writing when writing was the new technology. He hated it. And he thought it would make people dumber. Because before writing, all intellectuals had to hold everything in their head. And so people trained and you know, worked on that very hard. And they practised dialogue and debate and all of that. And then all of a sudden we were able to write things down. And he hated it because when you write things down, it's not a debate anymore. You know, you cannot debate with the book. It's written and it cannot change it. So he absolutely hated it. And I realised, isn't that exactly the same way we look at AI today? It is.

So I think yeah, it's hard to imagine what the future will look like, but in a way it happened before. You know, I don't think it's that new. Now we use, you know, we grow up writing and now writing is becoming obsolete. Yeah, it's weird, but also interesting to figure out what the next thing will be. I cannot tell you.

I would be just arrogant if I pretend I know what's coming. I think it's impossible by definition. Even the you know, the tech gurus that are in it and they give their you know, predictions, Oh, what the world will be in ten years. I think it's all very arrogant because we are literally at the event horizon. We're at the point where the technology is changing so fast that it's impossible to see beyond the horizon. So we really don't know where we're going.

Jen (00:25:57)

What does GenAI say if you ask it how we're going to be using it in 10 or 15 or 20 years? Have you tried that? I haven't.

Tullio (00:26:05)

No, I haven't. No, I haven't with AI. I don't know if it's the right tool to use for that because it's still based on our knowledge.

Jen (00:26:15)

Hmm, what exists. Yeah.

Tullio (00:26:16)

Yeah, it's still based on what exists. So it might give you an answer based on what it read in some sci-fi books.

Jen (00:26:24)

Yeah. Well, that could be fun though, right?

Tullio (00:26:26)

Oh, it could be fun for sure, could be fun for sure.

Jen (00:26:29)

Michael and I like sci-fi. It just means we'll get to look forward to a sci-fi future.

Michael (00:26:34)

Yeah, we should just get all of our ideas from sci-fi and just, you know, get AI to scrape all the sci-fi and say, you know, "Based on this, give us a blueprint for you know, how we're going to build our future". I'm only joking. But I do, I am a big fan of sci-fi and a lot of science is influenced by sci-fi, I think.

Tullio (00:26:57)

And what we're looking at with AI is science at work.

I mean, it is a product of the scientific endeavor and possibly the ultimate technology.

Michael (00:27:12)

Yeah. Hmm... Well, the umm...

Tullio (00:27:14)

Yeah, I know. I was certain it was going to get to, the conversation was going to get to a point where we all go, hmm...

Jen (00:27:24)

Which we've all just done. We've all just reached your point of silence where we're just not sure what to say next.

Tullio (00:27:31)

But in a way, this is what I wanted because I realised that because I looked into it so much, I had those moments when I really got... I really got to thinking, Okay, wow. It's impossible to predict what's going to be in 10 years, in 20 years. But most people probably haven't gotten to that point yet.

So I'm glad if this podcast episode triggers that response and makes people reflect. Because yeah, there's a lot we need to work out as a society, how we're going to deal with this AI thing, which goes well beyond the science communication sphere.

Jen (00:28:12)

Yeah, of course.

Tullio (00:28:13)

But in a sense, it is all a science communication exercise because it is a technology, it is science, and talking about it is a form of science communication.

Jen (00:28:25)

Yeah, of course it is. But I think Tullio, I think we need to have a regular... You can be our regular six-monthly... You know, we need to check back in with Tullio and what the latest advances and changes have been. So we'll just keep inviting you back to be our you know, AI correspondent. So Tullio, tell us what we should be worrying about next, please.

Tullio (00:28:46)

Worrying about?

Jen (00:28:48)

Well, getting excited about, worrying about. Just come back in six months and tell us what's you know, what's changed.

Tullio (00:28:55)

Yeah, I'm happy to do that. Absolutely. Now that I went down the rabbit hole, I've never come out. Now I'm committed to training researchers on how to use it right. And so I'm compelled to stay at the forefront of this thing. And yeah, so more than happy to come back and give you the latest.

What I see now, the leading edge is, is that we're moving from not only having large language models that write stuff for you or they make an image for you, but they actually do things for you. And that's when we start calling them agents. And so the agent is basically when the machine has a high degree of autonomy and it doesn't need you there to prompt it constantly to do the next step and the next step.

It comes up with a plan. and then execute that plan based on all the tools it has access to. So it can go on the internet, do some research and bring it back and write something. And then review it and criticize it all on its own. And then improve it and then generate images for it. And you know, you can have some fun designing workflows.

I haven't gotten into that a lot yet, but that's where the leading edge is. And ultimately everything we do on our computer, AI will be able to do. I've seen the first examples. There's one made by Copilot, Microsoft Copilot, where you literally share the screen with the computer and you do something on the computer. It could be I don't know, transferring data from Excel to some other place or vice versa. You basically demonstrate like you would to another human. You click here, you click there. You do this, you do that. Stop. And then the machine understands the process and then does it.

Jen (00:31:04)

Amazing.

Tullio (00:31:05)

It's nuts. Yeah. Some days I wish I was a plumber and... because in a way they [are] a lot safer.

Michael (00:31:15)

Well, what about the robots, Tullio? There could be robot plumbers in the future.

Tullio (00:31:21)

Yeah, I think we're safe for another few years from the robot plumbers. And also, I think the... it would be interesting. I think the houses built by humans will be still serviced by humans because they're too messy. And you know, are non-linear, non-predictable for a machine to work in.

But I think soon we will have 3D printed houses made by robots, designed for robots to fix from day one. And those will be a lot cheaper of course and maybe even better, who knows? I think those would will not need humans at all for fixing stuff. But the stuff made by us, it's so messy that probably still need the plumber for quite a long time.

Michael (00:32:15)

Well, those AI houses sound great. I'll need to get one of those, complete with a robot chef who can cook for me. And hopefully not, you know, channel the rage of some chefs and throw pans at me. But...

So I mean, is it fair to say then Tullio? I mean, it's hard to predict what the future is going to be, but you're optimistic? Cautiously optimistic?

Tullio (00:32:38)

I was... Let's say I choose to be a tech optimist. And there's lots of issues on this planet. I do believe technology can get us out of a lot of them. I don't know, the moon shot, like the best case scenario my view would be if this AI thing becomes really smart enough that lets us crack nuclear fusion. And then we have unlimited energy with that. We can really fix our climate situation and suck CO2 out of the atmosphere at the scale it needs to happen. That will be a future worth pursuing. Well, I will leave you with that one.

Jen (00:33:29)

Well Tullio, I am going to be hopeful. I'm going to be hopeful that your very optimistic view is also a realistic view. And I'm really serious. I really would love to invite you back repeatedly so that you can give us some AI updates.

But I'll just say once again on behalf of the whole STEMM world who's listening, thanks for the huge amount of time and effort you've put in to becoming such an expert in this space.

Because you know, we all have our own experiences using AI more or less or being fearful of it or negative about it or positive about it, whatever it is. You know, we're all you know, at the front of this wave that's kind of taking over the world.

And I just think it's fascinating to talk to someone who's so much more educated about it than I am. So thank you for making the time to talk with us today, Tullio. And I'm going to stick with the optimistic view because fortunately, I am a born optimist.

Tullio (00:34:25)

Me too. I think you live a better life as an optimist.

Jen (00:34:29)

Agreed.

Michael (00:34:31)

Hear, hear. Thanks. Thanks, Tullio. We'll see you in six months.

Michael (00:34:31)

Thanks for listening. If you enjoyed the episode, we'd love you to tell a friend about the podcast or leave us a review. And you can reach out to us on social media.

And we're very excited to announce that we're launching a new newsletter called the ChitChat. So if you are interested in receiving some hot tips, or advice on science communication, or to know what Jen and I have been reading or watching or interested in lately, or to just hear about some interesting science communication events that are happening, then this is a newsletter for you and we would love you to subscribe to it. So check out the link in the show notes or our posts on Instagram. And thank you so much to Ben, Dave, Restia and Wahyu for helping us get this newsletter set up and established.

And also a big thank you to our production team, Steven Tang and Madeleine Kelly.