See show notes for this episode: S 02 | Ep 51 To Master the Machine, Get Amazing at Being Human.
0:19 Alex Shevelenko: Today's guest, Paula Goldman, is at the center of the defining question of our AI era:
0:26 Alex Shevelenko: How do we get the enormous upside from AI without losing our humanity?
0:32 Alex Shevelenko: Paula is Salesforce's first ever—and I'm gonna read this title here—Chief Ethical and Humane Use Officer, where she has helped to shape the company's approach to AI,
0:46 Alex Shevelenko: and trusted AI in particular.
0:48 Alex Shevelenko: She's also an author of the upcoming book, Manage the Machine: How to Harness Human-AI Collaboration at Work.
0:57 Alex Shevelenko: And most importantly, she's a dear friend and an amazing human being.
1:01 Alex Shevelenko: Paula, so excited to reconnect here and get your thoughts... well, how can we save ourselves and empower ourselves at the same time?
1:13 Paula Goldman: Well, it's such a pleasure, Alex. It's been a minute, and it's always really fun to have these conversations.
1:19 Alex Shevelenko: So, tell us a couple of things, right? Your role is not a job role that exists in any job description ever.
1:30 Alex Shevelenko: I think you're the first of a kind.
1:32 Alex Shevelenko: What have you learned from shaping the AI dialogue around trusted AI at one of the most important enterprise technology companies of our time?
1:46 Paula Goldman: Yeah, it's funny because... well, first of all, congratulations on saying the title correctly. Not an easy feat.
1:53 Alex Shevelenko: I had to read it! I had to read it very slowly, split my IQ level. Thank you very much.
2:00 Paula Goldman: So, you know, the role was actually created something like seven years ago, and I have to credit the leadership of Salesforce—obviously Marc Benioff, but the executive team at the time who really understood that technology is moving super fast and that the blueprint for how to use it well isn't necessarily written in stone somewhere. So how should we think about that, knowing that as an enterprise business, trust has always been the number one value, right?
2:35 Paula Goldman: Our technology has to work like we say it does, or our business is at stake. And so, the answer to that question was really about process and rigor, and finding a way to actually raise the questions internally.
2:51 Paula Goldman: I guess I'll say two things. One, it's interesting that while no one else has this unique title, there are more and more trustworthy AI and responsible AI jobs going around, like at the frontier labs and elsewhere. It's really awesome to see colleagues and to learn from them at other companies.
3:10 Alex Shevelenko: In fact, they're hiring philosophers, right? I heard like, finally philosophers can get out of Nietzsche-land and into the real world.
3:20 Paula Goldman: Well, I mean, I have to say, my mother was not thrilled when I went to get a PhD in anthropology. She's probably happy... well, I'm not so sure, but here we are, right?
3:33 Paula Goldman: I feel like it's actually part of the work, right? A lot of the work is very technical, and I have a really cool cross-disciplinary team where I have data scientists, engineers, political scientists, and I'm an anthropologist. The cool thing is that you kind of need to think about these problems from all these different angles to get the right answer. Part of it is about culture. It's about getting the product right, but it's also about creating a culture internally where people are asking the right questions—and that can't just be me or my team, it has to be everyone.
4:14 Alex Shevelenko: Hm, that's beautiful. Well, I have to echo your sentiment about Salesforce, both on trust and looking ahead.
4:18 Alex Shevelenko: As you know, I worked at Salesforce at one point, and then we had the good fortune of being a partner in the early days. Someone from the partner team said, "Hey, I know you guys are pretty small—a startup, two guys and a dog." In fact, we didn't even have a dog, we just had two guys. But they said, "We could use that thing in us, so why don't we try to use you to improve our human user experience for our customers and partners?"
5:00 Alex Shevelenko: Eventually, after passing through probably the most rigorous trust and security review process that we had yet to pass back in those days—which was impressive—we deployed. It was a true commitment to get it done.
5:21 Alex Shevelenko: It ended up creating a great partner experience, and then Salesforce went ahead and actually communicated to the partners about how they care about the partner experience and how they work hard at simplifying it.
5:41 Alex Shevelenko: They were almost running ads on LinkedIn to do that, and I was just blown away. Those ads and whatever they did were better than anything that we've ever done to this day for our own marketing—just marketing to their own partners, not even prospects or customers, on how they want to create a better experience for them.
6:03 Alex Shevelenko: I thought that was a legendary example of a mature company that may not need to do that work as hard, right? But they still had the motivation on the one hand to seek out innovation, and on the other hand to put that innovation through the wringer to ensure that nothing would go wrong, and then bring that to improve the human experience of the partners where they really cared about who you are and what you want to accomplish. Those are obvious questions that we rarely actually go through.
6:45 Paula Goldman: I love that story. It's such a cool story. Thank you for sharing it.
6:50 Paula Goldman: It made me think of so many things, and there's a story in the book that kind of echoes that. Before I tell it, though, I wanna say I think everything you're talking about is becoming more, not less, important as AI grows more powerful. As everyone's personalizing messages and automating marketing and how you reach humans... obviously your work also needs to be legible to AI, right?
7:20 Paula Goldman: That's its own challenge and its own opportunity. But how you reach people and really connect with them makes a massive, massive difference, right? And how authentic that is, how the message stands out, and how you deepen those relationships.
7:37 Paula Goldman: So, the story—I have a chapter in the book that's about AI and marketing. The story I went really deep on is actually a well-known story: it's about the Harley-Davidson turnaround in the '90s, right?
7:51 Paula Goldman: I started with the history. Harley-Davidson was down and out. Japanese competitors... you know, they basically were a breath away from doom. They realized that they had this one really important asset, which was their community—the community of loyal customers. They went really, really deep on building these Harley communities.
8:24 Paula Goldman: I talked to the woman that was in charge of this strategy that ended up creating all these incredible, monetizable experiences—whether it was a museum that had travel to it, or all these different things that the Harley Owners Group (HOG) were creating for these communities of enthusiasts.
8:43 Paula Goldman: The important part, though, was that it was centered on the people, like these dealerships and local businesses, all of which were part of this network of partnerships that turned the company around. And it was that which turned the company around. Now, fast forward to where we are with AI, right? Think about how much—
9:10 Alex Shevelenko: Hold on, Paula, was that before people were getting Harley tattoos?
9:13 Paula Goldman: No, that is what caused people to get Harley tattoos!
9:16 Alex Shevelenko: So even before then... so before then, people weren't getting Harley tattoos?
9:20 Paula Goldman: Yeah, no, this is like... when you think about Harley, Harley is not just a motorcycle brand to people—it's an identity. It's a source of pride.
9:29 Paula Goldman: The person that I talked to, a woman named Laura Lee—who's now doing a whole bunch of boards, but at that time led their enthusiast services—talked to me about how she would get letters from people. There would be a letter from a mom whose child had been killed in service during a war or serving in the military, and she would send a photo of learning to ride a bike for her son. Or a woman who really wanted to learn how to ride a bike... so they created enthusiast groups for women that wanted to ride motorcycles. She camped out at a store in Wisconsin and would get these fan notes from them, right?
10:14 Paula Goldman: It's this kind of visceral enthusiasm, this connection—this idea that the company is serving a part of your identity and your aspirations. The Harley Owners Group allowed people to have access to leadership opportunities and business opportunities that they wouldn't have had otherwise.
10:36 Paula Goldman: I would argue the same thing with our Salesforce ecosystem, right? Our Salesforce ecosystem is built on the strength of the business opportunities that it creates across the board—whether it's small businesses, large businesses, or people that are Salesforce admins at different companies, right?
10:54 Paula Goldman: And I have to—I'll add a little plug, because I met an admin who has a tattoo of Salesforce!
11:03 Alex Shevelenko: What?!
11:04 Paula Goldman: Yeah! Which is kind of even more impressive for an enterprise software brand, right? But it was the community. Somebody helped her get a job when she was going through a tough time.
11:15 Alex Shevelenko: And we all know that the Salesforce Dreamforce event, which is coming up, is like Disney for B2B.
11:27 Paula Goldman: That's a great type of experience and way of describing it, yeah.
11:32 Alex Shevelenko: So it has this... there's a lot—some of it is probably deliberate, right? Some of it is deliberate, but genuinely people connect. I think a lot comes from Marc and promoting a community—a community of communities. It almost feels like that as well.
11:58 Paula Goldman: That's right, yeah.
12:01 Alex Shevelenko: So what I'm hearing is that this Harley story is kind of going to be essential for all of us.
12:04 Paula Goldman: I think so. And it'll look different for different businesses, but okay, so here we are in this world where people's attention is already inundated. It's not just businesses that have access to AI agents—so does your customer, so does the consumer. They're going to have the ability to filter things out, too, right? So the authenticity of how you build these relationships and how you deliver these messages...
12:42
Paula Goldman: ...and really make them shine for people makes a massive difference, which is why I think what you're doing, Alex, is so cool and even more and more relevant as this challenge becomes bigger and bigger.
12:55
Paula Goldman: And I think it will also look like things like Dreamforce. It'll look like things where you're curating community for people. It'll look like things where both the AI side and the human side will have to come together to continue to be competitive.
13:13
Alex Shevelenko: Yeah, it's so interesting. You know, we were chatting about industries that particularly require trust. Everything requires trust because we're inundated by a lot of noise, right? But then there are these regulated industries or functions—investor communications as a function, or insurance, pharma, and health as a service, infrastructure... things that are very much observed by the government, right?
13:46
Alex Shevelenko: So they have this really amazing challenge. On the one hand, they need to provide evidence, especially in the world of easily created blah, right? You need substance. You need something that feels like there is 200 pages of gold and evidence underneath whatever it says—there are years of experience.
14:19
Alex Shevelenko: But if you show up with the 200 pages of that gold, it's not gold, right? It's like, no... maybe back in your anthropology PhD days, you were able to crank out these books, right? But some of us, like me, were never able to do it in the first place. It's certainly harder now when we're bombarded with Instagram and distractions everywhere.
14:44
Alex Shevelenko: So on the one hand, it feels like we need an experience that starts easy—meets us where we are—but that allows us to go down the trusted path that is relevant for us. Or maybe it's guided a little bit, because a really thoughtful communicator will guide you on a journey. Like the best teachers, they don't just improvise, right? They kind of have a plan of how you select what's relevant for you or not.
15:19
Alex Shevelenko: Part of being trusted is being able to say, "We're not a good fit for you, right? This is not us." But being able to do that while having AI working in the background—getting people the right information through enabling agents to read your brilliant thought leadership and know that it's trusted on the one hand, but then when the humans get there, having the journey adapt to the way they want to consume at that particular point in time on their device—that's not trivial.
16:00
Paula Goldman: No, it is not trivial.
16:02
Alex Shevelenko: It's not trivial. For many, it's essential, right? Like, that's gonna be your book, right? That's your book—it's like, we need the substance. Get the right stuff to the machines so they help us, but then when I, the human being, land there, I don't want some machine-generated BS that is pseudo-personalized, because I'm going to have a scan for that.
16:26
Paula Goldman: You'll know, you'll know, yeah. Exactly.
16:29
Alex Shevelenko: That's exactly right, yeah.
16:29
Paula Goldman: Yeah, and I think you mentioned a few challenges that I think are really interesting. So one is, particularly for regulated industries, but really for everyone: how do you have that sort of governance that ensures not only that what you're saying or sending out is accurate, but that you're not tripping over things that could either get you in trouble or be really embarrassing?
16:50
Paula Goldman: As you know, there's been no shortage of stories about things going wrong. Way back in the day, pre-this wave of AI, there was a fast-food restaurant that automated their marketing and sent out a promotional email to their German customers celebrating Kristallnacht.
17:14
Alex Shevelenko: Not intentional.
17:17
Paula Goldman: Not intentional. And you sort of see the modern-day equivalents of that with AI. It's interesting because specifically when it comes to content governance, there are ways to use AI to set the right parameters in advance. A lot of what my team works on with our agent-first team is: how do you create essentially a control pane where the person setting up an AI agent is able to say, "Here are the topics; please don't go outside these topics"?
17:52
Paula Goldman: And then, "Here's the tone," and also, "We're going to set a filter for toxicity, and we're going to set a filter for these issues that might be specific to our particular product and the particular sensitivities that we may not want to stumble onto." So that's an interesting challenge.
18:12
Paula Goldman: I think that's a solvable... it's not totally solved, but there are a lot of good controls around it. I feel like part of it is basically what we're saying: you're mixing deterministic and probabilistic.
18:34
Alex Shevelenko: Exactly, yes. Are we gonna nerd out on that? Should we nerd out on that for a second?
18:38
Paula Goldman: Well, yeah, for folks that are not familiar with what those are, why don't you explain them? I'm sure you'll do a better job than I would.
18:46
Alex Shevelenko: Yeah, I mean, I think that there are many, many amazing things about this wave of AI, and typically when people talk about this wave of AI, they're talking about essentially this wave of frontier models that are not... they're not gonna give you the same answer to your question if you ask it twice.
19:09
Alex Shevelenko: That's what's called probabilistic. They're kind of just... there's going to be some variation in what they're going to create. And that's actually beautiful when you're using it to augment your own creativity, because there's going to be a little bit of chance that goes into it, versus how we thought about software in the past where there is a right answer and it's binary.
19:28
Alex Shevelenko: That's why you had all the old flowcharts of the past—like the whole customer service trees that would be like, "If they say A, get them B." But those were also very frustrating because they couldn't respond to you fluidly, and they couldn't understand if you said... let's say you said you wanted to get a return, but instead you said, "Where's the refund?" Maybe it misunderstood what you were saying, even though it was the same intention.
20:03
Alex Shevelenko: So the answer is not one or the other—it's a marriage of both, right? Because if you're... let's say you're using AI for customer service to extend that analogy: you definitely want that fluidity. You want someone to be able to pick up the phone, get your customer AI agent, have a conversation that makes sense, and not get them super mad, right?
20:31
Alex Shevelenko: But you also don't want that AI agent promising a refund for something that is out of policy. That happened, and there were lawsuits about that—that was another famous case that came out in the early days of AI.
20:48
Alex Shevelenko: The answer really is—and this is how Salesforce thinks about it—combining the determinism, all the sort of things that we've learned from decades of business of what these workflows need to look like, with the dynamism, the non-determinism, of cutting-edge AI models.
21:03
Alex Shevelenko: The reason people get confused is because we have two parts of our lives, right? One is where we're writing an email or trying to psychoanalyze our relationships with ChatGPT or whatever, and any insight is better than no insight, type of thing. There's not as much at stake, at least from a business perspective.
21:37
Alex Shevelenko: But then the moment you have a regulated workflow or a high-impact, financially dependent decision, you could get in trouble really, really quickly.
21:45
Alex Shevelenko: I think a lot of people are... look, we're an early innovation company, so I don't want to dismiss other companies like that, but even I see these AI companies popping up that are basically planting a flag, testing some ideas. They may not be real—they're testing things, but they look real because it looks clean and they use some agent to do this or that. Then you start getting overwhelmed with the noise. Oh my gosh, there's so much noise.
22:27
Alex Shevelenko: So therefore you need to go back and say, "Okay, this is a high-value workflow or a compliant workflow, where the revenue or cost impact of this could be dramatic to our business. We need to put some rigor into it, and we need to put people who will be around—who are not experimenting, who are like the Salesforces of the world." Or if you're picking startups, those who care about regulated industries like yours and have skin in the game.
23:04
Alex Shevelenko: It's a bet on this. Because of the noise, it almost feels like there will be a flight to quality.
23:17
Paula Goldman: I think that's true. I think that's true. I feel like there's a lot of hype and fear of missing out. To be clear, I actually think experimentation is extremely important, right? So if you go in and say, "I have to have this all locked down and totally figured out before I get started," you're going to miss the boat, right?
23:39
Paula Goldman: That said, there are particular areas of work where you want to make sure you're experimenting within a safe sandbox.
23:47
Alex Shevelenko: Experimental accountants—that's how you get in prison!
23:51
Paula Goldman: Well, here... no! Here, for example, we have an incredible customer named 1-800Accountant, and the business basically does financial advice for small businesses. Every year around tax time, you can imagine they're totally inundated.
24:10
Alex Shevelenko: Yes, yes.
24:12
Paula Goldman: Most of those inquiries are like, "Has my tax return been filed? What's its status?" So it's a perfect question for AI, so long as it's within a reasonable number of guardrails. You probably want that person to be authenticated; you don't want to just be giving away someone else's personal information.
24:31
Paula Goldman: But that's fairly trivial to use AI for right now, and transformative. But the minute someone asks, "Should I restructure my business on account of tax strategy?"—you don't want AI answering that question, for a number of reasons.
24:50
Paula Goldman: One, this is a very regulated category—financial advice is a regulated category. Second, when someone's asking a question like that, not only do you want to make sure they get a quality answer, but that is a great opportunity to deepen the business relationship, right?
25:06
Paula Goldman: So another piece of really important human-AI collaboration principles is: when is the human-AI handoff? What is the right time? Especially here, we're talking about when AI escalates to a person. Setting those rules in advance and keeping them in mind to understand as they evolve, right? Because the capabilities will change and the landscape will change. But I think those types of things will become standard questions when we're implementing AI.
25:35
Alex Shevelenko: Yeah. I think it's interesting that you brought up frequently asked questions. We have AI generating questions when you load something in—it reads it and generates its questions. Then we know that for some industries, those questions may be great from AI per se, but they're probabilistic. We actually know that there is a deterministic set of questions that are the right frequently asked questions.
26:08
Alex Shevelenko: And so we had to add that. It sounds basic, but we saw ahead of it, and the customer side was asking, "Well, what if I want to combine the FAQs together with the AI, right?" Because we all know the FAQ concept.
26:26
Alex Shevelenko: So you're giving, on the one hand, an organic ability to ask questions, right? But on the other hand, you're providing a guided experience based on what you need to accomplish and what you expect the clients need to accomplish.
26:40
Alex Shevelenko: And you could change those FAQs over time based on the data, because now you're actually smarter about what people really think about, right?
26:49
Paula Goldman: Yeah.
26:49
Alex Shevelenko: And that is a great... so the accounting example is great because we could almost think of a number of industries where you could take out the routine from people.
27:05
Alex Shevelenko: One of our customers that I was telling you about reminds me of tax folks around the submission deadline—it's employee benefit advisors and brokers who support HR people with employee benefits decisions in the US, which happens around the November timeframe all at the same time.
27:31
Alex Shevelenko: Oh my god, how much freedom and ability to do higher-value things it gives to people's lives if they could outsource some of the things that are not amazingly value-add.
27:49
Paula Goldman: Yeah, focus on the human component.
27:51
Alex Shevelenko: That's right. Because everybody's worried and talking about, "My job is going to be gone, da da da." But we're finding that you just need to be more human and lean into your strengths—the things only you can do. Those could be the really fun parts of the job.
28:11
Paula Goldman: I think that's absolutely right. What we are going to find as AI gets more and more powerful is that these human skills—relationship building, creativity, judgment, situational judgment... I actually think the scarce resource may not be AI. I think it might be human judgment and how we design these AI systems to make the best of it.
28:36
Paula Goldman: I deliberately called the book Manage the Machine, because I think that's what it is, right? We're used to thinking about management as managing people. It turns out that—obviously AI is not a person and shouldn't be treated like a person—there's a lot of similarity in terms of learning how to delegate.
29:01
Paula Goldman: Learning not only how to give a task in a way that AI understands how to do it well and can figure it out, right? So some wiggle room—there's no point in telling it exactly every single thing, because then why bother? But you also need to give the right context and the right guardrails, and then see how it does and keep expanding it.
29:21
Paula Goldman: I heard over and over from people in different industries—lawyers, people at tech startups—the same thing: you're experimenting in a similar way as you would delegate to a person on your team.
29:40
Paula Goldman: The skill of how we manage—and that's just one part of it—is going to be really important because all of a sudden we have this collaborator. It's not an inert piece of software, it's not a person, but it's doing work and we need to figure out how to manage it.
30:00
Alex Shevelenko: What's your take on the skill of curiosity?
30:07
Alex Shevelenko: I think I was telling you: one of the reasons I started the podcast was to remind myself that I need to shut up and listen to people smarter than me, learn to ask questions, and follow the thread without overprogramming it—to kind of follow the conversation. I see that sometimes AI does a good job in this area, and sometimes...
30:36
Paula Goldman: Yeah, well, there's a danger. There's a danger.
30:40
Paula Goldman: I'm really glad you asked this question because that was another one of the big pieces of how we manage AI: we need to be setting it up in a way that preserves that judgment, that taste-making, that curiosity.
30:57
Paula Goldman: There is a risk—it's well known and has been studied forever—that people, especially outside of their domain of expertise, tend to trust AI output way too much, right?
31:09
Paula Goldman: So there is a risk. It's the same risk that people are talking about now, calling it cognitive offloading, right? Where it's like, "I need to understand this topic" or "I need to write this thing, so I'm gonna ask AI to do it."
31:25
Paula Goldman: There's nothing wrong with that on its face, but there are situations where you really want to be able to preserve the mental effort and energy that goes into clarifying your own thoughts or learning, and not just offloading.
31:44
Paula Goldman: To give you an example: I have a whole chapter about product innovation in the book. There was this story from IKEA where their innovation team wanted to break the boxy prototype of a couch, right? They sell a lot of couches—what could the couch of the future look like?
32:05
Paula Goldman: So they brought in AI, and it kept defaulting to the meme; it kept bringing up the same sort of boxy prototypes. They had to force themselves to think before they brought in AI again.
32:19
Paula Goldman: They started thinking about other metaphors, like camping or gathering or whatnot, and ended up with this really cool foldable couch that you could carry—it was just a few pounds, and it ended up in a museum exhibit.
32:37
Paula Goldman: The point of that story is that there are techniques to preserve the human side of thinking, creativity, and discernment. Sometimes that's before the AI runs; sometimes it's after the AI runs where you've got a really critical decision.
32:55
Paula Goldman: Like, let's say that same financial advisor that we talked about hypothetically, right? You might imagine that they're using AI to brainstorm answers to a query about financial advice—that's something people do.
33:11
Paula Goldman: But you need that person to be accountable for that advice; they need to own it. So there's a little bit of positive friction. Sometimes people set it up so that they ask people to actually write up their own rationale before they're submitting something.
33:27
Paula Goldman: There are lots of different techniques, but again, on both sides, this notion is that it's not just about getting the best from AI—it's about getting the best from people as well and finding the systems that bring them both together.
33:41
Alex Shevelenko: Ironically, this reminds me a little bit of a Mike Tyson quote: "Everybody has a plan until they get hit in the face." I think AI is that plan—it's the sort of analytical system that makes it easy to avoid decisions or avoid things. And then you need to actually meet humans with rigor, and then you get, "Oh wait, that doesn't work!" Then you can maybe go back to AI and refine it, right?
34:19
Paula Goldman: That's totally right. But good job! We brought in my mom, we brought in Mike Tyson—that was unexpected. Good job.
34:26
Alex Shevelenko: We've got an anthropology degree! So, I think everybody is interested in figuring out how to manage the machine. Where can they find your book, when is it coming out, and what's the big nugget that you want to encourage people to do differently?
34:48
Paula Goldman: Yeah, well, thank you for asking. It is available for pre-order now, and it comes out September 15th wherever books are sold—Amazon, but also your local bookstore.
34:58
Paula Goldman: And the nugget really is this: the companies that are getting the best results from AI are obviously managing the technical side of it, but they're also managing the human side of it. They're seeing it as a human management challenge.
35:13
Paula Goldman: So all of these things we talked about in terms of building the skills, retaining judgment, creativity, curiosity—all of that... honestly, I had to write the book to get through tons of case studies and understand how deep this was, but it's not just a nice-to-have. It's actually how you get the right results from AI—by building the right human-AI collaboration.
35:37
Paula Goldman: So that's the takeaway. It's full of really practical tips, org by org: marketing, sales, service, and so on. Honestly, I wrote the book because I wanted to be in conversation with people about this and see it take more institutional form. So I'm really excited to hear from people, hear their thoughts, see what they're doing, and learn what their best practices look like.
36:04
Alex Shevelenko: Can people find you on LinkedIn? Where's the best place?
36:07
Paula Goldman: Yeah, LinkedIn is the best place.
36:10
Alex Shevelenko: So I'll summarize it as I see it: to get better at AI, you need to get amazing at being human.
36:18
Paula Goldman: Yes! Yes.
36:21
Alex Shevelenko: Thank you so much for coming, writing the book, sharing your nuggets, and being a great human, as everybody can see here.
36:28
Paula Goldman: Oh, thank you so much. This was really fun!