Ethics in AI with Justin Tauber from Salesforce

0:02 Georgie Healy: Founders scale faster on Deel. Set up payroll for any country in minutes. Hire anyone anywhere. Get visas handled fast and get back to building. Visit deel.com/dayone. That's D-E-E-L dot com slash day one.

0:22 Justin Tauber: Privacy in what we put into the LLM. How ethically dodgy is this? Ethics and innovation are not necessarily contradictions. It'll slow you down to be ethical. Like, to try and think of the most ethical response, someone else has already gotten to the solution. What is shadow AI and what kinds of companies are most at risk of enforcing this? I'm sure you do have to guide businesses. How do you guide them when the future has many scenarios?

0:51 Speaker C: Prediction is not as important as rehearsal. Aspiring to be ethical is like aspiring to be musical. You don't get musical, you learn an instrument.

1:00 Justin Tauber: Yeah. And then you're musical.

1:01 Speaker C: And then you can generalize. And then you can generalize and go, oh, okay, 'Okay, now I understand.' Often the ethics happens in the adoption rather than in the actual innovation itself, especially where innovation is just, 'Hey, we've got this new thing, there's new capacity, new thing we can do.' How we use it is really important.

1:16 Justin Tauber: My hunch is that when people say they have a GenAI strategy, what they mean is that they have a GenAI policy. It's not quite the same. What did you mean by that, Justin?

1:29 Georgie Healy: Hello and welcome to In the Blink of AI, the show where we mix AI tech and Louvre burglars, and also about how AI ethics can breed innovation instead of stifling it. I'm Georgie Healy, and this week I'm speaking to Justin Torbay, or Taubay, I guess, if we're being French. He's the GM for Agentic Technology, Trust and Adoption at Salesforce. Many of you listening are probably like me, AI evangelists for the most part. But as I'm often reminded on my TikTok platform, there are some AI doomers out there as well. I had someone comment on my platform the other day that AI is an evil on the world. They want AI fully removed. And why I'm excited to have this conversation with Justin is we're able to go deep on ethics and morality around AI while also coupling that with a love of innovation. Let's unpack some misconceptions. Let's talk about the arts. Let's tackle some AI hacks and hot takes and tell me what you think. I'd love to hear your comments on this one. Thanks, Justin, for being on the show.

2:41 Speaker C: You're listening to a dayone.fm show.

2:44 Georgie Healy: I'm thrilled to be partnered with Stripe for today's episode. Did you know that Stripe Startups offers early-stage venture-backed startups access to Stripe fee credits, expert insights, and a focused community of builders. We love builders on In the Blink of AI. Apply today at dayone.fm/stripe.

3:09 Justin Tauber: Okay, we're good to record. Um, what did you have for breakfast this morning?

3:14 Speaker C: I had a protein shake because I made the mistake of going to the gym and subjecting myself so much to my personal trainer. So terrible.

3:21 Justin Tauber: So terrible.

3:22 Speaker C: I mean, what a toxic way to start Yeah, I call him my personal sadist, and he smiles when I say that.

3:30 Justin Tauber: I bet he does.

3:31 Speaker C: Yeah, he loves that. He just laughs more the more I'm in pain.

3:34 Justin Tauber: I'm a really friendly person, but I will abuse, like, the person who's doing a gym class or something like that. Like, it brings out the worst in me. Like, they'll be like, "And another set," and I'll be like, "Fuck." All right, Justin, thank you so much for joining in the Blink of AI. I've been looking forward to this one.

3:55 Speaker C: Me too, Georgie. Thanks for having me.

3:56 Justin Tauber: Look, we start the show with an AI hack of the week, something that inspires joy, delight, something that the listeners can try at home. Would you like to start us with one?

4:10 Speaker C: Yeah, I mean, I have a work one that I use. I use— we've just released Slackbot internally, which is a new AI tool which I use to actually— can use it to catch up with what all my team is doing. Over the course of the week or the month when I need to tell the rest of the organisation how great they are and what they've been up to. It gives me a great summary because it knows what conversations they've been in, so I don't need to go into—

4:35 Justin Tauber: Not doom scrolling months of messages.

4:37 Speaker C: Exactly. And that's great. It helps me schedule check-ins with them and all that kind of stuff. And that's really useful. But the one that I think is much more inspiring is actually what my wife's been doing. She's been retraining as an interior designer. And she's been collecting art supplies for the last 25 years. The most high-end. We've been to Japan a couple of times, and a key part of going to Japan is buying art supplies.

5:01 Justin Tauber: Oh, they're amazing at this stuff.

5:03 Speaker C: Pens, Textas, I'm not going to— markers. I don't know what the technical terms are.

5:08 Georgie Healy: Yes.

5:08 Speaker C: But there's just wealth of art supplies. And finally she decided to retrain as interior designer. And I thought, fabulous, at last, all of this incredible investment in art supplies is going to be worthwhile. And then she discovered that she could do better renders with ChatGPT of renders of her, uh, of her plans. They're not as accurate, they're not perfectly accurate, they're not, but they're great for concepting and for getting the mood of a space together. So, um, it turns out all those art supplies are useless as we always thought.

5:37 Justin Tauber: Yeah, and I don't know that they necessarily appreciate and value oil paints and the like.

5:43 Speaker C: I don't, I, I don't know. I think there's a massive—

5:46 Justin Tauber: Maybe they do if she's listening. I think that they're a —investment.

5:50 Speaker C: I wouldn't be surprised if she's already done the calculations on that and not telling you.

5:56 Justin Tauber: Yeah, yeah, secondhand fashion, secondhand art supplies, it's very, it's very much her wheelhouse. Yes, um, she sounds like a cool, uh, person that's, you know, really thinking outside the box with AI, and I do love that.

6:08 Speaker C: Well, she's a late adopter for everything, so this was a big surprise. I had to get her her first email address. She swore that Facebook would never take off, thought Twitter was a complete waste of time, But for some reason, ChatGPT in particular has captured her imagination.

6:23 Justin Tauber: You've perfectly articulated this, this technical revolution that is AI. I was, I was a founder very briefly, but in the time of cryptocurrency and the metaverse and all these things, which, let's be honest, probably not as sexy currently. And then with AI, it was Immediately compelling. And your wife's example of actually using the tools quite early is, is quite an interesting sign in that, isn't it? It's not just drinking the Kool-Aid for the sake of it.

6:59 Speaker C: Actually, when I think about it, it's actually a really interesting case of multimodal because it's not— although it's not multimodal in the sense of going from text to voice or from text to image, it's actually going from one sort of image to another sort of image. And yet those, even within images, there are different kinds of structures that they need to have or don't need to have for different purposes. And being able to translate between those things is, I think, the thing that's really blowing everything up at the moment.

7:27 Justin Tauber: Amazing hack. Love that one.

7:30 Speaker C: She'll be very happy to hear that.

7:31 Justin Tauber: Oh yeah, it's actually my favorite. Sorry, Justin. Yeah, I do prefer that one. Mine is, mine is Bit of a random one. I've been reading a book called Genius Makers. And what got me onto this book is another podcast, Acquired. I don't know if you listen to Acquired. Incredible show.

7:52 Speaker C: Mm-hmm.

7:53 Justin Tauber: And they did this deep dive on Google and all the history. And one of the sources they used a lot was this Genius Makers book. It went from the origins of AI to now and starting with academics in the University of Toronto and no one thinking that neural networks would ever be useful or work, to today. I can strongly recommend the audio on Spotify too, if you are like me and want to blitz through it quickly. But the journey is fascinating, and there's key figures. It would make a great movie. Like, there are figures who were a big proponent of AI in the early days, like Geoffrey Hinton, who is now anti-AI. And it's just very— Yeah. Interesting to see the journey that we've been on. Elon Musk's role before and after OpenAI, and yeah, just some big characters in the space and how they've evolved their stance.

8:52 Speaker C: Yeah, I mean, it's a great testament to how much the space is evolving and how hard it is to make predictions this time. I think maybe it's worth mentioning the other way I have started using things like ChatGPT and Gemini is to build scenarios. Future scenarios is really an interesting way of using it. So instead of thinking, let's make a prediction about the future, why don't we make many predictions about the future? And so it's kind of called scenario— it's scenario planning. It's a very, a very great alternative technique to traditional strategy work if you have a very uncertain future in front of you. So by kind of exploring the diversity of futures that you have, you have an opportunity to go, okay, what are all the things that we should do now to help us survive and thrive in as many of those futures as possible? That way you kind of make yourself resilient to the different changes that are coming.

9:48 Justin Tauber: This is perfect. To drill down on what you do day to day, I know your role has a lot of innovation, which scenario planning I'm sure is important for, but also ethical considerations. Scenario planning very important for. Tell us about your day job, what you do, where you work, for the listeners that might not already be familiar.

10:12 Speaker C: Um, yeah, well, uh, look, I work at Salesforce. I've been there for the last 8 years, um, and I've been on a bit of a journey in Salesforce. So I, having been, uh, you know, kind of innovation design, strategic innovation space, so I was helping our customers for many years to articulate their needs and to use design techniques to help them articulate their needs because we have this incredibly broad and deep product and it wasn't clear. You could do anything, what should you do? So using— and that's not a particularly unusual thing to do, it was just really interesting place to do it inside a technology company where you're doing that in a— as a pre-sales activity to help people understand what's possible. Um, but Over the course of that time, I was heavily involved with our Office of Ethical and Humane Use. And I have this background in philosophy, philosophy of cognitive science and embodied cognition and things like that. And as a result of kind of sticking my nose into how we were talking about AI and how we were talking about the ethics of AI, eventually someone said, Justin, you need to take some responsibility for this stuff. And so now I run A team that focuses on making, enabling customers to understand what an agentic enterprise might mean for them. And so that involves solution engineers, technical architects, but also business consultants and specialists in things like workforce transformation to be able to help customers just get a grip on what an agentic enterprise looks like, start imagining that, and also help them to prototype and, and see how the technology can actually work in the context of their environment and their organization.

11:51 Justin Tauber: Incredible, because there's so many headlines around the ethics of AI.

11:58 Speaker C: Yeah.

11:59 Justin Tauber: And even the future of work and things like that, that it's, it's important to have that stuff translated and communicated in a very real way, I'm sure.

12:11 Georgie Healy: As a startup founder, you are juggling multiple priorities from the expected, like finding product-market fit, to the unexpected, like customer requests for SOC 2 or ISO 27001 certification. Achieving compliance is time-consuming, and time spent on that is time away from the needs of the business. And that's where Vanta comes in. Vanta is the all-in-one solution for startups to become compliant quickly. And build a security foundation with ease. With a combination of automation, an extensive partner network, and a security marketplace containing 385+ pre-built integrations, Vanta provides the necessary tools and expertise for startups to achieve compliance seamlessly, no matter how urgent your needs are, and at every phase of growth. Over 10,000 leading companies, including Cypherstash, Handle, and Indetted, trust mentor to automate compliance so they can focus on growing the needs of their business. Here's the important part: startup listeners of the show get $1,000 off if they go to dayone.fm/blink.

13:30 Justin Tauber: I've got quite a few quotes from you that I'd love to unpack. Some good ones, some really good ones. Um, what about this one? My hunch is that when people say they have a Gen AI strategy, 'What they mean is that they have a GenAI policy. It's not quite the same.' What did you mean by that, Justin?

13:52 Speaker C: Well, look, I think that quote is like almost a year old now, so I don't know whether it's—

13:58 Justin Tauber: We need more fresh quotes.

13:59 Speaker C: We need fresh quotes. Let's make some. Let's make some today.

14:02 Justin Tauber: We could do that today.

14:04 Speaker C: I think around the time people were talking about, we have a GenAI strategy, we're going to use this LLM and not this LLM, right? Which is more like a policy. I mean, it's also a very technology-specific policy. And around the time I was trying to argue, encourage people to think a little bit more broadly about—

14:26 Justin Tauber: And why? Why is that important?

14:28 Speaker C: Well, uh, well, for a couple of reasons. One is it's a massively multi-affordance technology. You can use it in all sorts of ways. So people were trying to respond to that diversity of use by trying to say, well, we'll just pick the LLM that's safe, and that way we'll have absolved ourselves of the responsibility of worrying about how it's used. So it's a kind of a technology-first approach to safety, which, well, it doesn't really work. It's like saying we'll only use safe hammers.

15:03 Justin Tauber: Mm-hmm. Safe chainsaws. Safe chainsaws. Safe axes.

15:08 Speaker C: Yeah, yeah. A lot of it's how you use it that makes all the difference. So it's the same.

15:13 Justin Tauber: LLM could be used for, yeah, scenario planning or some really sinister stuff. It's still the same LLM. Yeah. Yeah.

15:22 Speaker C: And that's not to say there aren't better or worse LLMs for each individual use case. But if you start with the use case and what sort of trust you need to defend in that use case, and then which LLM can best accommodate and is competent in terms of that kind of trust, then you're in a better position. But it's so, you know, and so the strategy part is really going, well, what sort of use cases are we going to focus on first and where is the opportunity for our organization and what sort of augmentation or automation or assistance is going to be the most valuable in our organization. Because then you can flush out the use cases and you can start to see patterns in them. And then you can build policies and safety mechanisms and compliance systems that will help keep you safe while you explore those spaces. But trying to imagine that you can keep yourself safe in— across any use case under any circumstances is, uh— Yeah. Is the thing that's just going to get you either into analysis paralysis or it's going to encourage you to do dangerous things.

16:30 Justin Tauber: Yes. You can't just wash your hands of it once you've said, oh well, you said I could use this platform, I'll just do anything in that. There's some thinking required. That's great. All right. What about this one? There's a lot of uncertainty about what the final AI stack will look like, but that doesn't need to be everyone's problem. What did you mean by that?

16:51 Speaker C: Yeah, I mean, there's, you know, I think we're seeing more and more this stack evolve, including tools being built on top of LLMs to add another layer, like Cursor being a great example of that. We have this whole, this whole, including data, you know, data preparation, data harmonization tools in order to make agents, for example, or work more effectively. If you try to make that everybody's problem, like inside an organization, if you make the call center operator needs to understand the data science behind the response that they're getting from, um, from some sort of assistant tool, that's just completely impractical. And what we need to do is lower the, the cost of safety for those people to be able to make it, um, uh, feasible for them to be able to behave and use these tools, these incredibly powerful tools in ways that are both beneficial to them, but also safe for them and for the people that they're serving.

17:50 Justin Tauber: Mm. Trust is about character and competence.

17:56 Speaker C: That's, well, I mean, I really can't take the credit for that one. I think that's Rachel Botsman.

18:01 Justin Tauber: Oh, really?

18:01 Speaker C: Who came up with that framework.

18:04 Justin Tauber: But it's a good one. Also, I personally think that you're also very smart, clearly. I think they, yeah. Right, okay, they just dropped that. They're like, oh, I think he said that, yeah. Yeah, I mean, it's— But you can— You agree with that quote by Rachel?

18:18 Speaker C: Yeah, I think sometimes you assume that trust is about, do I think you have good intentions?

18:22 Justin Tauber: Yeah.

18:23 Speaker C: Right. Which is a very important part of it, but it's not the only thing. It's like you might have very good— Like, for example, if you said to me, look, I'm going to organise your Halloween party this evening. And I go, wow, you really want to do me a favour. You like, I would love to organise a great Halloween party, but you're really not good at organising parties and it's going to be a crap party.

18:43 Justin Tauber: It probably would be, for the record. I'm a very good participant of Halloween parties, less so the planner, unfortunately.

18:53 Speaker C: Yeah, so it's like, I'm not questioning your intentions, but I'm questioning your ability to fulfill them. It's really, and actually I think we, it's this kick I'm on at the moment is interpreting what businesses do in terms of promises. If you just talk about promise making and promise keeping, like, can you— are you making the right promises? Are you intending to keep them? Do you— are you actually capable of keeping them? You know, you can get a long way by just asking those questions.

19:20 Justin Tauber: In this time of AI where we genuinely don't know what the future holds, to your point earlier, is it hard to— to— I'm sure you do have to guide businesses, but how do you guide them when the future has many scenarios in front of it?

19:35 Speaker C: Well, I think the key thing is not to do it for them. Like you need to do it with them. I think there's a really nice way out. We have a futures team inside Salesforce who just focuses on this stuff 24/7, both largely internally for our executive to be able to try and anticipate the future 'cause it's hard for us as it is for everyone else. Yeah. There's no magic crystal ball in San Francisco.

19:57 Georgie Healy: That would be good.

19:58 Speaker C: That would be good. Maybe there's one at the top of Salesforce Tower that I haven't been previously.

20:02 Justin Tauber: You'd put it at the top though.

20:03 Georgie Healy: Yeah.

20:04 Justin Tauber: Yeah, yeah. So everyone can benefit from the crystal ball.

20:07 Speaker C: Everyone just looks up at South Samsara, they can see the future.

20:10 Justin Tauber: Iconic for you guys.

20:12 Speaker C: Yeah. And so one of the things they talk about, which I really love, is sometimes the benefit is not that any of these scenarios that you come up with turns out to be true. The point is that you've had a conversation about what might happen and you've had everyone in the room focusing on that range of possibilities of what might happen and talking about how it affects their part of the organization and what they can do to respond to it or how they could respond to it. It's like, I think people do this with business continuity and disaster recovery planning and things like that. They, they play out a scenario.

20:46 Georgie Healy: Mm-hmm.

20:47 Speaker C: But the point is not so much prediction is not as important as rehearsal. You're rehearsing for the future so that when and if a scenario turns out to be true, that some hybrid of these things or with a slight variations in different ways, you can— don't need to take the time to have everyone build a language for talking about what's going on or understand what they could do or who they need to talk to. And when they talk, they have— they have already a shared language about— they can use to talk about it. It's like that scenario, but it's a little bit different. Like, this, this is how I think we should respond. And it's the speed of the response that's actually more important in the case where something unpredictable happens, a kind of black swan moment.

21:29 Georgie Healy: Prediction is less important than rehearsal.

21:33 Justin Tauber: We need to add that. Yeah, that's iconic. That's the best quote. Brilliant. And last one, an important one, I think, for specifically for your role, Justin. Yeah. Ethics and innovation are not necessarily contradictions. Talk to me about that, because often, often we do think it'll slow you down. To be ethical, like to try and think of the most ethical response. Someone else has already gotten to the solution because they're not weighed down by morals and values.

22:06 Speaker C: I guess there's two parts to that. One is that, well, now I've committed to there being two parts to it and I'm—

22:11 Justin Tauber: There's gonna be two.

22:13 Speaker C: There's just one part to it. Well, let me give you an example, right? Like maybe we just don't see the impact of ethics That's one thing. The innovation that comes out of ethical considerations. Like, I mean, just to give you a really banal example to start with, the guy who invented the bendy straw had a daughter who could not grasp straws. So—

22:36 Justin Tauber: Really?

22:37 Speaker C: He needed to find a way of allowing her to drink soda. I think it was in the States, right? And so invented the bendy straw. And you're like, bendy straws are cool. Like, but he did it out of love, right? Out of love and care for someone who couldn't do— had a mismatch with the way the world worked around her. But even on a grand scale, like, we— you look at LLMs and generative AI depends on a machine-readable internet. There's a reason we have them. There's a reason we have them. And I remember the birth of CSS and what we had before that, and a world of MySpace-like websites it was not machine-readable in any way. But what, what changed it was actually accessibility rules. Accessibility rules forced developers to work out how to make machine-readable websites or force them to adopt the practices. Good practices of writing accessible code ended up being machine-readable code because a screen reader needs to— is a machine. It needs to read the code as a machine. And so it's thanks to like WCAG, for those who remember, you know, the accessibility guidelines, thanks to those accessibility guidelines that we have generative AI because we wouldn't have access to the range of data across the internet that we need to actually train generative AI.

23:58 Justin Tauber: I actually genuinely want to have a little reflection today on innovation born out of ethics. That's, that's an amazing take. I have no good ones. The only thing I can think of is I visited Europe in June. It was lovely. It might be once in 10-year trip. I've got young children. But the lids on plastic bottles now, it's more a sustainability thing than an ethical thing. I don't know whether—

24:25 Georgie Healy: Yeah, yeah.

24:26 Justin Tauber: But they force the lid to stay on because now there's an attachment to the bottle itself. And it was so that the lids didn't end up in— oceans and things like that. But the number of times I've like lost the lid and then I've just got this like water bottle without a lid or it's gotten under the car seat or something like that, it's actually quite helpful. Maybe it's not necessarily a popular take. Some people might like to remove it. But for me, I was like, this was done for sustainability reasons, but actually better usability for me as well.

24:58 Speaker C: Yeah, they always have halo effects. Like the really simple way of keeping it in mind is that everyone suffers from temporary disabilities. When you've got a headache, you've got a cognitive disability. When you're carrying a small child, you are one-armed. Yeah, been there. Yeah, right. You sprained your ankle, you're now, you know, like everyone has temporary disabilities regularly, but for some reason, and if the world's not built to cope with those things, it's, you have a smaller scale of population, a smaller scale of moments that you can address with your technology if that technology is not able to address those moments of disability. And they're the best— the people who know best how to cope in those moments of disability are the people who have to live in them permanently.

25:48 Justin Tauber: That's why you need to be innovative too, I'm sure. Justin, I'm starting to understand your ethical and innovation overlap now. Better and better.

25:56 Speaker C: Well, it's somewhere like, you know, like we've been talking about customer centricity for a very long time. Like the whole digital transformation was focused— digital transformation era was really focused on customer centricity. It's just widening the lens slightly to the people who are around the customer, to other stakeholders who are nearby, or different kinds of customers. It's, it's, it's not that big a step. This is the thing, I think there's one of these things about ethics we kind of like, and, uh, Philosophers are as guilty of it as anyone. And, or maybe just a way of avoiding doing ethics that people go, it's very hard, or you have to get it perfect, or all these kinds of things. And I think it's, it's not that hard, and it's not, it doesn't require perfection. It just requires being a little bit more decent in a little bit more, more situations and constantly aspiring to that and going, what, what's the one thing, extra thing that we can do to make the way we're working a little bit more decent? And, you know, if you keep that attitude, then it's really, it's always easy to find things, and you'll find those extra benefits and those innovations that come with it.

26:55 Justin Tauber: Is it a muscle, Justin? Like, have you— you've done it for quite some time, thinking of the ethical considerations. Have you noticed in your team and inspiring others that, that it, it's something that you can develop over time, that ethical thinking framework almost?

27:10 Speaker C: Um, yeah, absolutely. Yeah, I mean, I— big shout out to the Cranlana Institute, does a lot of this stuff for ethical leadership training and things like that. And one of the nice things about that training that they do is they remind everyone that you have all these skills that you use every day. It's just a matter of pointing them at a slightly different problem. Like, executives have all this strategic thinking and operational— operationalization skills. There's no difference. You can just apply those skills. There's not a special set of skills or a special magical kind of perspective that you need to take to be ethical. It's to be more ethical. I, you know, there's no being ethical doesn't mean anything. It's like, it's also like aspiring to be ethical is like aspiring to be musical. You don't get musical, you learn an instrument. You do one, you try to learn one instrument.

28:01 Justin Tauber: And then you're musical.

28:02 Speaker C: And then you can generalize and go, okay, now I understand.

28:05 Justin Tauber: So yeah, yeah, totally. Yeah. Okay. I love that. We can all, we can all work on this and be ethical. What is shadow AI and what kinds of companies are most at risk of enforcing this?

28:19 Speaker C: Well, shadow AI is really just the idea of the people in your organization are using AI, but in ways that you don't know. And so you can't control firstly whether the kind of AI they're using is fit for purpose, whether it's adding risk to what they're doing, or whether it's even leaking information about your— out of your organization in a way that would be leaking confidential information out of your organization. So I think as one of these interesting stats that we saw when talking to workers, there was a 160% jump. I think I'm going to get that stat wrong. Hopefully that's right. 160% jump in the usage of AI by Australian workers.

28:56 Justin Tauber: Yeah.

28:56 Speaker C: But I don't think there's a— when that's a wreck, that, that, that's by from talking to the workers themselves rather than talking to the businesses where those workers work. And you kind of go, okay, well, how are they using it? Do— does the organization know how they're using it? Have they put safeguards in place?

29:13 Justin Tauber: Very interesting. I interviewed amazing CTO of EY, and she said, you know, they're trying to encourage workers to use AI more and more so that they can help their customers more and more. And there was a stat about, you know, I'm going to get this stat wrong now, like 70% of the workplace is using AI. And she's like, I'm looking around the room, 70% of the room is not using AI. It's making me think of that shadow AI. It's like, well, are they using it for their work and productivity and improving their role, or are they shadow AI-ing?

29:48 Speaker C: Yeah, I mean, you know, it's, I think you want, it's really attention. You want to maximize the amount that people can explore and find for themselves use cases that are actually valuable to them and to their work. At the same time, you want to do that in such a way that you're not adding risk for the organization or for the individuals involved, right?

30:08 Justin Tauber: Yes.

30:09 Speaker C: So, it's about facilitating in a safe way.

30:11 Justin Tauber: Yeah, I think that's brilliant. We are going to zoom out a little bit and I want to get your perspective on some semi-hot topics around AI and ethics. Number 1 being around the LLMs and potential bias being trained into them by the engineers that, that do, do the, um, human feedback, right? How, how concerned should we be about bias within an LLM? Um, I use ChatGPT or Gemini daily. Is that risky? And what, what, what do you think about bias?

30:43 Speaker C: I'm more worried— I'm less worried about the engineers. I'm more worried about the internet. Because really the LLMs are not trained on engineers. They're trained on the internet as a whole. And so recognizing the internet is a weird and crazy place with a whole lot of biases built into it. And those biases do translate through to the LLMs. But LLM itself is kind of this incredibly powerful source. It's the adoption of that LLM and how you build that into your systems and processes that makes a big difference to whether those biases translate into your actual processes or not. Sure.

31:18 Justin Tauber: Yes.

31:20 Speaker C: So to give you a very tangible example, uh, recruitment company, the company that matches job applicants and candidates, right? And, uh, you— we know that it's a high-risk situation to try and use an LLM to do that matching because it's going to make biased assumptions about the job and who's appropriate for the job and biased assumptions about the candidate depending on who the candidate is. So it'll And we know about those kind of biases, especially gender biases, race biases, but even these kind of arbitrary biases that kind of come out of nowhere. Like, I know in the early days we were talking about using machine learning tools, and that was a classic example of machine learning tools actually ending up prioritizing candidates who were closer to the office, who lived close to the office as better candidates.

32:09 Justin Tauber: Interesting.

32:09 Speaker C: Because they tended to stick in the role longer. Because they weren't— they just didn't have as big a commute, which is not a great skill test.

32:16 Justin Tauber: No.

32:16 Speaker C: Or another machine learning tool that ended up privileging candidates who went to a particular Ivy League school or played football for a particular Ivy League school because the organization had a history of hiring ex-Ivy League footballers into roles, which is like just a great mirror. So, so, but what I do know about is like It's really about choosing the right tool for the right part and the right task within that process. So it might not be in the candidate matching that you want to be doing using generative AI to do the matching. You might want to use a very, very, uh, repeatable mathematical function to do that. But in order to get the job application and the job description into a format where you can do that mathematical comparison, you need to use generative AI to translate what's in that job application into that formula. So changing the structure of things is really great, generative, generative AI use case. But the other benefit of not using generative AI on the matching is that you can, you can have transparency and contestability. You can, you can show, if challenged, exactly how that matching was done and what features were matched. And you can also— Which allows someone to say, hey, that wasn't, that wasn't fair. And you can, you can even publish if you need to what formula you're using for that. So and this is why, especially the solution engineers and technical architects in my team, the thing that they've had to specialize in is not generative AI or agentic AI or predictive AI, but in analyzing and and breaking down use cases in such a way that they distribute the right part of the job to the right kind of technology. And that's, you know, where bias is going to be highly impactful. Don't rely solely on a generative AI tool to be able to do it.

34:17 Justin Tauber: Interesting. I love that.

34:18 Speaker C: And we've even been developing, I think this is the other thing, is there's also places where probabilistic tools are useful and appropriate and places where deterministic tools are much more appropriate, where you want to go back to kind of traditional programming structures. So where you want a thing to do the same thing every time, don't use it, don't use an instruction for an agent to do that.

34:41 Justin Tauber: I'm going to put you on the spot. Can you give an example of a deterministic?

34:46 Speaker C: To give you an example, well, a great example that we see coming up over and again is people try to give instructions to agents like when you get this kind of complaint, the first thing you do is this, and then the second thing you do is this, and then the third thing you do is this, and then you want to send it off to this, get this information, and then you want to do that.

35:05 Georgie Healy: Clear workflow.

35:06 Speaker C: One, doing one, one thing after another in a really specific order. It always gets a little bit more complicated as soon as I think about it, but just having a step-by-step workflow like that especially when that workflow gets quite long, agents can get confused with a very long set of steps.

35:26 Justin Tauber: Long horizon tasks. I learned this word recently.

35:30 Speaker C: Long horizon tasks. Yeah, that's right. That's even— But it can be just the normal steps.

35:33 Justin Tauber: Patting myself on the back to be able to put that in a sentence. Yeah.

35:36 Speaker C: Yeah. I mean, you know, we know that LLMs aren't great at like counting the number of Rs in strawberry.

35:42 Justin Tauber: Yeah.

35:43 Speaker C: And things like that. They're very good at creative and creative, creative activities, translation activities, generation activities. But, um, but where you want something that's more logical, are there neural networks? I mean, the thing to remember is that we— the whole neural network and the precursors of deep learning and machine learning and now generative AI was coming out of the failure of a different kind of AI, which was symbol processing AI, which was based on trying to codify the logic of everything that we do in our everyday lives. And that failed. But that doesn't mean that there wasn't some goodness in that. Yeah. And so we need to— we actually probably need a combination of both kinds of approaches. And that's just to ensure that when we're trying to do logic, take logical steps and work as humans, we're great at this. We're great at this switching between logical mode and non-logical mode.

36:43 Justin Tauber: It's funny hearing these long horizon tasks and deterministic, even booking a restaurant, you'd think there would be a set series of steps.

36:49 Speaker C: Yeah.

36:50 Justin Tauber: Call the restaurant, ask for a booking on this date for 2 people. But when they're like, is the bar okay? And it's like, well, I don't know what the bar is. Like, we don't have restaurant seating, but we do have this. And they're like, ah, I don't know what that means. Whereas normal human would be like, is the bar okay? 'Actually, no, that's not what I wanted.' Yeah, yeah, yeah, yeah.

37:13 Speaker C: Ironically, that's one of the things that, that agents, I think, are best at, is when the customer, for example, doesn't know exactly what they want and they're trying to explore the options. That's where they incredibly outcompete older technologies like chatbots.

37:25 Justin Tauber: Yes.

37:26 Speaker C: Um, especially, again, tree-based deterministic chatbots required you to start at a certain point and end at a certain point. But if you're coming in and you don't know what the options what options are available to you, and you're trying to explore the options that are available to you, you need a nonlinear process, an adaptive process. So a process that's going to adapt to the conversation that you're having rather than a process that's preset. So it's horses for courses. It's just a matter of being able to combine those things in the right kind of way.

37:55 Justin Tauber: Temptation to talk about AlphaGo and go deep on that is very high, but—

37:59 Speaker C: Right.

38:00 Justin Tauber: But I've got questions I am dying to ask you. Privacy in what we put into the LLM.

38:06 Georgie Healy: Yeah.

38:07 Justin Tauber: How ethically dodgy is this? Like, we don't know how much consent we're giving the LLM, who's gonna see the conversations we're having. I'll speak for myself here, Justin. I do talk about like emotionally charged topics that I'm probably not wanting completely broadcast to all of OpenAI necessarily. How do you feel about how people interact with an LLM from a privacy and ethical standpoint?

38:33 Speaker C: I think it's a very difficult— I mean, everyone has their own idea of what they're willing to share. I think the thing where it becomes an ethical challenge, if you like, is where people don't have options. So for example, there was a story recently about— actually, a friend of mine was in involved in it where she took her daughter to a specialist and the specialist had basically one of the new transcription services that would allow that, that would listen in to the specialist appointment.

39:10 Georgie Healy: Yes.

39:11 Speaker C: And then convert that into an update to an electronic medical record or a case note or something like that, right? And she said, Well, she called them up and said, look, uh, it's because it's a child, it's my child, I'm not comfortable with that being recorded and transcribed and processed by this AI tool that you're planning to use. And she was told to find another specialist, which is where you start to tie—

39:38 Justin Tauber: Huh.

39:38 Speaker C: Uh, compromising or making a particular, a particular call about what privacy is acceptable with access to health services, for example, then, then it starts to get—

39:50 Justin Tauber: That's a fascinating case because I'm a weird example. I'm prolific on social media. My husband and I made the decision to not have the children's faces and stuff on social media. Completely understand both sides. That's just our personal take. Because I'm like, I know I'm an oversharer. I would not want my mum to do that to me as a child. Yeah. And it's still the same tool. Yeah. Okay with it for me. Not okay with it for them. Yeah. Yeah. I think that's a fascinating example.

40:21 Speaker C: I mean, I think that's a good example of where you're going. You even might be resistant because they're not competent to make the decision for themselves, but actually you're also not competent to make the decision for them because it has far-reaching effects.

40:36 Justin Tauber: Yes.

40:37 Speaker C: Whereas, you know, that's one of the reasons why you might be cautious in that situation. Yeah. At the same time, I'm watching my boys who are both teenagers really having to, I mean, my God, they've been through a lot in the last 6 years.

40:48 Justin Tauber: Yeah, those poor, not a great time to be around in COVID and the social media.

40:55 Speaker C: And a shout out to my son Gabe who just finished his HSC yesterday. So he's just—

41:00 Justin Tauber: He's loving life now.

41:02 Speaker C: Yeah, he started year 7 during COVID which is like, you mentioned starting year 7 during COVID and then you have the social media, rise of social media and all those kinds of things.

41:11 Justin Tauber: We didn't know how to grapple with it. Like we don't really have that much data on this stuff. Still.

41:15 Speaker C: Yeah. Yeah, it is. And it is very difficult for everyone to make a call about what the appropriate level of privacy is, I think. Uh, and so, but having the right to make a call for yourself is almost the more important thing. Yeah. Because, and providing support, again, it's maybe it's like character and competence in a certain extent. It's building, building your own perspective on it, but also having the capacity to actually enforce the values that you want to enforce is really important as well. And that's, and that's where having options like being able to fall back to a human where you don't, don't want to deal with an agent, for example, is really important. And having those transitions be seamless, I think, are underappreciated. We just focus on the power and excitement of the new thing. But ultimately, ultimately, maybe another way of putting it is you would talk about ethics and innovation. It's like often the ethics happens in the adoption rather than in the actual innovation itself, especially where innovation is just, hey, we've got this new thing, there's new capacity, new thing we can do. Yeah. How we use it is, is really important.

42:23 Justin Tauber: Yes. I love your take as well on even both as adults, both using an LLM. I might be completely fine telling the world that I'm 'Going to a Halloween party tonight,' and you'll be like, 'That's actually quite personal, and I don't need people to know about my amazing Halloween party tonight.' Two people just with very different perspectives on what is fair game.

42:48 Speaker C: Yeah, I mean, if you went, 'I'm going to a Halloween party tonight, and here are all the people who are coming with me.' Ah, yeah, yeah. See, there we go. That's a very different thing.

42:57 Justin Tauber: Very different thing, isn't it? Definitely. Grey area in terms of age and certain perspectives on what we care about.

43:08 Speaker C: And I mean, there was a whole— people start developing strategies as well. I think it's interesting that people develop strategies to kind of manage that balance as well. Like, I know at one stage lots of teenagers were deleting and recreating their social media accounts, like, almost continuously.

43:27 Justin Tauber: Really?

43:27 Speaker C: To try and stop the acquisition or the accumulation of data about them.

43:32 Justin Tauber: So teenagers are smarter than me. Yeah. They know what's up. Yeah. Amazing. We are going to dive a little bit more into agentic AI. This is something I know that you're very passionate about. Salesforce have their annual conference called Dreamforce. Obsessed with the name. I'll quote your CEO. We're entering the era of agentic enterprise. Humans and AI agents are the new operating model. How do you see humans and AI agents working together in a way that excites you and makes you, with your ethical standpoints, really happy?

44:07 Speaker C: I think there's lots and lots of ways that the humans and agents can work together. I think there's, I don't know, I was trying to think of a good analogy for this the other day. This might be a little bit nerdy.

44:18 Justin Tauber: Okay, very nerdy analogy. Please, I love a nerdy one.

44:22 Speaker C: I've been coming up with all these analogies, but—

44:24 Justin Tauber: Shower thought analogies. Yeah.

44:28 Speaker C: So like, because I'm often asked to kind of differentiate what's different about agentic AI, like agentic AI is a very— is actually should be treated as quite a different thing. People assume it's the same thing as generative AI and generative AI definitely powers agentic AI, but they're different and they're both different from predictive AI, which is predictive AI was all about classification. And so I have this musical analogy. Which goes back to Beethoven. So Beethoven sat down and he basically, he wrote this thing called the Well-Tempered Clavier, which basically defined the musical notes, the relationship between musical notes that we have in, that we use every day from C to C.

45:11 Justin Tauber: Yes.

45:11 Speaker C: On the piano, that's, he basically tuned the piano in the way we tune it now. And so every other note that we talk about in Western music is tuned according to his pattern. So what he did was he took a continuum of music and he allowed us to find, to differentiate and classify things within it. So, sorry, a continuum of sound and basically classify each of those bits of sound into a discrete note. And that's exactly what predictive AI does. It takes a continuum and allows us to classify things into discrete groups. Generative AI, by comparison, is more like the electric guitar. Electric guitar is the moment an electric guitar comes out, especially with distortion—

45:53 Georgie Healy: Oh yeah.

45:54 Speaker C: —is a way of mucking with sound and actually making it worse in some ways. It's like less predictable.

46:01 Justin Tauber: According to my mother, so much worse.

46:03 Speaker C: Like Beethoven would be mortified.

46:05 Justin Tauber: Yeah. Yeah.

46:06 Speaker C: But it unlocked this huge expressive capacity. It's by having this randomness injected into the playing of music that we managed to find all these different songs and all these different ways, all these different musical opportunities. Within it. I think generative AI is like that. And, and GenAI is like a drum machine. You can set it and it will operate without you. Now you can— so you can have two sorts of relations with that drum machine. You can go, I'm going to get that drum machine going and then I'm going to leave the room and just— and I know it will keep going and I don't need to worry about tempo. That's one way of looking at it. The other way of looking at it is like— and the way I would use a drum machine is like, well, I'm just going to jam along to the drum machine. Like, I'm going to riff and I'm going to, you know, have a layer of creativity over the top of that, over the top of that drum machine. And so I think that's a nice way of—

46:57 Justin Tauber: That's a beautiful way.

46:58 Speaker C: Thinking of the relationships, like what things would you like to be able to set and kind of not need to manage, you would keep going so you can focus on the things that are actually more expressive of yourself or more are going to communicate something in a certain way or allow you, you know, do something that's uniquely human over the top. And so, you know, we shouldn't see it— and so you don't— a drum machine doesn't replace music. It just allows people to build layers and layers. And, you know, let alone drummers come in and they produce layers and layers of sound over the top of that as well.

47:36 Georgie Healy: It also reminds me—

47:37 Justin Tauber: I love this analogy of you can create more output, or, you know, I don't need to hire a drummer, use the drum machine, and I can riff off that. With the workforce, you know, I can be more— I can be a better employee. I can do more. I can achieve more of my critical thinking if I'm not admin calendar ops all the time.

48:00 Speaker C: Yeah, but you need to know which songs are appropriate to have a drum machine on and which songs you need live drums on. I mean, True. And you know, and that requires some thoughtfulness.

48:10 Justin Tauber: Yes.

48:11 Speaker C: Really. But you know, I mean, maybe that analogy only goes so far, but, but yeah, anyway, it's one—

48:18 Justin Tauber: I loved it. I really loved it. We have, we have a lot of startup founders on the show.

48:24 Georgie Healy: Yeah.

48:24 Justin Tauber: And the complexity of scaling AI in enterprise, I feel like is scary because of the impact it can have. But for startups that build fast and break things, you know, people say like it's wild, wild west out there. They'll build something that, you know, we can't control. Yeah, you can't. But I'm going to ask you to tell me which is scarier or riskier.

48:49 Speaker C: Startups or—

48:50 Justin Tauber: Yeah, big enterprise when adopting AI.

48:53 Speaker C: Yeah. Yeah. I mean, it's kind of a really nerdy way of thinking about this, I suppose, but kind of depending on which ethical framework you're kind of taking to the problem, like if you're, you know, this kind of duty-based ethics or deontology, they call it, which is really you have certain duties and those apply regardless of scale, right? So from that perspective, moving fast and breaking things is, is kind of worse because you're not adopting all the responsibilities you have. At the same time, often when you're, you know, but from a utilitarian perspective, for example, where it's about maximizing the happiness or minimizing the harm, overall, then you're more worried about what happens at scale, right?

49:33 Justin Tauber: Context. But I don't know whether that's useful. That is useful. No, it is. You broke down the different ways that things can, can get murky.

49:42 Speaker C: But they— I think the other thing to add on that is really there's, uh, there's kind of, uh, things change with scale. So what might be appropriate for a startup to try and experiment with. At a certain point, you have different sorts of responsibilities as you get larger. And one of the interesting things about working with enterprises is that they are quite constrained by regulations and things like that, and they've grown large in a context where there is quite a lot of regulation and compliance, and they need to accommodate that already.

50:14 Justin Tauber: So, um— Listening to the history of Google is very interesting for that alone. Yeah. And even DeepMind, which now is in Google, it of trying to work within the corporate beast. And yeah, to finish our chat today, I'm going to reward you with spicy rapid-fire questions. How does that sound? Okay. Firstly, what is one area of your life, Justin, where you're like, I want to lose the guardrails in this area? This could have— this is where I do not want to have too much framework and thought and—

50:52 Speaker C: Yeah, I mean, I don't know. I was struggling with this one 'cause I feel quite, I feel like I have a lot of freedom and autonomy to kind of work out what the appropriate things to say are. So, and what the appropriate things to think are and how to structure my work. I'm very lucky in the sense that at Salesforce we do a lot of natural continual job crafting.

51:12 Justin Tauber: Really?

51:13 Speaker C: Where we continually are kind of re-articulating or redefining our roles as we go. Most people don't have that kind of— No. Freedom. And I think it's, um, it's underappreciated how important it is when things are changing to have that freedom to job craft for yourself and have the autonomy to job craft. Um, but I mean, I don't know, I just, like I said, Gabe's just finished his HSC, and just watching the joy that comes with being free of those, the guardrails of a curriculum, of subjects that you don't have a lot of choice over, all those kind of things.

51:46 Justin Tauber: Do you know what he might do? After school?

51:48 Speaker C: Yeah, he's gonna, I think he's gonna go into ancient history.

51:51 Justin Tauber: He's like, modern history is too modern for him. Oh, so cool. So cool.

51:54 Speaker C: Yeah, going deep into ancient history. But he's been doing this, like I remember when I was at school, we had 6 column inches on a topic in the Encyclopedia Britannica. But he's been, he's had thousands of hours of YouTube videos on the Assyrians and the Babylonians and Alexander the Great. And he just wants to go deep, deeper and deeper into that world.

52:13 Justin Tauber: I wonder how AI will impact that. Like, can we render the, the Library of Alexandria or something like that. Wouldn't it be amazing?

52:21 Speaker C: I think there was an amazing use case for— they used a machine learning tool to take this scroll that had been curled up and was ancient, very, very ancient scroll, and actually read what was on it without unrolling it. No. Which is pretty cool.

52:38 Justin Tauber: That's super cool.

52:39 Speaker C: Because by unrolling it, you could destroy it. You could not unroll it without destroying it. So—

52:43 Georgie Healy: I'm going to ask you an ethical question.

52:47 Justin Tauber: The Louvre robbers that stole those artifacts.

52:51 Speaker C: My wife tells me they're very attractive robbers.

52:53 Justin Tauber: Were they?

52:53 Speaker C: Everyone's talking about how attractive the Louvre robbers are.

52:56 Justin Tauber: Is that because they were wearing like construction outfits? Maybe there's some like, you know, maybe that's helping.

53:01 Speaker C: No, no, it's quite hilarious.

53:03 Justin Tauber: Actually attractive. I've lived in France for a year and did not see anyone that attractive. So, yeah. So, you know, missed a trick there. Do you want the jewels to be found again and returned, or are you kind of like, fair game, they— there was no security, they should learn their lesson. Maybe they'll be a bit more careful.

53:29 Speaker C: Yeah, right. Well, I don't really know about that. I think, yeah, I don't know. I think the jewels probably reasonably belong to the whole French people rather than just those guys.

53:39 Justin Tauber: I'm actually quite sad about it, if I'm honest. It is quite sad.

53:41 Speaker C: But what's the interesting thing I learned was that the Louvre actually does not insure its collection because it can't. It would be prohibitively expensive to insure it. So instead they take the money that they would have spent on insurance and invest in their own security systems.

54:02 Justin Tauber: Not enough.

54:03 Speaker C: You're just like, well, yeah, absolutely. But that's quite interesting, isn't it? Like, it's so valuable that it's uninsurable.

54:10 Justin Tauber: That is fascinating. Oof.

54:14 Georgie Healy: Yeah.

54:15 Justin Tauber: Might need to have a side hustle coming. No, I'm not. I'm not, for the record.

54:18 Georgie Healy: You're not?

54:18 Justin Tauber: I'm not gonna—

54:19 Speaker C: Stealing jewels? Yeah.

54:20 Justin Tauber: Is that—

54:20 Speaker C: Quite cool though. Right, Cat Burglar.

54:21 Justin Tauber: Yeah. Good Halloween costume.

54:23 Speaker C: Good Halloween costume. Not a great profession.

54:26 Georgie Healy: Maybe not.

54:27 Justin Tauber: Maybe not. Maybe doesn't scale the way I would like it to. Right. Okay. Right. You could only really do one.

54:34 Speaker C: Right.

54:35 Justin Tauber: That's true. It's not Ocean's Eleven.

54:36 Speaker C: Maybe that is your secret. Your secret.

54:39 Justin Tauber: Secret. Now I've ruined it. They're going to be like using AI to, to facially map my— All right. Is Australia slow and lagging when it comes to AI?

54:53 Speaker C: I look, I don't— I think it's too early to say. And actually, slow and lagging for what? Like, if— Should we really be going after building our own foundation model or something like that. I'm not sure. That requires an immense amount of data. The scale of the investment is not the kind of scale of investment that Australia could probably do. Will there be opportunities in the future to create, I don't know, an Aussie Lingo LLM or something like that to help Australian businesses actually de-bias other LLMs, for example, or make them more appropriate to Australian context or something? That might be appropriate. It's about finding the kind of role that Australia will play in this new economy. Um, Singapore— in 1984, Singapore, uh, was like— had failed, had failed in an attempt to modernize its workforce and stuff like that. And it was only had 70% of the GDP of the US per capita GDP. By 2024, it had 170% of the GDP per capita of the US. And that's because they found a particular spot in the development of global shipping with the containerization of shipping and the global logistics happening. That, that started well before Singapore got involved, but they found a particular role. They were kind of late to that party. Containerization started in Vietnam War, for example. They were late to that party, but they found a unique role that made them invaluable to the flow of goods around the world and so became incredibly wealthy as a result. I think the right question to ask is, what is Australia's role in this new AI-powered economy that we're entering into? And, you know, we have this reputation for doing regulation, like for example of our banking sector through the GFC. We got a reputation for kind of like regulating our banks really well and making them safe from things like, from risky investments and things like that. We also could be a country that is known for safety products, for safety features, or even more ethical AI, or enabling the adoption of AI in more situations.

57:09 Justin Tauber: Like we did with social media, first country to increase the age ban of—

57:14 Speaker C: Exactly.

57:14 Justin Tauber: Yeah, that is really interesting and something that I'm really proud of. Right. Yeah. Same with gun laws and things like that where it's like, I'm really proud of the— it's innovative in a way to think future forward about the impacts. Totally is.

57:29 Speaker C: It's like saying, well, wouldn't it be great not to have to worry about that problem? Like, you know.

57:34 Justin Tauber: Yes. That's actually— It prevents me moving to the US if I'm completely honest.

57:38 Speaker C: Right. Exactly. So, you know, and I think we talked about it in passing as well. Like, you look at the EU, which has gone hard on their AI Act. And people say, oh, that's going to slow down the EU. It's like, I see that as a market for safety products. Yes. That's who we'll sell them to.

57:57 Justin Tauber: Yes. Oh my goodness. What did they say in there? If I can find it quickly. The EU AI Act is a strict risk-based and enforces significant penalties for non-compliance. Not necessarily a bad thing, Justin.

58:12 Speaker C: No, I think, but I think what they do really well is they help us categorize risk. They help the rest of the world categorize risk because we can represent it. We might have a different view on which risks are acceptable in each country, but having categories of risk is really useful for everyone to be able to speak the same language when they're talking about how they're— what sort of risks they're able to take responsibility for. And every product and service takes responsibility for some kind of complexity or some kind of risk on your— like even your toaster. Your toaster takes care of the risk of electrocuting yourself while heating up bread. So true, you know what I mean?

58:48 Justin Tauber: Like, you don't even think about it though, do we?

58:50 Speaker C: No, no, we don't. But that's exactly the job of the toaster manufacturer, to ensure you don't need to think about it.

58:55 Justin Tauber: We know, we made it that way.

58:56 Speaker C: And that's the benefit, that's the benefit that they're offering you. So, um, I, I think there is a whole range of opportunities for us to, um, to explore there, which we might be very well placed to do because, I mean, we used to play this role like 100 years ago, like with you know, suffrage and all these kinds of innovations, social innovations, which actually was, you know, if I compare this time to any other time in history, I'd say the Belle Époque, like 1870 to 1910, is like a good example. Explosion of technologies, social transformation as well as technological transformation. What role do we play in that? We, we were leading the world in social innovation. So why not do the same here?

59:41 Justin Tauber: Beautifully said. This has been an absolute pleasure, Justin. Thank you so much. Um, I would love to give you the board now to shout out to where people can find— I really recommend your LinkedIn, by the way. You post on there some really thoughtful thinking, especially in this time of AI. Uh, where can people find you? What would you like to, to close with?

60:02 Speaker C: Uh, where can people find me? I guess, yeah, you can find me on LinkedIn, absolutely. I also publish a Substack from time to time. But yes, if people want to go deep, that's a very philosophical Substack warning label attached. But yeah, and also just please come along to Salesforce events and let me know if you're coming.

60:23 Justin Tauber: I will be there. Thank you so much for joining In the Blink of AI.

60:27 Speaker C: Thanks for having me, Georgie. It's a pleasure.

60:30 Georgie Healy: Thank you for listening to In the Blink of AI this week. I wanted to shout out Shout out to 3 special commenters this week. I read these on the YouTube channel. Arafat Tassin, Sail Pros, and Mick Hewer. Love your work, guys. Thanks for commenting. If you want a shout out on the show, make sure you engage with me so I can see it, I can interact with it, and frankly, I can continue to make the show exactly how you guys are responding to it. I will see you next week, and I will be keeping an eye on the socials. I will see you all there. Bye.