Latest / Humanity Working / How To Get Serious About AI

Transcript

How To Get Serious About AI · Humanity Working

• 0:05
  SPEAKER_03: And you can take a step back and you have the time to think about how you can help grow the business.
• 0:10
  SPEAKER_03: And good things are going to happen.
• 0:12
  SPEAKER_03: And I personally have a very optimistic view of the future.
• 0:15
  SPEAKER_03: I mean, I know that I'm in the business, so it's kind of in my nature to be optimistic, but I really do think we're going to see productivity gains like we've never seen before, not just in tech, but in all sorts of uh businesses, nonprofit, society, community service.
• 0:29
  SPEAKER_03: I think we're going to start to see some real productivity gains.
• 0:36
  SPEAKER_01: Hi there.
• 0:36
  SPEAKER_01: Welcome to Humanity Working.
• 0:38
  SPEAKER_01: Now, today's episode is a really interesting one.
• 0:40
  SPEAKER_01: We've got Boaz Ashkenazi, who is the CEO and co-founder of Simply Augmented, and he's also the host of the Shift AI podcast, which is absolutely fascinating.
• 0:50
  SPEAKER_01: If you've got any interest in the world of AI, strongly recommend you listen to it.
• 0:54
  SPEAKER_01: And that, of course, is the topic that Boaz and I are going to be talking about today, AI.
• 0:58
  SPEAKER_01: We're going to be looking at it from a couple of different angles, but mainly what AI is doing in the world of work today and then what we think it will do over the course of the next few years to disrupt and change how we do work.
• 1:10
  SPEAKER_01: But before we get to that, I've got a really, really simple request for you.
• 1:14
  SPEAKER_01: Now I know everybody's always asking you to like and subscribe, like and subscribe, but it actually really is important.
• 1:20
  SPEAKER_01: If you can, if you're watching us on YouTube or if you're listening to us on your favorite podcast, do take a moment to rate the podcast and let other people know why you like it.
• 1:29
  SPEAKER_01: It really does make a difference.
• 1:31
  SPEAKER_01: It will help other people find us, and it will help us also do a great job of recruiting really good guests for you.
• 1:36
  SPEAKER_01: And on that topic, if you have any ideas as to additional guests you'd like us to feature, please email us at info at humanityworking.net.
• 1:44
  SPEAKER_01: That is info at humanityworking.net, and we'll be sure to do our best to get them on as a guest.
• 1:50
  SPEAKER_01: All right.
• 1:50
  SPEAKER_01: Enjoy the show.
• 1:54
  SPEAKER_01: I kind of want to start with a little bit of a of a future view.
• 1:57
  SPEAKER_01: Everybody has a um, I think a probably a good understanding of what, for example, generative AI, uh, things like chat GPT, etc., mean for them today.
• 2:07
  SPEAKER_01: But I'd love us to fast forward, let's say, five to six years.
• 2:10
  SPEAKER_01: What are we, 2024 today?
• 2:11
  SPEAKER_01: So let's let's imagine a world in in 2030.
• 2:15
  SPEAKER_01: What do you think the working experience is likely to be for humans in in say five or six years, in terms of how humans are interfacing with AI, what is the role they're playing?
• 2:28
  SPEAKER_01: What is the role that AI is playing in that future world?
• 2:32
  SPEAKER_03: Yeah, it's a really interesting question.
• 2:34
  SPEAKER_03: And I think that we're starting to see a little bit of it now, but not so much in the chat GPT sense where you're prompting a chat, but I think it's gonna be a lot more invisible.
• 2:45
  SPEAKER_03: I don't think we're gonna have conferences on AI in five years.
• 2:48
  SPEAKER_03: I don't think we're gonna have podcasts about AI.
• 2:50
  SPEAKER_03: I just think it's gonna be everywhere, and I think it's gonna be part of all the software that we use.
• 2:56
  SPEAKER_03: And so I think from a human standpoint at work, you're gonna be using software tools that you may use today.
• 3:03
  SPEAKER_03: You know, you might be logging into Salesforce, you might be logging into Microsoft Tools or Notion or some other tools, and AI is gonna be embedded in everything that you do.
• 3:14
  SPEAKER_03: I think also that we're gonna find that some of the monotonous things that we did previously, we're not gonna have to do anymore.
• 3:23
  SPEAKER_03: It doesn't mean that we're not gonna have to work.
• 3:25
  SPEAKER_03: It's just gonna be doing different kinds of work.
• 3:28
  SPEAKER_03: And I'll give you an example of that.
• 3:29
  SPEAKER_03: A salesperson, for example, who spent a lot of time taking unstructured data and putting it into structured data is not gonna have to do that anymore.
• 3:38
  SPEAKER_03: What do I mean by unstructured data?
• 3:40
  SPEAKER_03: Think about all the times previous to a lot of these note-taking tools that we now have where you had to take down notes about a conversation, take those notes and put those into a form, put that form and you know, put that into Salesforce.
• 3:53
  SPEAKER_03: And there was this method of kind of getting all this unstructured information into a digital backbone that was a big part of your job.
• 4:02
  SPEAKER_03: And a lot of the sales guys I've talked to, that was the part of the job they hated the most.
• 4:06
  SPEAKER_03: They wanted to be on phone calls with clients, not being a data entry person.
• 4:11
  SPEAKER_03: And I think that there's going to be a lot of those kinds of things that we just don't have to do anymore.
• 4:18
  SPEAKER_01: Yeah.
• 4:18
  SPEAKER_01: It's intriguing because the, you know, the this six-year time horizon is um is one that I like to play around with quite a bit because we know as as human beings, right, we tend to underestimate in general.
• 4:31
  SPEAKER_01: We tend to kind of like underestimate the um the amount of change that will happen in, say, let's say a 10-year time horizon or a 20-year time horizon.
• 4:39
  SPEAKER_01: But we tend to overestimate the amount of change that will happen to our lives in the next year or so or something or something along those lines.
• 4:46
  SPEAKER_01: So this is sort of like comfort comfortably, for me, comfortably sat in the middle.
• 4:51
  SPEAKER_01: Um I guess to sort of like to probe a little bit a little bit deeper on that, it sounds like a it sounds like a interesting future, uh a future where work is potentially more rewarding.
• 5:04
  SPEAKER_01: If I were to force you to be, I'd say a little bit more dystopian, right?
• 5:09
  SPEAKER_01: Because none of this is certain, right?
• 5:11
  SPEAKER_01: What would be the I said, give me dystopian Boaz.
• 5:15
  SPEAKER_01: What would the what would the dystopian uh six-year time horizon be for you?
• 5:20
  SPEAKER_03: Yeah, I think that there are gonna be a lot of people in our society that are gonna be out of work and needing some sort of universal income.
• 5:30
  SPEAKER_03: I don't know if it's gonna be in five to six years, but let's imagine dystopian Boaz was saying in six years there was gonna be uh that kind of scenario.
• 5:38
  SPEAKER_03: I think we have to have a hard question about what that looks like.
• 5:42
  SPEAKER_03: And I know that in the United States we don't like to talk about that, but I think it's gonna be something that we're gonna have to talk about and society is gonna have to deal with it.
• 5:53
  SPEAKER_03: Because I think that there's gonna be a certain stratus of jobs that are just not necessary anymore, and there's gonna be some people that don't want to uh learn some of these new tools and are not gonna be interested in kind of operating in some of these uh other uh areas.
• 6:10
  SPEAKER_03: So I I think that in for stability in a dystopian world, there's gonna have to be universal income.
• 6:16
  SPEAKER_03: I really believe that eventually.
• 6:19
  SPEAKER_01: Yeah.
• 6:20
  SPEAKER_01: Um this is I think one of the more intriguing aspects as to the nature of this revolution, right?
• 6:29
  SPEAKER_01: If we take the first piece of the answer that you gave.
• 6:34
  SPEAKER_01: I loved it because you said that you know, we're not gonna have conferences about AI.
• 6:38
  SPEAKER_01: It's just gonna be part of it's just gonna be part of what we do.
• 6:41
  SPEAKER_01: And and history would suggest that, right?
• 6:43
  SPEAKER_01: I mean, we're no longer having um like detailed conferences about, I don't know, lower levels of the networking stack or you know, exactly and things like that, right?
• 6:53
  SPEAKER_01: It just becomes part of our uh part of our everyday lives.
• 6:57
  SPEAKER_01: But taken on face value, that can mean, oh well, it's just not important anymore.
• 7:02
  SPEAKER_01: No, it what it really means is that it's it's just become so central, so central to to who we are, so central to our work environment, so central to our lives that it's that and so ubiquitous that it's not the new disruptive thing that is we've sort of taken on board the disruption.
• 7:22
  SPEAKER_03: Yeah, but it's impossible to live without but impossible to live without impossible to live without, exactly.
• 7:30
  SPEAKER_01: Um but that the very fact that that is happening and that is happening at a um at the pace at which it it is is happening, um can lead us, I think, potentially into situations where um we're not paying enough attention to make sure that this is happening in a way that uh allows us to maintain society in the in the way that we hope.
• 7:52
  SPEAKER_01: Um do you think that there are lessons to learn from previous technology revolutions that will um and the way we've handled them either positively or negatively, um, that will allow us to be able to handle this one potentially better?
• 8:06
  SPEAKER_01: I'm thinking of things like social media, for example, or other examples that you might want to think of.
• 8:10
  SPEAKER_03: I might go back even farther.
• 8:12
  SPEAKER_03: I think there's two parts to it.
• 8:13
  SPEAKER_03: You know, there's the metaphor that's used a lot to describe AI, and that's electricity, uh, you know, the advent of the automobile.
• 8:21
  SPEAKER_03: And one of the things that's really interesting about that time when you go back and look at the news coverage is how much they talked about electrocution and the and the dangers around electricity.
• 8:32
  SPEAKER_03: And now, I mean, we don't really think about, I mean, yes, you can still get electrocuted, but we figured out ways of kind of really mitigating the risks associated with electricity, and it's never going away.
• 8:45
  SPEAKER_03: The same thing is true with the automobile.
• 8:46
  SPEAKER_03: There were real fears, and there were articles about the dangers of driving.
• 8:51
  SPEAKER_03: And yes, people still get in accidents, but it's just really mitigated now.
• 8:55
  SPEAKER_03: And I kind of feel like the monster AI robot that's gonna take over humanity that that narrative is misplaced.
• 9:03
  SPEAKER_03: I could be wrong, but I think it's misplaced.
• 9:05
  SPEAKER_03: And I think in the future, there's still gonna be that um risk, of course, with AI and in certain kind of security and privacy issues.
• 9:13
  SPEAKER_03: But I just think that as a society, we're gonna figure out ways to deal with it.
• 9:18
  SPEAKER_03: Not perfect, but figure out ways to deal with it in the same way we have with electrocution.
• 9:24
  SPEAKER_01: Can we dive into that metaphor a little deeper?
• 9:28
  SPEAKER_01: I'm sure I'm really struck by it because as you were describing that to me, I I was along I was on the ride with you, and at the same time I was thinking, I think there's something else here too, though.
• 9:40
  SPEAKER_01: Because maybe a lesson from the automobile is not that we shouldn't have been worried about that, but maybe that we didn't have the foresight to be worried about in quotes the right thing, right?
• 9:56
  SPEAKER_01: Because you think about it today, right?
• 9:59
  SPEAKER_01: Automobiles are massive contributors to climate change, and nobody in 1900, because because with the concept didn't even exist, right, nobody in 1900 was was thinking that through, right?
• 10:12
  SPEAKER_01: So the it's possible that with regard to this, that we're worried about the crazy AI robot that's gonna take us over, but we're not necessarily worrying about downstream effects that that might impact us in a very severe way.
• 10:30
  SPEAKER_01: That like the so as in the similarly with the social media example, right?
• 10:34
  SPEAKER_01: When social media launches, people worried about a whole bunch of different things.
• 10:37
  SPEAKER_01: I'm not sure that many of them were worried about things like teenage suicides.
• 10:42
  SPEAKER_03: Yeah, that's a really interesting point.
• 10:44
  SPEAKER_03: I hadn't thought about that.
• 10:45
  SPEAKER_03: Yeah, but I think you're right.
• 10:47
  SPEAKER_03: And I'm not sure what those things are gonna be, but we're already starting to see little things around the way that people, how comfortable people feel writing, how creative people are when they have these agents, assistants, helping them along the way.
• 11:03
  SPEAKER_03: And I find that fascinating too.
• 11:04
  SPEAKER_03: I just wonder how that's gonna evolve.
• 11:06
  SPEAKER_03: You know, how are people how comfortable are people gonna be with a blank piece of paper in front of them in five years?
• 11:12
  SPEAKER_03: Where you know they really feel like they need something to help them get started.
• 11:15
  SPEAKER_03: And maybe that's okay if we can figure out how to manage that.
• 11:19
  SPEAKER_03: But I think there is gonna be something lost.
• 11:21
  SPEAKER_03: I'm just I'm not sure exactly what the all of the downstream effects are gonna be.
• 11:26
  SPEAKER_03: But I think it's a really interesting point.
• 11:29
  SPEAKER_01: And I suppose the answer to that is that we will figure that out if we if we're robust at studying early not just the AI itself, but the human AI interface, if you like.
• 11:45
  SPEAKER_01: The what the changes in human behavior that are happening as AI becomes more part of our day-to-day lived experience.
• 11:58
  SPEAKER_03: Yeah, I I think you're right.
• 11:59
  SPEAKER_03: And I I mean, we as a company are so focused on the workplace.
• 12:04
  SPEAKER_03: You know, it's kind of an extension of the work that we did at Meta.
• 12:06
  SPEAKER_03: Now we're just doing that for other companies.
• 12:08
  SPEAKER_03: And so we're starting to see like what happens, not so much in a chat-related interface, but just when you go to work and the job that you were doing now has AI as kind of uh supercharging that experience, what that does to you and how you uh how you go to work, how you show up for work.
• 12:25
  SPEAKER_03: And that's really interesting.
• 12:25
  SPEAKER_03: And I think we're gonna see more of that.
• 12:28
  SPEAKER_03: And in some ways, it's not gonna be so dystopian, in some ways it could be very positive.
• 12:32
  SPEAKER_03: You know, um, more and more people will have an opportunity to be more strategic and work on the business, not in the business.
• 12:38
  SPEAKER_03: You know, you hear this a lot with owners of a business getting dragged into the minutiae of the business and it's preventing them from growing.
• 12:45
  SPEAKER_03: And I think the same thing is true for other people in the office, you know, what that does to you and how you uh how you go to work, how you show up for work.
• 12:53
  SPEAKER_03: And that's really interesting.
• 12:54
  SPEAKER_03: And I think we're gonna see more of that.
• 12:56
  SPEAKER_03: And in some ways, it's not gonna be so dystopian, in some ways it could be very positive.
• 13:00
  SPEAKER_03: You know, um, more and more people will have an opportunity to be more strategic and work on the business, not in the business.
• 13:07
  SPEAKER_03: You know, you hear this a lot with owners of a business getting dragged into the minutiae of the business and it's preventing them from growing.
• 13:14
  SPEAKER_03: And I think the same thing is true for other people in the office.
• 13:18
  SPEAKER_03: You know, when you can take a step back and you have the time to think about how you can help grow the business, good things are gonna happen.
• 13:25
  SPEAKER_03: And I personally have a very optimistic view of the future.
• 13:28
  SPEAKER_03: I mean, I know that I'm in the business, so it's kind of in my nature to be optimistic, but I really do think we're gonna see productivity gains like we've never seen before, not just in tech, but in all sorts of uh businesses, nonprofit, society, community service.
• 13:42
  SPEAKER_03: I think we're gonna start to see some real productivity gains.
• 13:46
  SPEAKER_01: So that leads us um rather neatly into kind of discussion of your organization and and how it came about.
• 13:52
  SPEAKER_01: So take us back to uh to when you first started, simply augmented what was what was the idea and how has it evolved over time?
• 14:02
  SPEAKER_03: Yeah, I mean, I need to step back one more step because um when I was at Meta, we were working uh in a company that had sent everybody home.
• 14:12
  SPEAKER_03: And you know, when I first started, everybody was at home.
• 14:15
  SPEAKER_03: And then all of a sudden people started trickling back.
• 14:18
  SPEAKER_03: And uh, but when what we were building were productivity tools for all sorts of folks in the organization, not just engineers.
• 14:25
  SPEAKER_03: So we were building uh chat interfaces and other kinds of chat bots, but we were also building VR uh headset experiences and AR experiences for people in the data centers, facilities, security, uh, you know, all sorts of folks in that organization.
• 14:40
  SPEAKER_03: It's a big organization.
• 14:42
  SPEAKER_03: And when ChatGPT came out, one of the engineers on my team, a young engineer, very, very uh smart and savvy in machine learning and AI, approached me and said, you know, we really need to start a company and, you know, we need to do this.
• 14:56
  SPEAKER_03: This is going to be big.
• 14:57
  SPEAKER_03: And this was kind of early, you know, right when ChatGPT came out.
• 15:01
  SPEAKER_03: And it I really um it was because of him that I got much more interested early on.
• 15:08
  SPEAKER_03: And yeah, it made a lot of sense to take what we had learned about how to make employees more productive and just take that outside uh into the world.
• 15:15
  SPEAKER_03: It just so happens that because of our size, you know, we're a small agency, we tend to kind of our our customer profile lives in this medium-sized business zone.
• 15:26
  SPEAKER_03: You know, we're we're trying to help businesses with, you know, 100 to 300 employees, try to figure out where Gen AI can fit into their workflows.
• 15:34
  SPEAKER_03: So that's kind of our obsession, is just understanding.
• 15:37
  SPEAKER_03: A lot of people, they understand ChatGPT and they're learning about how to prompt engineer, but they're not really sure exactly what the power is of large language models in their existing workflows.
• 15:48
  unknown: Yeah.
• 15:49
  SPEAKER_03: And so that's what we're really focused on.
• 15:51
  SPEAKER_01: Okay.
• 15:51
  SPEAKER_01: So for our audience that, you know, they'll they'll they'll have a, I think, a general understanding of what generative AI is, and they may have a some understanding of what a large language model is, but probably most of their familiarity with it will be using something like ChatGPT or Gemini or something along those lines.
• 16:08
  SPEAKER_01: So I wonder if we could just if you just give us a little bit of a primer about large language models specifically and what what this means if you're not using, for example, out-of-the-box GPT.
• 16:21
  SPEAKER_03: Yeah.
• 16:22
  SPEAKER_03: I think out-of-the-box GPT is a good reference point because I think people understand that.
• 16:27
  SPEAKER_03: But there's a lot of misconceptions about generative AI.
• 16:29
  SPEAKER_03: And one of the misconceptions is it understands facts.
• 16:33
  SPEAKER_03: And it it's trained on facts so that it could be a very good language processing tool, but it doesn't really understand facts in the way that we do, which is why you see a lot of businesses point those models at internal proprietary data.
• 16:48
  SPEAKER_03: If you give it data and if you give it unstructured data, especially, and I'll explain that in a second, it does really, really, really well understanding what that is.
• 16:57
  SPEAKER_03: So I'll give you a perfect example of some of the ways that we're doing this, and it'll give you a kind of a sense of the way, the power of large language models.
• 17:04
  SPEAKER_03: So we have a customer that is uh in the retirement account space, the self-directed retirement accounts, and they get a lot of inbound content from their customers.
• 17:13
  SPEAKER_03: And it's all over the place, different formats, some are PDFs, some are unstructured data that lives in uh documents and transcripts.
• 17:21
  SPEAKER_03: So it's a very dispersed data.
• 17:23
  SPEAKER_03: Inside their company, they have very, very structured digital backbone.
• 17:27
  SPEAKER_03: And there's an ERP system, and there's all sorts of structure.
• 17:30
  SPEAKER_03: So when you think about like previous to generative AI, how would you get all that unstructured data into a structured system?
• 17:37
  SPEAKER_03: You'd either make your employees do it, you would have them fill out forms and get it into a way that it would be able to be put into a structured table, or you could have your customers do it.
• 17:47
  SPEAKER_03: And you see this a lot on websites, right?
• 17:49
  SPEAKER_03: You send them a form and you make them do it.
• 17:51
  SPEAKER_03: The customer experience is not a bad one.
• 17:54
  SPEAKER_03: I don't want to fill out another form.
• 17:56
  SPEAKER_03: And the employee experience slows you down.
• 17:58
  SPEAKER_03: But now what we have is we have these tools we just didn't have before.
• 18:02
  SPEAKER_03: We have tools that can read those documents, no matter how dispersed they are, make sense of it at a very, very high accuracy and put it into a structured database in an amazing way.
• 18:12
  SPEAKER_03: It's uncanny.
• 18:13
  SPEAKER_03: And the same thing uh is true for another client that we have that's doing uh the same thing with unstructured data.
• 18:21
  SPEAKER_03: This can be done within uh conversations like we're having that turn into transcripts.
• 18:25
  SPEAKER_03: Now all of a sudden you have a bunch of data that can be brought into uh to into a structured form.
• 18:30
  SPEAKER_03: And so I think that's where the real power is, is being able to make sense uh of all of this stuff in a way that we just couldn't before.
• 18:37
  SPEAKER_01: Yeah.
• 18:38
  SPEAKER_01: It's funny when through our work, when we work with people on uh on AI readiness and we work with organizations around um strategically adopting particularly generative AI type models.
• 18:53
  SPEAKER_01: Um our focus is really on um on how to make the the humans um ready to bring AI into their team, as it were, and what this means for organizational dynamics.
• 19:07
  SPEAKER_01: So for example, one of the uh one of the analogies that we use is that you're almost taking old school managers that were supervisors, you're almost making every individual contributor in your organization a manager because they're now coordinating a new team member, which is AI, right?
• 19:23
  SPEAKER_01: And that every team now has at least one new member, AI, in it.
• 19:29
  SPEAKER_01: And so this does a lot for like team dynamics and organizational dynamics, how the company should be structured, how it should be put together, and all this kind of stuff, all the human elements associated with it.
• 19:39
  SPEAKER_01: Um but that um is sort of tied, I would say, with a kind of general understanding from people about what these tools mean mean to them today, um as almost like a supercharger for their capabilities.
• 19:55
  SPEAKER_01: Zach Ratner, who was recently on our podcast talking about the stuff, said, you know, making you a little bit better at most of the things that you do or making you a little bit faster at most of the things that you do.
• 20:04
  SPEAKER_01: Right.
• 20:05
  SPEAKER_01: But I what I love about your approach as a supplement to that is what you're what you're talking about is almost like going through here are the here are the things that the business does, right?
• 20:18
  SPEAKER_01: Almost using almost like a lean manufacturing approach or something like that.
• 20:22
  SPEAKER_01: Say here's how everything that the business does breaks down into you know into these different processes.
• 20:27
  SPEAKER_01: And then here's how we can use these tools as tools to optimize those processes more effectively.
• 20:38
  SPEAKER_01: And I think we need to think inside organizations, we need to do both.
• 20:41
  SPEAKER_01: We need to think about what this means from a from a process optimization perspective and what it means from a human organization perspective, if organizations really are gonna realize the the true potential of this type of technology.
• 20:57
  SPEAKER_01: Would you agree with that?
• 20:58
  SPEAKER_03: Yeah, I agree with you.
• 21:00
  SPEAKER_03: And I I've got another example that I think does both in a kind of interesting way, because I think the the superpowers for humans is important, especially if it's targeted on things that are very high value for the business.
• 21:11
  SPEAKER_03: So let me give you an example.
• 21:12
  SPEAKER_03: Uh commercial real estate company that we work with, a global one, they're called Savils, and they have uh brokers in their company.
• 21:19
  SPEAKER_03: And the brokers are really the revenue generators in those cut in that company.
• 21:23
  SPEAKER_03: And one of the things that they used to be stuck on was that they would try to go into very large data sets to try to get information that was critical for them to sell.
• 21:34
  SPEAKER_03: And what that would look like is they would say, they would ask questions like, how many uh leases are there in the Detroit, greater Detroit area that are coming up for lease for this particular sector of industry?
• 21:47
  SPEAKER_03: You know, and they would use that to prospect and go out.
• 21:49
  SPEAKER_03: But they would have to go and talk to a data analyst to get that information.
• 21:53
  SPEAKER_03: They would they wouldn't know how to write SQL queries and they wouldn't know how to kind of suck that information and visualize it.
• 21:58
  SPEAKER_03: But now With Gen AI, they can ask a question using natural language, just English.
• 22:04
  SPEAKER_03: Yeah.
• 22:05
  SPEAKER_03: They can ask that question, you know, 50,000 leases, where are they right now in this greater area?
• 22:09
  SPEAKER_03: And in the background, a function call is creating the SQL query, going out into the large database, authenticating into that database, coming back out.
• 22:18
  SPEAKER_03: And actually, they are doing it in ChatGPT Enterprise, which is interesting.
• 22:22
  SPEAKER_03: So we built an extension on top of ChatGPT Enterprise.
• 22:24
  SPEAKER_03: So the interface for them actually looks a lot like chat, except there's a bunch of steps that are happening in the background that they used to have to do manually that now they don't.
• 22:34
  SPEAKER_03: That's affecting the top line of that business in a dramatic way.
• 22:38
  SPEAKER_03: And I think that's how businesses need to think about if they look at value, effort, and risk associated with Gen AI, that's the magic.
• 22:45
  SPEAKER_03: Because there's some things that Gen AI doesn't do good, even if it's a highly valuable thing.
• 22:50
  SPEAKER_03: There's just regular machine learning or it's conventional software sometimes.
• 22:53
  SPEAKER_03: It's better.
• 22:54
  SPEAKER_03: It's not good for everything.
• 22:55
  SPEAKER_03: And I think there's also this misconception that Gen AI is this magic thing that solves everything.
• 23:00
  SPEAKER_03: And it just doesn't.
• 23:01
  SPEAKER_03: It's really good at language processing right now.
• 23:04
  SPEAKER_00: Hey, welcome to the Mid-Roll ad.
• 23:06
  SPEAKER_00: As a podcast listener, I used to hate mid-roll ad breaks.
• 23:09
  SPEAKER_00: But these humanity working podcasts are packed with useful information.
• 23:13
  SPEAKER_00: They're dense.
• 23:14
  SPEAKER_00: And science tells us that almost nobody can remain fully focused for more than 20 to 25 minutes.
• 23:19
  SPEAKER_00: So we don't want you to miss the important stuff.
• 23:21
  SPEAKER_00: So here's a tiny little break for your brain.
• 23:24
  SPEAKER_00: And while you're taking this break, here's another fun fact for you.
• 23:27
  SPEAKER_00: If you do a one-day course on soft skills training in your organization, 70% of the information is lost in a week and 90% in a month.
• 23:35
  SPEAKER_00: That's right, only 10% of the information remains, and it's probably not the most useful stuff.
• 23:40
  SPEAKER_00: The name of the restaurant that did the catering, maybe, how bad the coffee was, whatever.
• 23:44
  SPEAKER_00: Billion Minds knows that, so they've designed their soft skills development with one aim only: to make sure that the learning translates to behaviors that make your organization thrive.
• 23:53
  SPEAKER_00: Every experience in Billion Minds is a learn-do exercise of about 10 minutes, but by the end of each one, employees have tangible results they can bring to work.
• 24:02
  SPEAKER_00: Not only that, but Billion Minds helps you as an organization optimize your infrastructure, policy, and culture to make sure that individual learning translates to your organization's thriving.
• 24:12
  SPEAKER_00: As a humanity working listener, you can get free access to our guide on how to build a culture of learning in your organization.
• 24:19
  SPEAKER_00: It explains what works and what doesn't work based on years of research and how people actually develop in organizations.
• 24:26
  SPEAKER_00: You can download it today at billionminds.com slash humanity working.
• 24:30
  SPEAKER_00: So not really an ad so much as a gift to you.
• 24:33
  SPEAKER_00: And now back to the podcast.
• 24:36
  SPEAKER_01: When we started the conversation, you talked about misconceptions.
• 24:41
  SPEAKER_01: Um and one of them was about, you know, like the idea of AI knowing facts and things like that.
• 24:47
  SPEAKER_01: And I thought that's something I want to talk to you about in a little bit more detail as well, because it's something that we've seen in our work, and I don't really know if we have an answer yet.
• 24:56
  SPEAKER_01: Um what we've seen is that it's quite confusing for the humans in the mix to understand how AI in quotes makes mistakes.
• 25:12
  SPEAKER_01: And because I think the mental model that people hold in their heads is that this this is just like um regular technology.
• 25:22
  SPEAKER_01: It's like Excel on steroids or something along those lines, right?
• 25:26
  SPEAKER_01: And it's just an it's an upgrade to our existing uh to our existing version.
• 25:30
  SPEAKER_01: But the thing is that if anybody's used any form of AI, whether it's uh generative AI like Chat GPT or I don't know, or if they drive if they do full full self-driving in a Tesla, what they'll notice relatively quickly is it's not that it uh doesn't get things wrong, it's that it gets things wrong in ways that uh don't uh normally compute with the way that we expect expect those things to go wrong.
• 25:58
  SPEAKER_03: Yeah.
• 25:58
  SPEAKER_01: Um and so and so for for example, that can lead to things like overtrust from a human.
• 26:06
  SPEAKER_01: Like um, and we did an experiment in our uh in our organization whereby we asked a group of people to ask GPT um some questions about something that they were expert in, that they personally were expert in.
• 26:19
  SPEAKER_01: And they would obviously critique the responses in some detail.
• 26:24
  SPEAKER_01: And then take those very same people having already got answers that they considered to be suboptimal from GPT and have them ask it questions about a field that they were not expert in.
• 26:34
  SPEAKER_01: And even though GPT had just got the answers from their perspective wrong in their own field, we asked them to rate the validity of the answer from GPT in a field that they were not expert, and they rated it incredibly high.
• 26:47
  SPEAKER_01: Because of this, we think, because of this underlying expectation that it's sort of like Excel.
• 26:52
  SPEAKER_01: In other words, if you ask Excel equals three plus four, you expect Excel to come back with seven.
• 26:58
  SPEAKER_01: How do you think as humans that are interacting with this and we're dealing with something that is inherently probabilistic in nature in terms of the answers that it's that it's giving, how do we as humans adapt and adjust to the fact that we're dealing with something immensely powerful, but also um almost human in the way in which it can make mistakes?
• 27:21
  SPEAKER_01: Do you know what I mean?
• 27:22
  SPEAKER_01: So it's it's like how do we how do how do we bridge that as we're trying to be truly productive and effective with it?
• 27:28
  SPEAKER_03: So interesting.
• 27:29
  SPEAKER_03: You know, the I'm gonna give an example of a project that we worked on that that dealt with this.
• 27:34
  SPEAKER_03: But I will say that, you know, for a long time we've been writing software in a really deterministic way.
• 27:39
  SPEAKER_03: Uh-huh.
• 27:40
  SPEAKER_03: And when you do, when you do testing, where we used to do testing, you would just know this is a bug because it's not working the way that I thought it was going to.
• 27:48
  SPEAKER_03: And now though the bets are off, you know, with a lot of this because it's hard to be deterministic anymore and it's hard to understand how to test some of this software.
• 27:58
  SPEAKER_03: We uh we had a project where, and this is live now, so people, your audience, if we put this in the notes, can go check this out.
• 28:05
  SPEAKER_03: Absolutely will.
• 28:05
  SPEAKER_03: This was this was a company uh in Kirkland, Washington that was a tax advisor.
• 28:09
  SPEAKER_03: It's called Formations.
• 28:11
  SPEAKER_03: It's actually an interesting company because they're really tax advisors for businesses of one.
• 28:16
  SPEAKER_03: That's their cut, that's their customer profile.
• 28:19
  SPEAKER_03: People that have very, very small businesses and are trying to figure out should I be a sole proprietor or an S-corp?
• 28:24
  SPEAKER_03: How should I do?
• 28:25
  SPEAKER_03: How should I deal with my taxes?
• 28:27
  SPEAKER_03: And they had uh some folks on staff that were amazing tax experts related to this particular field.
• 28:33
  SPEAKER_03: And they wanted us to build a conversational AI experience that answered tax gens for people that were businesses of one.
• 28:41
  SPEAKER_03: It was very, very important that it got the answers right.
• 28:45
  SPEAKER_03: And what we found at first was when we when we broke up all the content that we were given from this company and just had large language models go directly to that content and give back answers, it was getting confused.
• 29:00
  SPEAKER_03: But it was getting confused in a very, very convicted way, confident, you know, and so it was hard to know.
• 29:07
  SPEAKER_03: Like it sounded good, you know.
• 29:09
  SPEAKER_03: And and so we but the experts in the business were looking at what was coming out on the other end, and they were saying, this is not right, this is not acceptable.
• 29:16
  SPEAKER_03: Yeah and so what we had to do was we had to just tell the large language models to not use any of its general knowledge, and we had to point it at a curated database that this company curated that we knew was clean, up to date, and correct.
• 29:35
  SPEAKER_03: And we told the large language model because we knew that it was processing language in a in a great way, we did it, we did certain things where we said, if you don't have the answer, you have to tell the user that you don't know that you need to direct them to a human.
• 29:50
  SPEAKER_03: And we just forced that.
• 29:52
  SPEAKER_03: And and the accuracy dramatically increased when we did that.
• 29:57
  SPEAKER_03: Yeah.
• 29:57
  SPEAKER_03: And so I think that there is this notion now, and this kind of feeds back into your question about workers.
• 30:03
  SPEAKER_03: You know, in 1996, there was this famous quote about content being king, and we had these content managers.
• 30:10
  SPEAKER_03: That was a job, content manager.
• 30:12
  SPEAKER_03: I think there's a new job out there, and that's knowledge manager.
• 30:16
  SPEAKER_03: Because content, as you know, with the work that you do, is so easy to create now if you have a back-end knowledge base that's that's good and clean.
• 30:25
  SPEAKER_03: You can create chat experiences, you can create blogs, you can create articles, all this stuff can get created from this golden database.
• 30:32
  SPEAKER_03: And so I think that managing that knowledge inside a business, the brain of this business now becomes even more important.
• 30:38
  SPEAKER_03: And there's going to be people, workers, that that's their job is to make sure that that knowledge is good and that is up to date.
• 30:45
  SPEAKER_03: And then we learned a lot from that experience.
• 30:47
  SPEAKER_03: And I think that a lot of people that are starting to use recall augmented generation and other kinds of techniques to go into proprietary databases are learning some of the same things.
• 30:57
  SPEAKER_01: You might have very interesting choice on your website, um, and uh which which I find really, really intriguing and I and I'll I'll mention, I'll just say I love.
• 31:10
  SPEAKER_01: You go to the homepage of your website, um right at the very start, right there, um is basically an AI chat experience right in front of center.
• 31:23
  SPEAKER_01: Um Robin.
• 31:25
  SPEAKER_01: Yeah.
• 31:26
  SPEAKER_01: It's unusual, right?
• 31:28
  SPEAKER_01: Even for organizations that are uh that are in this field, particularly if they're consulting in nature or something like that, that's not how they that that's not front and center.
• 31:37
  SPEAKER_01: Um what was the decision behind doing that and what kind of feedback do you get on that?
• 31:43
  SPEAKER_03: Yeah, so we built that, and the the name of the the conversational AI agent is Robin.
• 31:52
  SPEAKER_03: And but behind Robin is a database that we call a KGA database, stands for known good answers.
• 32:00
  SPEAKER_03: And we learn this through that experience that I talked to you about with the tax accountant.
• 32:04
  SPEAKER_03: So we can go in and curate a known good answer database, and we have the confidence that Robin will answer questions about our business in a way that is accurate.
• 32:14
  SPEAKER_03: You know, we feel very confident about her ability.
• 32:16
  SPEAKER_03: And if she doesn't know, she will say so, just like you know, the example that I described before.
• 32:22
  SPEAKER_03: And we thought it was important to surface that for customers that they could go in and ask us about pricing, ask us about how we work, ask us about our process and get answers.
• 32:31
  SPEAKER_03: And you know, it's interesting to think about the ways that a prospecting tool like this can give people answers, but also ask questions because Robin asks questions too.
• 32:42
  SPEAKER_03: And that's the thing that's different about Chat GPT than Max, who is this uh the the uh accounting, tax accounting uh bot and Robin, is that they will answer your questions all day long, but they'll also ask you questions too, because they're getting smarter the more that you talk to them.
• 32:59
  SPEAKER_03: And they're trying to, in Max's case, qualify you.
• 33:02
  SPEAKER_03: In Robin's case, she's not as hard of a qualifier as Max, but she's still asking questions and trying to understand about your business like a good salesperson would.
• 33:11
  SPEAKER_01: Yeah, and I would encourage people to go to the website.
• 33:14
  SPEAKER_01: Um, even I mean, hopefully they'll go to the website and um want to use your services, but even if they don't, I'd encourage people to go to the website and start interacting with it because what I think it does to your point, um, is it helps break through a couple of the myths that we're already forming about what these uh what these tools are like, right?
• 33:33
  SPEAKER_01: You made a very important point.
• 33:34
  SPEAKER_01: I want to make sure people don't miss it, right?
• 33:37
  SPEAKER_01: This whole idea of, you know, convicted, I mean, some people call it convicted lying, but this uh this whole idea of like sort of responding to everything with high with high conviction, um, that's a design choice, right?
• 33:50
  SPEAKER_01: It's a it's it's a design choice to be able to answer it in in that particular way, and then just as they all do, just have a little disclaimer say this may or may or may not be right, right?
• 34:00
  SPEAKER_01: There are alternatives to that, right?
• 34:02
  SPEAKER_01: It would be it is possible as a design choice to be able to say, to say, I don't know.
• 34:08
  SPEAKER_01: Um I've also seen uh implementations occasionally where, for example, um it says um I don't know, but this is my this is my low confidence guess or something along those lines, right?
• 34:20
  SPEAKER_01: These are all the these are all ways in which the responses can give an indication as to how much confidence the AI model has in the answer that it's giving you, which can be helpful or useful uh or useful information.
• 34:33
  SPEAKER_01: And then of course the other aspect of it is that there is a um we're starting to see models that are, to your point, not just answering questions, but asking questions.
• 34:43
  SPEAKER_01: I've seen models, I'm sure you have too, that are like trained in the Socratic method effectively.
• 34:48
  SPEAKER_01: And so they're they're the questioner and you're the one that's doing the thinking and they're drawing and they're drawing stuff that's out of you.
• 34:54
  SPEAKER_01: And people and I think that will really help people understand that it that just what they see with Gemini, what they see with GPT is not the world and it just getting smarter and smarter versions of that.
• 35:07
  SPEAKER_01: It's gonna be much more nuanced than that, right?
• 35:10
  SPEAKER_03: Yeah, and I think that um the other uh just you'll probably put this in your liner notes, but it's uh max.formationscorp.com.
• 35:18
  SPEAKER_03: You can see kind of how Max is different than Robin.
• 35:20
  SPEAKER_03: And that's the other thing I think that's interesting is that we have a lot of control over the persona of these agents.
• 35:26
  SPEAKER_03: And I know you know this because you play around with a lot of these, but they can have different personalities, they can have different goals.
• 35:32
  SPEAKER_03: You know, Max's goal is to qualify you because the problem that the company had was the salespeople were getting bombarded by unqualified leads.
• 35:40
  SPEAKER_03: So we were saying at the beginning, that's Max's goal, but it doesn't need to be.
• 35:44
  SPEAKER_03: And I think people should understand that there's a lot of nuance to the way that uh you can design these agents in the in the background.
• 35:53
  SPEAKER_00: Sorry for the second interruption.
• 35:55
  SPEAKER_00: There's still a few more minutes left in the episode, but I wanted to take a quick second to give you some timely information.
• 36:00
  SPEAKER_00: If you're listening to this episode, you probably realize that getting good at working with AI is about way more than just prompt engineering.
• 36:07
  SPEAKER_00: It's really about getting used to working with a new AI member of your team and figuring out how to adjust your working style to make the most of it.
• 36:14
  SPEAKER_00: Billion Minds has spent the last year researching the people that do that best, and they have brought it all together in a great course called Thriving with AI.
• 36:22
  SPEAKER_00: They're taking registration for their next cohort right now, and you can get a 10% discount by using the discount code HW.
• 36:29
  SPEAKER_00: So visit HTTPS colon slash slash www.billionminds.com slash AI dash readiness dash program.
• 36:40
  SPEAKER_00: Or just click the link in the show notes.
• 36:42
  SPEAKER_00: Uh anyway, you got the link.
• 36:44
  SPEAKER_00: Use that discount code to make sure that you are ready to thrive in a world increasingly dominated by AI.
• 36:50
  SPEAKER_00: Billion Minds, making all of us ready for the future of work.
• 36:54
  SPEAKER_00: And now, back to the podcast.
• 36:57
  SPEAKER_01: Boas, let's talk briefly about co-pilots.
• 37:00
  SPEAKER_01: Um, some people are becoming more familiar with co-pilots either through Salesforce or through um or through the Microsoft co-pilots, the um the various releases that Microsoft's done with that.
• 37:11
  SPEAKER_01: Um again, for uh for anybody who hasn't really explained, uh hasn't experienced a copilot, if you could explain what that is, how that works, but also where you see those going.
• 37:21
  SPEAKER_01: Do you see them again kind of fully blending into the background, even though there's a lot of marketing spent on them?
• 37:28
  SPEAKER_03: Yeah, it's a great uh great question.
• 37:30
  SPEAKER_03: So, copilots, you think about them as an assistant, an agent that lives next to whatever you happen to be doing as you're doing your normal work every day.
• 37:41
  SPEAKER_03: So the classic example, and our engineers use this and get you know 40% uh increase in productivity is GitHub Copilot.
• 37:48
  SPEAKER_03: So GitHub copilot lives next to your coding, and these software engineers are coding away, and it's recommending what it thinks should be the next line of code or the next lines of code.
• 37:59
  SPEAKER_03: And you can accept that or you can change that, but it's kind of always happening.
• 38:02
  SPEAKER_03: You see this a lot with Microsoft copilots uh write a better email or you know, uh do a better job, you know, uh adjusting the article that you happen to be writing.
• 38:12
  SPEAKER_03: You know, if you think about like recommendation of writing a little bit farther ahead of you, always giving you a little bit of coaching, that's a copilot.
• 38:19
  SPEAKER_03: And what we're starting to do is starting to build custom co-pilots because there's a I think that some of the bigger companies are gonna be building co-pilots for the software that we normally use as workers, you know, uh, whether it's uh whether it's a Microsoft tool or whether it's a Notion or a Salesforce or other kinds of tools that we just use every day.
• 38:42
  SPEAKER_03: But there's gonna be some things that we want copilots for that are not so common.
• 38:47
  SPEAKER_03: You know, and we're starting to think about how do you have that?
• 38:50
  SPEAKER_03: How do you leverage a large language model that sits next to some form you're trying to fill out or some website that you're trying to scrape and do research on something?
• 38:58
  SPEAKER_03: You know, there's a bunch of things that are a little bit, I just think you'll have copilots for all sorts of stuff.
• 39:03
  SPEAKER_01: So huge field, massively changing field.
• 39:07
  SPEAKER_01: Um, and so it's obviously something that that uh that your business is built about, uh is built around.
• 39:14
  SPEAKER_01: But it's also something that very clearly is a personal passion of yours, and I know that um that's leading to some of the discussions you have on the Shift AI podcast, which I talked about at the outset here.
• 39:25
  SPEAKER_01: Yeah.
• 39:26
  SPEAKER_01: Um for those in the audience that are not familiar with the Shift AI podcast, talk a little bit about the origin of that and uh and the nature of the guest feature and what your goal is with it.
• 39:37
  SPEAKER_03: Yeah, the podcast was started actually when I was at Meta.
• 39:40
  SPEAKER_03: And we were doing a lot of productivity tools and we were thinking a lot about the future of work, and I wanted to be able to highlight people that were thinking about the future of work.
• 39:51
  SPEAKER_03: And as AI took hold, more and more of those people were in the AI space.
• 39:55
  SPEAKER_03: And when I left uh Meta, GeekWire, which is a local uh Pacific Northwest media publication, uh, started to pick it up, and the guests that were part of the Pacific Northwest were showing up in GeekWire.
• 40:09
  SPEAKER_03: And so it kind of got more popular than it was when I first started.
• 40:14
  SPEAKER_03: And what was exciting about the time that that happened was that places like Seattle, places like San Francisco, you know, we're starting to see a lot more energy in the developer community, in the investor community, at the universities, with people thinking about this content.
• 40:30
  SPEAKER_03: And so I started putting more people on the show that were uh that I wanted to elevate that were in the community.
• 40:36
  SPEAKER_03: And so on the Shift AI podcast, we have founders, so young founders that are doing really interesting things with businesses.
• 40:42
  SPEAKER_03: If anybody's used read.ai, David Schim was on the show, or Udley, which is another tool that was incubated outside of the um the AI2 incubator in Seattle.
• 40:52
  SPEAKER_03: We've had politicians on the show like Joe Wynne.
• 40:54
  SPEAKER_03: Joe is a state senator in Washington, and he just uh you know put together the AI task force bill that just went through the Senate.
• 41:01
  SPEAKER_03: So we're, you know, we're thinking about it from a policy standpoint.
• 41:04
  SPEAKER_03: Then we have investors from Madrona, from AI2 that you know are talking about what they're seeing is they're getting pitched all day long, and they get a set, they get a kind of vision into this world that normally most of us don't get.
• 41:15
  SPEAKER_03: So I like to talk to them.
• 41:16
  SPEAKER_03: And then luckily in Seattle, you know, we happen to be in Amazon and Microsoft land, and we have the benefit of talking to these massive players that are you know really, really far ahead.
• 41:27
  SPEAKER_03: And so we try to get a lot of folks.
• 41:28
  SPEAKER_03: We had the head of Gen AI from Amazon AWS on the show.
• 41:32
  SPEAKER_03: And you get these uh insights that are just uh fascinating.
• 41:35
  SPEAKER_03: But the reason really I like what you're doing, Paul, is that I try to keep it personal.
• 41:39
  SPEAKER_03: If anybody listens to the show, they know I ask questions about people's background, I ask them about their mentors and their family.
• 41:45
  SPEAKER_03: Because I want to kind of not just make it all about AI hype.
• 41:50
  SPEAKER_03: You know, I want to understand like personally where these folks are coming from and how this is impacting them.
• 41:56
  SPEAKER_03: So it's a it's it's a it's a nuanced approach to technology.
• 42:00
  SPEAKER_01: Okay.
• 42:01
  SPEAKER_01: I want to finish this off with uh with a question that almost like brings us full circle from from where we started.
• 42:08
  SPEAKER_01: Um so when we started, if you remember, we were talking about where the world's gonna be in in six years' time.
• 42:15
  SPEAKER_01: And of course, that's gonna mean that as individuals, as humans, we're all doing, we're all doing different things at that point.
• 42:21
  SPEAKER_01: So what's your what's your what's Boaz's day gonna look like in 2030?
• 42:28
  SPEAKER_01: My day.
• 42:29
  SPEAKER_03: Well, I happen to be the kind of person that enjoys talking to other people.
• 42:36
  SPEAKER_03: It just so happens that for the last several years, it's been on calls like this.
• 42:42
  SPEAKER_03: In the last year, it's been a lot more in person.
• 42:44
  SPEAKER_03: I imagine that in 2030, I'm gonna continue to be interfacing with people a lot.
• 42:50
  SPEAKER_03: And I'm I'm sure that a high percentage of that is gonna be on my computer screen, but not all.
• 42:57
  SPEAKER_03: I think that um there is a lot of manual work that I do that interfaces between those conversations.
• 43:04
  SPEAKER_03: I think a lot of that's gonna go away.
• 43:06
  SPEAKER_03: You know, the the way that this conversation is cataloged is gonna just disappear and it's gonna show up for me in ways that are impactful for me.
• 43:13
  SPEAKER_03: And I think there's gonna be a lot of other things that um are part of my everyday that are co-pilot-like, recommendations.
• 43:20
  SPEAKER_03: Hey, you haven't talked to this guy in a while, you should probably reach out, or hey, you know, you might want to think about you know, doing this, you haven't done this in a while.
• 43:27
  SPEAKER_03: Just like tips and and things that are uh that I have to remember that I'm no longer gonna have to remember.
• 43:33
  SPEAKER_03: And so I I I'm excited about the future, and I'm excited about uh being able to continue to engage with people.
• 43:40
  SPEAKER_03: Just I I think I'll have some some friends helping me along the way, some AI friends helping me along the way.
• 43:47
  SPEAKER_01: Well, here's to a future, an AI future where AI is enriching the human experience at work and outside of work.
• 43:54
  SPEAKER_01: Um it's a great positive note to end on.
• 43:56
  SPEAKER_01: Boaz, thanks for joining us on the pod today.
• 43:58
  SPEAKER_03: Paul, thank you so much.
• 43:59
  SPEAKER_03: It was great talking with you today.
• 44:01
  SPEAKER_00: Humanity Working is produced and edited for Billion Minds by Matt Neal.
• 44:05
  SPEAKER_00: Our interviewer is Paul Slater.
• 44:07
  SPEAKER_00: Music is by Dolo Records.
• 44:09
  SPEAKER_00: If you like this podcast, please subscribe, like, and recommend us to others.
• 44:13
  SPEAKER_00: It helps us reach more people.
• 44:15
  SPEAKER_00: If you want to learn more about how Billion Minds gets companies ready for the future of work by building adaptable, resilient employees, look for them on LinkedIn or visit their website at billionminds.com.