00:00Can the machine be managed?
00:02I argue yes, absolutely.
00:04And what's interesting is obviously there's no shortage of commentary about AI.
00:08It is moving very, very fast.
00:10But what's interesting, having worked across now thousands of companies, large and small,
00:14is that the companies that are getting the best results from AI,
00:17they're not just thinking about the power of the model.
00:20They're thinking about how they design how people on AI work together to get the best outcomes.
00:25And that's really what the book is about, is how do you design that
00:28so that people can do bigger, better things and get trustworthy results?
00:32Give me a sense, though, because we talk about this existential threat of AI.
00:36I feel like for a while the conversation was always, oh, it's going to take my job, take this.
00:40But ultimately, the existential sort of question is about this idea of what does this make humans?
00:45If you believe in this idea that intelligence becomes a commodity,
00:48because we can all just talk into our chatbots and get it,
00:51where does the human fit into all of this process longer term?
00:55It's interesting because AI is really powerful.
00:59But as we all know, it has failure modes.
01:01We've read about them in the news.
01:03You've talked about them, like the hallucinations,
01:06the lawyers that are getting cited for court cases that are made up.
01:11And these failure modes continue.
01:13So that's part of the reason why we need guardrails in our products
01:17that allow businesses to get the right results.
01:19But it's not all of it.
01:20Like, I talk to salespeople who are using AI.
01:24They're super happy to not have to hear no on cold calls anymore
01:28and to be able to concentrate on building trust with their customers
01:32that helps them reinterpret their problems.
01:35Those relationships, that judgment, the creativity,
01:38that is what AI used well helps unlock for people.
01:43Well, let's talk about the judgment aspect
01:44and exactly what does human judgment look like
01:48in a world of, I guess, AI intelligence?
01:51Well, I'll tell you.
01:53You know, one of the most robust use cases for AI right now
01:58is in customer service, right?
02:00But one of the most important questions
02:02that companies using AI for customer service need to ask is
02:05when to stop using AI, right?
02:08That is to say, we have a customer 1-800-accountant.
02:12They do financial advice for small businesses.
02:14And, you know, if someone calls in and says,
02:17is my tax return filed?
02:18Great use case for an AI agent.
02:20Someone starts asking for tax advice or financial advice.
02:24Not only is that a regulated use case that requires a person,
02:26but also it's a chance to deepen the relationship
02:30to really give solid guidance
02:33that then makes a better customer relationship.
02:36Well, give me, I mean, in the book,
02:36you give a couple of examples of Walmart, Nike, Shark Ninja,
02:40a couple of them I'm probably forgetting.
02:41But is the idea that this is going to be about
02:44what the customer really wants
02:46or the client wants from their experience,
02:48whether it's a simple call to a customer center
02:51to, you know, whatever, you know,
02:52check the credit card this or that
02:54or something even more complicated like getting advice.
02:56How much do I, as the client, the customer,
03:00how much control do I actually have in this process?
03:02Do you mean if you're the one that's on the receiving?
03:04Yeah, on the receiving, yeah.
03:05Well, I mean, I think that's also something
03:07that needs to be factored in.
03:08Some people may have a preference to talk to a person.
03:10Some people may, you know, it was interesting.
03:12Studies say that when people are angry,
03:14they generally want to talk to a person.
03:15When they're embarrassed, actually,
03:17they sometimes want to talk to AI.
03:18And these are things that we're learning.
03:20And I think that's an important part of the process.
03:22But equally, people who are working and using AI
03:26have a lot of strong instincts
03:28on what it can unlock for them.
03:30And that's, I think, what the exciting part is
03:32about the moment that we're in.
03:33So there was this concept in the book
03:35you called human in the loop.
03:38Human at the helm, actually.
03:39You're not, but human in the loop.
03:41Sorry, I'm going, I'm getting there.
03:43You're not a fan of that phrase.
03:44Well, so human in the loop.
03:46Yeah.
03:47Comes from the Cold War.
03:48Yeah.
03:48And it, you know, was like all of a sudden
03:50we had technology that could detect an inbound missile.
03:53Mm-hmm.
03:55Who should decide what to do with that information?
03:58How do you make sure that a person
03:59makes a consequential decision?
04:01Kind of got flattened.
04:03Like we start, when we think human in the loop,
04:04it's like, well, the AI drafts something
04:06and then I'm going to approve the email and send it.
04:09But in a world of AI agents, like that kind of defeats the point.
04:12Like you can't have a person looking at every data point
04:15that the AI agent is working on.
04:17I mean, this was sort of the core of the hugging face issue
04:20and some of the other sort of, you know, gone rogue, if you will.
04:23It's the idea that a human is supposed to sort of be in the loop.
04:26They either weren't there
04:27or there was just too much information for them to even keep up.
04:29Well, what we say is that we need to design systems
04:32that we call, we call the principle human at the helm,
04:34where you can delegate really complex tasks to AI,
04:38but when there's an anomaly,
04:40when there's something that requires judgment,
04:41a new circumstance,
04:43a relationship that needs to be strengthened,
04:45that people are deliberately brought in
04:47at the right moments to exercise judgment
04:49and honestly to take accountability for those outcomes.
04:52So that's, I think, the imperative.
04:54That's the moment that we're at right now with tech.
04:56And, you know, when I read that line,
04:57I was like, you know,
04:58I completely thought of Dr. Strangelove.
05:00It gets to this idea, though, of trust.
05:02And, you know, when you hear a warning
05:04from Dario Amadei and some of these others,
05:06there's a backlash out there,
05:08not to what they're saying, but as to who's saying it.
05:11And I do wonder if some of these tech CEOs
05:14have sort of lost the trust of the public
05:17with regards, specifically with regards to this AI issue.
05:19Well, it's interesting that you use the word trust, right?
05:22Right. Salesforce, our top value is trust.
05:25Why? Because we knew that when we were founded years ago,
05:28we asked people, now it sounds clean,
05:30but we asked people to put their data in the cloud
05:32and they were worried about it,
05:34that we had to give strong mechanisms
05:36to enable them to trust us.
05:38And that's what we're finding right now
05:39with AI and the enterprise.
05:41It's a very different conversation
05:42than this broader conversation about AI.
05:45Companies that are using AI, they need it to work.
05:47They need it to work as designed.
05:50They need to know that it's going to escalate
05:52if there's an anomaly.
05:53They need to know that it's not going to talk about
05:55something that they've already said is off topic.
05:57Trust is central to adoption.
05:59And I think businesses have a real role to play
06:02in scaling that.
Comments