YT YouTube2Text

YouTube2Text → Transcripts → [Webinar] The AI Meaning Gap: New Research on Trust and Hidden Risks in AI Communications

[Webinar] The AI Meaning Gap: New Research on Trust and Hidden Risks in AI Communications — Transcript

by maslansky + partners · 8,623 words · 1,380 segments · language en · Watch on YouTube

Open in transcriber Summarize with AI Download SRT/VTT

Full transcript

0:05All right.
0:07Welcome everyone. Hopefully uh
0:10you can hear us uh fine. My name is
0:12Michael Mazlansky for those I don't
0:14know, and I have with me a colleague
0:16Jason Martisello who uh who has really
0:19led this research and I will be taking
0:21us through a lot a lot of it.
0:23Um
0:25but um
0:26uh we're going to jump in. I know that
0:28people are still joining
0:29uh um and as you can see we we dressed
0:32for the occasion by deciding to wear
0:34exactly the same outfit as we did this
0:36webinar. It's part of the value that we
0:38bring to every conversation like this.
0:41Um
0:42but uh um
0:43really really looking forward to
0:45presenting this uh to you. Hopefully you
0:46will you will jump in in the chat, ask
0:49questions, make comments
0:52uh as we go through um as I think
0:54ultimately this will probably be one in
0:57a series of pieces of work that we do on
0:59this because of the importance of the
1:01topic that we're going to talk about. Um
1:03and so uh we're going to go into what
1:06we're calling the AI meaning gap and uh
1:09what we think this research reveals
1:10about the importance of of building
1:13trust uh in your AI communications.
1:19So, question uh Jason, if you go to the
1:22next slide.
1:29Go to the next slide. Oh, you're there.
1:31Sorry. No, sorry, the previous slide. I
1:33was on the wrong slide. Um
1:35so, first question is
1:37why did we do this work? Um I think that
1:39there were there were really a couple of
1:41reasons driving it. Uh the first is
1:45uh
1:45it it seems like everybody is talking
1:47about AI in every conversation, whether
1:49it's a a business conversation or a
1:51personal conversation. Certainly among
1:53our clients, it seems like almost every
1:56conversation that we have has an AI
1:58component to it.
1:59Um, and at the same time,
2:01uh, investors are driving, uh, a lot of
2:05demand to hear what companies are doing,
2:07what investments they're making, how
2:08they're getting those investments to pay
2:09off. Uh, and from what we've seen from a
2:12lot of clients, the the AI narrative, so
2:15to speak, that, uh, that companies are
2:17building is really an AI investor
2:19narrative. And, um, that is a a
2:22critically important aspect of talking
2:24about AI,
2:25uh, but it's not the only thing. And,
2:28um, and for companies that are really
2:30focused on how to talk about AI for
2:33investors, on the next slide,
2:36um,
2:37one of the important things that we've,
2:39uh, that we see in a lot of the work
2:40that we do is that investor-focused
2:43communication is not focused only on
2:45investors, that media is covering,
2:49uh, what investors are communicating
2:51about. I see somebody saying they're
2:53getting feedback. Hopefully, it's only
2:55them who's getting feedback. Um,
2:57uh, that that any investor communication
3:00bleeds out to a whole host of other
3:02audiences, probably the most important
3:04of which is employees,
3:05who, uh, who are likely to be following
3:08the message that's delivered to
3:10investors and maybe taking away a very
3:11different message. Uh, it's going to
3:13communities, it's certainly, uh, being
3:15covered by the media, policymakers,
3:17customers, uh, etc. And so, companies
3:20need to be increasingly aware, if they
3:23are not already, about how that AI
3:25narrative is playing with these other
3:27audiences, even if they're just writing
3:29it for AI.
3:30And we're operating in an environment
3:32that I think probably everybody on this
3:34call recognizes is an extremely
3:37difficult environment. Uh, next. Um,
3:41you know, whether it is looking at
3:43stories about the future of work, uh,
3:45and the fear that many employees at many
3:48companies have around job cuts,
3:51the, um, seemingly overwhelming amount
3:54of messaging, particularly as we
3:55approach the the midterms of data
3:57centers, and, uh, and really what I
4:00would say is kind of the unprecedented
4:02backlash against data centers that we're
4:06seeing in many markets around the
4:07country.
4:09And then the the seemingly regular
4:11stories about AI safety concerns,
4:14breakouts, other issues. All of these
4:18have a really significant impact on how
4:21these other audiences outside of
4:22investors and even investors to a
4:24certain extent, but but particularly
4:26other other audiences are going to hear
4:30messages about AI.
4:32And so if we
4:34think about kind of where a lot of
4:36companies started with their AI message,
4:39the focus has been, how do we
4:42demonstrate AI leadership? How do we
4:45show that we are not behind our peers?
4:47How do we talk about the investments
4:50that we're making and the opportunities
4:53that we have, how good the future looks
4:55in a world where we are implementing AI.
4:57Because companies I think more than
4:59anything else
5:01feared being left behind or being seen
5:04as being behind in the race to adopt AI.
5:07Now, what we're going to talk about
5:08today and I think is important here is
5:11there are the frontier model companies.
5:13I think where we are more concerned is
5:16every other company in so many
5:18industries that are adjacent or
5:20completely separate, whether it's health
5:22care, energy, telecom,
5:26uh
5:26food and beverage, you name it, that
5:28that still are focused on communicating
5:31a narrative about what they're doing on
5:33AI and had been focused on leadership.
5:36And I think now given the the broader
5:38set of issues that we just talked about,
5:40really need to be thinking about
5:42something beyond leadership. They need
5:44to be thinking about acceptance. Is the
5:46message that they are communicating
5:47about AI going to be accepted? Are they
5:50going to be building trust or losing
5:52trust
5:54in the context of communicating about AI
5:56in this environment that we're operating
5:58in today?
6:01So as many of you know, probably all of
6:03you,
6:04we are guided by this idea that it's not
6:06what you say that matters, it's what
6:08your audience hears. And this is one of
6:11those areas
6:12where working with several of our
6:13clients, we had identified a gap that
6:17the AI narrative needed to evolve and
6:19needed to evolve quickly. And so we
6:21decided to take a broader look at it
6:23here
6:24and try and share some of what we've
6:26learned in that process.
6:31>> Awesome.
6:32Thanks, Michael. I want to first start
6:34by just running through a little bit of
6:37a practical exercise and poll. I just
6:39want everyone to just take a moment and
6:41just read this.
6:44I'm going to give you like 10 seconds
6:46here, 15 seconds.
6:48And this comes from an actual company's
6:50AI announcement.
7:02Okay.
7:04We're next going to move to a poll. I
7:05want everyone just to imagine a second
7:07that you are an actual employee of this
7:09company.
7:10Now, you're in the mindset of an
7:13employee, what what would you take away?
7:15What is the first question you'd ask
7:17yourself having just read that
7:18announcement on AI efficiency?
7:21I would love for anyone to just pipe
7:23their answer into the chat, whether it's
7:25A, B, C or D. What was What was the
7:29question that first came to mind as you
7:32read that announcement?
7:38Awesome. Seeing
7:41Oh.
7:43We have a poll coming up.
7:44>> There's a poll.
7:44>> There we go.
7:54>> Do we let let's pull that down. I think
7:55we've got it in the chat.
7:57>> Yeah.
8:05Okay, let me Let me re-share this again.
8:07Um
8:09Okay, so
8:10we've seen we've seen many different
8:12responses seeing all over the all over
8:14the map, the A's, the B's, we saw some C
8:16and D. Um
8:19and there's really no right answer to
8:20this. Everyone's entitled to their own
8:23question. The key takeaway from this is
8:26the you know, just to showcase that
8:28there's a dispersion in how people react
8:30to that same message is the first key
8:32thing is just seeing how broadly people
8:36perceive just one company's
8:37announcement. So, it reflects what
8:39Michael was just alluding to in terms of
8:41communicating for one audience and we're
8:43seeing different responses how it may
8:45resonate or you know, how people may
8:47have different questions or react
8:49differently.
8:50And particularly, I wanted to showcase
8:52some of the research uh findings we had
8:55on AI messages before we dive into you
8:57know, showcasing some of the research we
8:59did. But typically, with efficiency
9:01messages, we are we've been starting to
9:04hear a lot of um responses as such. Um
9:08you know, we will not need as many sales
9:10people or quote unquote people are
9:12losing you know, essentially this
9:13translates to people using losing jobs.
9:17In short, these have become proxies and
9:19heuristics for you know, job cuts,
9:22reducing head count and you know,
9:24getting introducing fear amongst you
9:27know, particular employees.
9:29Um and this is just you know, keying up
9:31the bigger idea that there is you know,
9:33there's a gap between what what we're
9:35saying and what people are hearing. So,
9:38in terms of today, we're going to go
9:39through an agenda. We'll spend a little
9:41bit of time breezing through some of the
9:44uh you know, what we tested. But we
9:46really want to spend a lot of time on
9:48you know, getting into where leadership
9:51and trust diverges and most importantly,
9:53the payoff. You know, what could we do
9:55about it? What is the winning language
9:57between what separates um you know, high
10:01uh high trust and high AI leadership
10:03from those that don't build trust in the
10:05AI narratives that they produce. And
10:08ultimately, we'll share some of our
10:09implications
10:11um
10:12and uh wrap things up. And we'll save a
10:14little bit time for questions at the
10:16end, but as Michael alluded to, if
10:17there's questions, comments, feel free
10:19to throw them into the chat. We'll We'll
10:21try to bring them in when they're
10:22relevant.
10:24So, first, you know, this whole
10:26presentation is around this gap, that
10:27divergence between what is taking place
10:30in the environment when it comes to AI
10:32communications. You know, there's
10:34certainly been a lot of investments and
10:36a lot of uptake in AI, but what does it
10:39look like on the other side of the
10:40fence? Um and we're already alluding to
10:43a gap, and that's essentially what we
10:45found when it came to, you know, several
10:48different things. And I'm going to first
10:50point to this top bar uh chart here on
10:54favorability. And one of the interesting
10:56things is, you know,
10:58and this is
10:59pretty pretty expected, I think, of most
11:01people. The more you use AI, the more
11:03favorable you are. You see that dramatic
11:06in uh uptick in terms of favorability,
11:09in terms of um
11:11in terms of how much you're using it.
11:13Now, the the interesting thing is, no
11:16matter how much you use it, it doesn't
11:18change your uh concerns around AI's
11:21future. So, all this to say is, people
11:24are really concerned about the future
11:27prospects of what AI's effects are on
11:30society.
11:31And that's certainly on the minds of
11:33many across all those that we sampled.
11:36In terms of favorability, when we This
11:39is These are net favorability uh This is
11:41a net favorability index, and it's
11:43compared and across different
11:45industries. So, AI relative to
11:47different, you know, other industries.
11:50And essentially, AI is, you know,
11:51ranking towards the bottom relative to
11:53other um industries.
11:56And what's more important is if we take
11:58a step back and look at this over time,
12:00um what Pew data shows is that this uh
12:04concern for AI is actually increasing
12:07over time. So,
12:09our study kind of validated that it is
12:10in a low place, um and, you know,
12:13looking at historical data patterns over
12:15time, it is worsening.
12:18And one of the interesting things is
12:19when you look at this. So, we looked at
12:21this amongst knowledge workers
12:24at different levels within an
12:25organization. And the thing is how you
12:28articulate
12:30your AI narrative and your strategy, it
12:33typically is going to resonate and
12:35you're going to hear it just how you say
12:36it up top. But the thing is when it that
12:38cascades down to our employees,
12:41it's not true. It's not same. They're
12:43not understanding it. So, for this,
12:45there's a quite a drop-off when it comes
12:48to understanding what AI is being used
12:51for in their organization and what it
12:53means for for their actual role.
12:56And when we look at this, you know, we
12:58across many different areas of
13:02of how they're understanding AI and some
13:06of the gaps in communication from higher
13:08levels of organization down to, you
13:10know, just you know, lower level or
13:12associates, there's quite a substantial
13:15gap and it's consistent across every
13:18different metric when it comes to
13:20understanding, being able to
13:23have confidence in that approach or and
13:27essentially, this just reiterates that
13:29this there's this broader gap with how
13:33communications of our AI approaches are
13:36are feeding through all the way down to
13:38our employees, all the way to our
13:39customers, and just more broadly into
13:42society as well.
13:44>> And I'll just add the
13:46the sample. We'll talk a little
13:48more about it in a second. This is a
13:51knowledge worker sample. If we think
13:53about what's happening in organizations,
13:55we did not look at field force. So, if
13:57you think about line workers in energy,
14:00you think about retail associates in a
14:02in a retail environment or store
14:04restaurant associates in
14:07in a a food and beverage context.
14:11We know from all the other work that
14:12we've done that messages that
14:14that struggle to get communicated even
14:16within HQ or among corporate employees
14:20are much harder to communicate more
14:22broadly out to the field force. And so,
14:24these numbers are likely to be much,
14:26much lower
14:28among the field force and in ways that I
14:30think are going to be really important
14:32for companies to to acknowledge and do
14:35something about as
14:38as AI starts to really impact even even
14:40whether or not they have jobs in the
14:41future.
14:46>> And to finish this section off, I'm just
14:48I'm leaving with a quote that says they
14:50they said AI helps with the boring stuff
14:53so people can focus on bigger tasks. I
14:55just hope it actually works out that way
14:58for my role. This came from an actual
15:01quote from from the study, which I'm
15:03just going to get into a in the sample
15:04in a moment. But essentially, this
15:07highlights some of the fear and
15:08uncertainty that is commonplace as AI
15:12takes a more center stage across all
15:15different aspects of life and work.
15:19There's a question around getting the
15:20slides. We will
15:22circle back after the webinar to be able
15:24to retrieve a recording and follow-up
15:27materials. So, yes.
15:29In terms of how we tested,
15:32so
15:33I'm going to dive into the kind of
15:35methodological approach, what we tested,
15:38with who.
15:39But first, I want to share a little bit
15:41about that hypothesis that Michael
15:43alluded to earlier. It's the the reality
15:46that, you know, AI ambition is taking a
15:49lot of central
15:51the center center stage in terms of
15:53investments, people putting money and
15:55effort behind pushing it into all
15:57aspects of organizations, our work, and
16:00personal lives.
16:01But the reality is this is one of the
16:03fastest-growing technologies, faster
16:05than cell phones, faster than the
16:08internet, and the ability for us to
16:10catch up as humans has not has not been
16:14sufficient. And this is creating what we
16:17are
16:18thinking as of as the trust gap. And
16:22this is what we set to measure out and
16:24to understand how we can help solve that
16:26with language and communication.
16:29So we So what we did, our approach for
16:32this, and we modeled this after
16:35a scientific academic study that had a
16:38very innovative approach to testing
16:40coms, which I thought was
16:42I definitely want to give some credit
16:43to, but we we we started with 10Ks. And
16:47looking at the item ones in 10Ks where
16:50people discuss their AI plans. And we
16:55took 15 narratives,
16:58the AI narratives from actual 10K public
17:02filings. And these were what we tested.
17:06And I'll
17:07I'll get to them in just a moment. But
17:09so that was
17:10a very important part of this study.
17:12In terms of who we we spoke with, we
17:14spoke with US general pop, nationally
17:16representative, 801
17:19individuals, and then knowledge workers
17:21at which Michael alluded to, which which
17:23was a part of a lot of the stats we just
17:26teed up with, which are
17:28individuals who are working behind
17:31computer most all
17:32all day. And then lastly, what what are
17:36some of the outcomes we looked at? We
17:37looked at five different outcomes. We'll
17:38break down each of these outcomes in
17:40just a moment. But they they're along
17:42the lines of trust, comfort, support,
17:44leadership, differentiation. Different
17:46outcomes we would want when it comes to
17:48developing AI communications.
17:52So as I mentioned, we test different
17:53narratives. We extracted these from the
17:5510Ks, from the item ones. There
17:59different Fortune 500 companies we
18:01pulled.
18:02Two of them we actually
18:05developed or as artificial stress test.
18:09Ones that were more extreme outliers
18:11that talk about AI employees. And then
18:14the second one was recursive
18:16self-improvement. Essentially the idea
18:19that a lot of these
18:21large language models are going to be
18:23operating
18:25independently of humans as they progress
18:28and get more
18:30efficient over time.
18:32And I will note one of these, I'm not
18:34going to disclose which one, we did play
18:37a role in in terms of helping
18:41actually is actual language that we
18:42tested in terms of developing for one of
18:44our clients.
18:47Okay, so in terms of how we scored, as I
18:50mentioned, it there's different metrics,
18:52outcome metrics. The first top two are
18:55ambition, which kind of map to what
18:56Michael was mentioning earlier,
18:58positioning ourselves as leaders in the
19:00field in terms of showcasing how we're
19:03maximizing the use of AI, how we're
19:05differentiating from our peers,
19:08ultimately looking to get credit from
19:09investors. We want to use it in a very
19:11strategic way. The second piece is these
19:15different metrics around acceptance, so
19:17trust, they're trusting in the approach,
19:19people are comfortable
19:21in how they're using it. And then, you
19:23know, support for it more generally
19:25speaking. And ultimately what these kind
19:27of ladder back to is that AI narrative
19:29score, which is kind of an index of all
19:32five of these kind of average to 100.
19:34And then we had certain diagnostic
19:36metrics which helped us play a role in
19:38understanding what are some of the
19:39nuances of why some were performing
19:42better than others, which kind of will
19:44shape some of the insights, the language
19:46learnings, which we'll get to towards
19:49the end.
19:53So, where language and trust diverge.
19:56So, amongst the findings, I think one of
19:59the most prolific findings is the
20:03in terms of after we showcased AI
20:05narratives was the limited amount of
20:09trust people had in the company's
20:12approach to AI. So, the AI narratives
20:15were certainly not getting to
20:20a point where people felt trustworthy of
20:23what they were intending to do with AI.
20:26And this is this is pretty profound
20:29largely because, you know,
20:31there's
20:32the trust in AI is going to be critical
20:35as you'll come to learn in just a moment
20:37to seeing what helps separate a lot of
20:40these companies
20:42from their peers in terms of building
20:45Go ahead.
20:46>> Well, I was just going to say, I mean, I
20:47think if if we if we take a step back
20:49and look at this for a second and think
20:50about in in your businesses whether you
20:54have been thinking about how to infuse
20:57AI messaging into your messaging whether
20:59it's directly to to customers
21:02or to other audiences and how in in many
21:05of those contexts we are we are trying
21:07to build messages that kind of sell the
21:09benefits of our products or the
21:11positioning of our business
21:13and we're trying to build trust.
21:15And here we took you have 15 different
21:18companies and what they are doing
21:20they're kind of best efforts at
21:22positioning their investments in AI and
21:26and finding that only 37% of Americans
21:30trust those companies in the aggregate
21:33in terms of how they're using AI. It it
21:35does not suggest that there is a lot
21:39out there as we think about how we use
21:41AI in our messaging
21:43a lot to suggest that adding more AI
21:45messaging in general is going to help
21:47build trust
21:49you know, across the different things
21:50that we talk about.
21:56>> I want to move next to another poll and
22:00I would love for you to read each of
22:02these. We have two different kind of AI
22:05narratives and I'm curious
22:07which of these AI narratives earned more
22:10trust. So, just take a a quick read of
22:13each and then when you have formed an
22:16opinion if you could just place your
22:18answer in the chat.
22:38>> Okay, we're getting a lot of answers.
22:42>> A lot of B's.
22:43>> A lot of B's. A lot of B's. Um
22:47It's good to good to see people are
22:48tracking along. Um
22:51There's a
22:51a lot of the B's are going to indicate
22:53that yes, we are we are thinking in the
22:56right direction that uh narrative B um
23:00is certainly going to, you know, score
23:02higher amongst uh in in terms of
23:04building trust.
23:06And
23:07when I look at
23:08the lay of the land in terms of how
23:10these scored out across the two the five
23:13metrics, um trust um versus kind of the
23:16AI leadership. And by the way, we group
23:19them AI leadership uh and
23:22differentiation into that ambition
23:25metric. And then trust, inclusive of
23:27comfort and support. So, and when we
23:30look at those across the different AI
23:33narratives we tested, what you'll see in
23:36this graph is that there is a delta
23:40between how they're performing on
23:42leadership and also on trust.
23:45And that is consistent across all of
23:48these. And this showcases that trust
23:51deficit um across all these you know,
23:54the bar for hitting AI leadership is not
23:56as high as building the trust. And that
23:59is critical because that's the big
24:01opportunity we see when it comes to
24:04finding the right words to showcase that
24:07we're not only going to be, you know,
24:09commanding a better strategic and
24:11differentiated position, but also how we
24:13can
24:14um build the trust of all of our
24:16stakeholders.
24:17And interestingly, when you look at it,
24:20you know, so we had the two different
24:21audiences,
24:22um
24:24and you know, when you look at these
24:25different dimensions,
24:28um on the left here, they're ranked on
24:30AI leadership. So, we could perform very
24:32well on AI leadership and perform not,
24:35you know, we can bomb completely on on
24:37on trust. The point is when it when it
24:40comes to winning on one metric, um it
24:44may actually, you know, kind of cause
24:46you to not perform well when it comes to
24:49trust. And, you know, some of these may
24:50be well-intentioned. Um for for
24:52instance, some organizations may be
24:55looking to get more investment and, you
24:57know, trying to shoot for, um you know,
24:59be more in aligned with some of the
25:01frontier models and usage of AI
25:04capabilities, where others uh but for
25:07most, they're trying to build broad um
25:10broad understanding and belief and trust
25:13across their employees, across their uh
25:16stake their investors, policy leaders.
25:19And as we take a a broader view in terms
25:22of how what we say has consequences for
25:24our different audiences, it's important
25:26to know winning on one particular
25:28dimension may actually not be winning on
25:31what we can consider the most important,
25:34and that is going to be trust.
25:37Um so, when we look at and this this
25:39gets back to when we scored some of the
25:42um
25:43the actual uh companies based on their
25:46AI narratives, you know, when it comes
25:48to where the widest trust gap um showed
25:51up, it it was more around many of the
25:54companies usage of language that is
25:56aligned with a lot of the frontier
25:58models. So, leaning into, you know, some
26:00of the digital labor, the agentic
26:02layers, the, you know, reduced human
26:05involvement, or, you know, instance uh
26:08robo-taxi, which involves no people. So,
26:11you know, more of that autonomy, these
26:12platf platforms that are, you know,
26:15really pushing the boundaries of the
26:16future, um certainly are where
26:19um that trust
26:21um
26:22there was that trust deficit and where
26:24leadership, you know, certainly rose up.
26:27Um on the bottom where there's the
26:28narrowest gap, there was a lot of
26:32um things that were uh oriented around
26:34how um
26:36certainly human oversight um and what
26:39how the human ultimately played a role
26:41in that broader view of how the
26:43narratives are connecting back to uh
26:46people, the organization it benefits,
26:49and how the stakeholders ultimately will
26:52um confer benefits over time. So,
26:55that's this is I think an interesting um
26:58point of view on that. And then when we
27:00look at leadership and trust and map
27:03this out. Um so, on the left axis we
27:05have a the acceptance from bottom um
27:08going up to
27:10uh higher trust, support, comfort. And
27:12then on the bottom on the x-axis we have
27:14ambition, and from the left to the right
27:17we improve on leadership and
27:18differentiation.
27:20And ultimately this is just
27:22a composite map of where we fall out
27:24just based on how we're we're performing
27:28on on on both of these different
27:29metrics. And the ideal state is being in
27:32the trusted leaders camp where we're
27:35able to position our AI
27:38uh narrative in a in a way where we're
27:40building leadership and differentiation
27:42from our peers, but also um we're
27:45trusted amongst our constituents, our
27:47stakeholders. And ultimately this will,
27:50you know, put us in a better place, you
27:52know, not only from a competitive
27:53strategy perspective, but also ensuring
27:56that we're not reducing um
27:58any trust when it comes to any of our
28:01stakeholders.
28:02>> And just to kind of pause on this for a
28:03second. So, a couple of things that that
28:05I think are important.
28:07Um first, while we're not saying which
28:10which of these narratives we worked on,
28:11it may be the one in the, you know, top
28:14um top of the box. Um
28:16you know, just
28:18that's up there in the top orange. But I
28:20think the bigger point is um uh is that
28:24there does not need to be a trade-off
28:26between communicating about leadership
28:28and building trust. You don't have to go
28:30far in one direction to talk about uh
28:34about your innovation or about the
28:36things that AI may be taking on
28:39without also kind of building support
28:42and acceptance with the with the
28:45constituents that matter most whether
28:46it's employees or the broader your
28:48broader customer population and that
28:50ultimately as we get to the language
28:52that
28:53that resonates the most it is really
28:55trying to to do the best that you can in
28:58terms of selling your leadership
29:00while also doing the best that you can
29:02in terms of maintaining or building the
29:04trust of your audiences
29:05and that is where I think you the world
29:08is evolving pretty rapidly to to a place
29:11or from a place where
29:13it was fine to just be a leader
29:16you know a couple months or a couple
29:17years ago because we were still in the
29:20early days of talking about AI
29:23but if you buy into the hypothesis now
29:25that you have all these other audiences
29:27that you have to
29:28make sure that you are also managing
29:30with your communications then then you
29:33cannot choose one over the other you
29:35have to choose both and that's what
29:36we're going to get into in terms of the
29:38the language recommendations.
29:46>> Well so when we look at
29:48some of the best narratives and you know
29:50kind of dissect them and understand how
29:53we can build trust while also
29:57maintaining AI leadership and
29:59differentiation what is what what stands
30:02out is not on how well they do on
30:06amongst you know the people that rate
30:08highly right amongst you know so on the
30:11left we have strong confidence top two
30:13box people who rate that trust comfort
30:17and support you know it's pretty flat
30:19right amongst the top tier and the
30:21bottom tier
30:23but where there's a bigger difference is
30:26on the bottom two box and you know
30:28that's trust comfort and support on the
30:30right amongst top and bottom tier
30:32narratives and essentially what this is
30:34telling us is that um and then this
30:37comes back to the broader thing is is
30:39it's not we're just not just building
30:41fans, we're actually reducing people
30:44that are resisting um our narratives and
30:47our companies and our positioning cuz
30:49ultimately, you know, AI has taken on a
30:53new meaning across the world and, you
30:56know, people that we may not intend for
30:58it to be relevant for, it is they're
31:01still grading. They are still hearing
31:02you and they are still, you know, having
31:06an opinion and perception that
31:07ultimately is going to weigh into how
31:10they think about companies uh more
31:12broadly. So,
31:13this is I think super interesting as we
31:15get into our next area.
31:18>> Just go go back one one second, Jason.
31:20The The other thing about this, I think
31:22we we've seen this in the responsible
31:24business work we did on ESG, we've seen
31:25it in the age of rage work, and we see
31:27it in a lot of certainly in a lot of the
31:29data center work that we're doing now,
31:31which is um
31:33which is often it it is more important
31:36or at least as important in building an
31:39effective message to um to reduce the
31:42negativity associated with your message
31:44as it is to kind of build the
31:45positivity. And that the the backlash
31:49risk in more and more situations today
31:52for companies poses a greater risk than
31:54kind of being a little bit more plain
31:57vanilla or a little less compelling on
31:58its face. And so, so we look at kind of
32:01the net or the the the impact of bottom
32:04two box uh perceptions of these
32:06narratives as being uh really a
32:09significant um um measure of how
32:11polarizing your message is, how much
32:13risk there is in communicating this
32:15message, and maybe tapping into a well
32:17of discontent uh that you just don't
32:19want to touch.
32:26So, we next want to move to three
32:28language lessons that we've taken away
32:30from uh this work and that we want to
32:33share with you all.
32:36And the three principles really come
32:38back to one, who remains in control?
32:42Two, who benefits? And then three, what
32:45makes this promise credible?
32:48And ultimately, each of these is going
32:50to, you know,
32:52connect how we can convey and build
32:56trust with our AI narratives and still
32:59while maintaining that AI leadership and
33:01differentiation.
33:03And we're going to showcase and break
33:05down each one of these. But all that to
33:07say, each of these, um, you know, is
33:10grounded based in some of those metrics
33:12that we learned. So, for instance, show
33:14who's in control,
33:16human oversight was the top attribute,
33:18which, you know, really showcased that
33:21humans are present. The second, um,
33:23around making the benefit clear, this
33:26was, you know, one of the highest
33:28priorities people wanted when it came
33:29to,
33:30um,
33:31certain specific communications that,
33:33um,
33:34um, making them clear and impactful was
33:36how they will benefit people. So,
33:38connecting that back and then that
33:40third, um, you know, this was actually,
33:43um,
33:44you know, the number one thing when it
33:46came to priorities people wanted to
33:48explain. So, ultimately, these were, you
33:50know,
33:50these showed up across many different,
33:53uh, areas in terms of, you know, what
33:55people are wanting when it comes to how
33:58we can make our AI narratives build both
34:00trust and leadership. And we're going to
34:02go through each of these in turn to kind
34:03of bring them to life a little bit more.
34:06We'll first start with, uh, show who's
34:08in control.
34:10Um, and you know, as AI becomes more
34:13ubiquitous in our environment, it is
34:16very much more important as we start
34:18relinquishing some of the activities and
34:20tasks to AI. It is becoming increasingly
34:23important to
34:25show who's in control, to talk about
34:27how,
34:29um, AI is going to be governed. This is
34:31somewhat a little bit around the
34:32regulation that people want to know
34:35where the the lines in the sand are when
34:38it comes to, uh, who's in control, uh,
34:41and how AI, as it becomes more powerful,
34:44where does that power, uh, who possesses
34:46that power? So, you'll see on each of
34:48these upcoming slides on the left, this
34:50is language that, um, loses trust. This
34:53is stuff that really taxes that trust
34:56gap. It, you know, kind of creates more
34:58of that deficit. And you'll see
34:59highlighted language here, but uh,
35:02excerpts from the AI narrative and
35:04specific things that are highlighted,
35:05but ultimately things that are on full
35:07self, uh, self uh, full self-driving,
35:11robo-taxi, AI robots. These are things
35:14that, uh, reduce that perception of
35:16control. They limit how we perceive
35:18trust in that AI strategy, that
35:21narrative, and ultimately that company.
35:23Um, and on the right, here's an example
35:25of, uh, a trust-building AI narrative.
35:29And you'll see here, "Make everyday work
35:31simpler and more meaningful by reshaping
35:33roles to emphasize uniquely human
35:36strengths such as creativity and
35:38leadership. Identifying areas where AI
35:40can automate repetitive tasks." So, you
35:43could start to see how that feels more
35:46trustworthy. You see the bounds of
35:48control and how, um, humans ultimately
35:51are being empowered, um, and where some
35:54of the lines are limited, uh, in terms
35:56of AI and in repetitive tasks. There's
35:58very few people who are going to
36:00disagree about AI removing repetitive
36:03tasks. So, the net Go ahead.
36:05>> Go ahead. Finish.
36:07>> Yeah, the net net here is really just
36:09making sure the human is present in the
36:12narrative.
36:14>> Yeah, so, I I think the
36:17um, we just finished another piece of
36:19work for a client. I think some may be
36:21on, on here, um,
36:23where we we did split sample testing of
36:26where we used the term AI and we either,
36:29uh, tested a similar phrase without AI
36:31or, uh, with technology, some form of
36:34cutting-edge technology in there. And
36:37uh, and we saw that, um, that in most
36:40cases adding AI into the message
36:43actually had a negative impact on
36:45favorability towards the message.
36:47And and so it becomes even more
36:48important that when you are going to use
36:50AI that it that it has its proper place
36:53in a sentence uh as much as in a message
36:56itself. That you know, here you can see
36:58that the reference to AI comes later in
37:01the message. It is a a kind of how we do
37:03the thing that we have set up at the top
37:05of the message versus on the left side
37:08where AI comes first. Um that there is
37:10all of this sensitivity now to the role
37:13that it's playing and uh and being able
37:16to kind of reinforce the fact that
37:18um
37:19uh that uh
37:21um that the human is in the loop uh is
37:24uh and in control is really critical.
37:26And and I see the comments on on super
37:28intelligence. We'll we'll come back uh
37:30once we get through this and um
37:33uh and talk a little bit about that.
37:36>> Yeah, one final note on this. It it it's
37:39not that autonomy is the villain in this
37:43story.
37:44It's really the ambiguity is, right? So,
37:47making sure that we are highlighting the
37:51control. Uh it's not to say that people
37:54don't like AI removing or making people
37:57more efficient. It's really having the
37:59understanding of the control is going to
38:02be the critical piece.
38:05On our second is making the human
38:07benefit um clear.
38:10And
38:12I think you know, when we think about
38:13clarity and human benefit, I think we're
38:17I think many miss um the next layer to
38:20get to the human benefit. So, what I
38:22think we find often in looking at a lot
38:24of these AI narratives is, you know, we
38:26end at efficiency and that or you know,
38:29productivity and you know, those
38:32necessarily are not translating fully
38:35down to what that means to employees, to
38:37customers. Um so, I think it's really
38:40about taking it a step later uh a step
38:42deeper and making that more concrete.
38:45It's like um
38:47So, for example in terms of time
38:48savings, like what does that make
38:50possible? And you'll see here we have an
38:52example on our left and on our right.
38:54Um
38:55Uh in terms of the losers trust, our
38:57customers can build and deploy always-on
38:59digital label digital digital labor for
39:02employees and customers leveraging
39:04autonomous AI across business functions
39:06that aim to increase productivity, lower
39:08costs, and drive operational efficiency.
39:11And then on the right, we have AI's
39:14potential reimagine healthcare including
39:16simplifying care navigation, reducing
39:18costs, transforming consumer experience,
39:21enhancing the quality of and access to
39:24care. Now, both of these weave in AI to
39:28you know, kind of broadly speaking
39:30improve productivity, experience, cost.
39:33I mean, you could see on the right
39:34though it's more specific and tailors
39:36back to actually that human benefit.
39:38These are things people want and desire.
39:41I mean, I think they broadly abstract on
39:43you know,
39:44business benefits, but it pulls it
39:47through to a more human benefit all the
39:50way to the right. And you know,
39:52ultimately it is efficiency, it is
39:54productivity, it is making more money,
39:56but at the end of the day tying that
39:58back and making that human benefit very
40:01salient as we could see in this
40:03healthcare example, um this is really
40:06going to help bridge that trust gap and
40:09also still showcase how you are still a
40:12leader in this space.
40:15>> Yeah, and I think you know, for many
40:16industries that are on here and that we
40:18work with, you know, tying it back to
40:21even if you are talking about
40:22productivity or lower costs or
40:24operational efficiencies, if that's what
40:25you what you want to talk about, it is
40:27connecting that to an affordability
40:29benefit to customers cuz that is
40:30presumably where some of those cost
40:32savings are going to go.
40:33Um but it needs to be made explicit
40:35because just the kind of the
40:36profitability benefits which um
40:39which investors may may want and want to
40:43see, um
40:45those benefits can be really dangerous
40:47in communicating out to uh to the
40:49broader public.
40:56>> So, our last principle is really back
40:59that promise with proof. And
41:02when we say, you know, adding technical
41:03specificity can that can establish a lot
41:06of the credibility that's needed when it
41:08comes to AI, but the question is is
41:10really
41:11uh whether an audience is going to
41:13understand why those details matter.
41:16So, I'm going to showcase a couple
41:17different examples. On the left, we have
41:19losing trust. With increased processing
41:22power, smartphones and PCs are becoming
41:25per pervasive AI platforms with complex,
41:28large generative AI algorithms running
41:31on device, enabling on-demand contextual
41:34AI use cases.
41:35Um
41:36and then on the right, we have we use AI
41:39to create better customer experiences,
41:41including analyzing more than 1.2
41:43trillion network alarms to spot and
41:45resolve issues before they disrupt daily
41:47life.
41:49So, what we're seeing here is that we're
41:52both have, you know, kind of, you know,
41:55different elements of proof and, you
41:56know, uh bring that down to concrete
41:59example, but
42:00on the right, this is really tying back
42:03and, you know, makes this into an
42:04understandable
42:06um kind of plain-spoken way to back up
42:10that um you know, the actual um promise
42:13that we're trying to make. So,
42:16you know, specific is important, but
42:18it's not the same thing as being
42:19meaningful when it comes to uh the proof
42:22we're trying to convey to our audiences.
42:30And lastly, now we want to just get to
42:32some of the implications.
42:34And, you know, we're going to take a
42:35step back from the examples, but, you
42:37know, just think through this more
42:38holistically and how this matters for
42:40you and your audiences and how we can
42:43tell better AI narratives um when it
42:46comes to
42:48um connecting our AI strategies um to
42:51the real world.
42:52Um and what we have here are just four
42:54different implications.
42:55>> Yeah.
42:57>> And Michael, do you want to take this?
42:58>> Yeah.
42:59>> So, um
43:00so I think, you know,
43:02as we think about where we are, we've
43:03covered some of this already, but the
43:04the the first and um
43:07I think most significant is that this is
43:09a different moment, and so a lot of the
43:11narratives that have been created
43:12already um need to be reevaluated in the
43:15context of a a midterm post-midterm
43:19environment, the the relatively new
43:22excuse me AI safety issues, and um and
43:25the impact of these data center fights.
43:27Uh what we are seeing across the country
43:29is that
43:30the the shift in attitudes towards AI
43:34and certainly towards openness to
43:36uh data centers um to the negative is
43:39moving faster than any issue I've seen
43:41in 25 years of doing this. Uh and it is
43:45there is an overhang impact um on every
43:48message that is associated with AI.
43:50Um it both for internally for uh for
43:53employees
43:54uh and for the broader audience. And I
43:56would say that even though investors may
43:58feel like they are a different audience,
44:00in in many situations, they are
44:02responding in similar ways to consumers
44:06to a lot of these messages.
44:08So,
44:09um uh evaluating the the messages that
44:12are out there in the context of this
44:13stakeholder environment, um that uh
44:16recognizing that that being a leader in
44:18AI is not the same as building uh as
44:21building trust in your strategy,
44:23um that there's probably a need, no
44:25matter what you're saying, to
44:26communicate a lot more with your
44:28stakeholders, especially your more
44:29junior uh or more distant stakeholders,
44:32if you're operating in that kind of a
44:34corporate office and field environment.
44:36Um
44:37uh the third I think is that we said
44:39this about ESG in the midst of the
44:41anti-woke backlash, that the year of
44:43performative ESG was over, and that
44:46companies really needed to focus on the
44:47substance. I think the same is now true
44:50with AI, that that companies
44:52excuse me, could get away with a lot of
44:55messaging
44:56around what AI was going to do or what
44:59our the pilot programs were going to do
45:01or what the plans were related to AI in
45:03organizations. And I think the bar for
45:07demonstrating that AI is now
45:10is now having a meaningful impact or
45:12that the investments are real and they
45:14are likely to turn into tangible
45:16benefits if they have not already is
45:18here.
45:19And that finally, you know, that
45:22shocking to hear this from us, but that
45:24the language really does matter. That
45:26there is a
45:27universe of words
45:30and phrases
45:31that that, you know, that we're seeing
45:33come up in in
45:35Excuse me.
45:37In many situations where which have
45:39become coded as negative in the AI
45:42context much as language gets coded in a
45:44political context. And that
45:47the last thing that you want to do with
45:48your language is send an unintended
45:51negative signal by using a word that
45:54means something internally. But it has a
45:56much more negative meaning externally.
45:58And so finding that that compelling and
46:01credible and effective frame becomes
46:04even more critical.
46:11Jason, you can take this.
46:12>> Uh
46:13Yeah, so um
46:15you know, to kind of connect the dots
46:17back to what is taking place. There is,
46:20you know, historic investments in AI
46:22capabilities are
46:25you know, I think doubling every 6
46:26months from some of the stats I've
46:27heard. And you know, there's no doubt
46:30around
46:31some of this technology that's powering
46:34how we conduct our business.
46:37Um
46:38the key thing is really how we turn that
46:40into commercial value. And I think, you
46:43know, we talk I I heard a lot of
46:46um you know, examples around the tech
46:49stack. And I think one of the missing
46:51layers in that tech stack is actually
46:53language
46:54um in terms of translating that into
46:57actual value. Um, because otherwise, I
47:00think there's a lot of, you know, room
47:02for air or risk when language is not
47:06accounted for when it comes to
47:07communicating AI capabilities. We've
47:10certainly seen it it it
47:12a lot of language can be very polarizing
47:15when it come to AI. And also, when it
47:18comes to the stakeholders we need to
47:19account for. There's a lot of cascading
47:21effects as we think about AI
47:23communications more broadly. So,
47:25language we feel is the missing link,
47:28the missing piece. It's probably, as
47:29Michael mentioned, not um, this is not
47:33um, we are not coming from we are a
47:35little bit biased maybe in saying that,
47:36but ultimately, we do feel there is a
47:38deficit when it comes to how we're
47:40thinking about and communicating
47:42um, in AI. So, when it comes to how we
47:45think about and help a lot of our
47:47clients in this space, we've done a lot
47:50of work over the past year, year and a
47:53half helping clients and customers uh,
47:56turn their AI capabilities into
47:59compelling narratives. Not so they're
48:01just winning in um, in leading in their
48:05um, marketplace amongst their peers and
48:08competitors, but also so they have
48:10buy-in across their uh, employee base.
48:14So, that their
48:16uh, customers actually
48:18see them and trust them in in the
48:20future. Um, not just around their uh,
48:23you know, their current capabilities and
48:25their products. So, you know, what does
48:26this look like? This shows up in many
48:28different ways in which we engage with
48:30clients. From helping, you know, find
48:32that sweet spot and balance of
48:34positioning on leadership and trust and
48:36finding the right differentiation as
48:38more and more people start trying to
48:40take credit for AI and you know, finding
48:42ways to carve out your own lane and
48:44still build trust. The second is really
48:47around um, employee AI narratives.
48:50And this is becoming more and more
48:52popular as, you know, what we say has
48:55direct consequences for our employees.
48:57And it's almost just as important, if
48:59not more to make sure employees are on
49:01board with our AI communications
49:03externally as they are internally. So,
49:06we've done quite a bit of work to, you
49:08know, ensure that our AI narratives for
49:12our clients work in across external
49:15communications as well as internal. And
49:17third, trust and reputation. You know,
49:20certainly a lot of our clients uh have
49:23specific needs when it comes to trust
49:25and reputation when it comes to AI and
49:27specific nuances with uh audiences. And
49:30this is where um we certainly come in
49:33for some bespoke engagements to really
49:35fine-tune what is that right language to
49:37win over uh specific stakeholders um for
49:40those particular um moments that really
49:42matter. And then, you know, I think
49:44number four is one I've I'm finding more
49:47often than not is the way in which our
49:50several clients I've seen, they all they
49:53they have different AI stories, but
49:55sometimes it's taking a step back and
49:57making sure those stories connect and
49:59ultimately align towards a broader
50:01vision in terms of how there's
50:02alignment. Um because how AI
50:05capabilities
50:06um within some of our clients, it is
50:08vastly different. So, how we have to
50:11articulate and ladder that back to a
50:13broader kind of enterprise AI story is
50:17really going to make the difference for
50:20a lot of the audiences um and
50:22stakeholders that really buy in and will
50:24make uh that a trustworthy as well as
50:27differentiated uh approach.
50:33I'm just going to end on one slide and
50:35then we'll open this up with for
50:36questions, but you know, and it's really
50:39about when you think about your your
50:41cust your um
50:43your constituents, your stakeholders,
50:45your customers, your investors,
50:48when you show up, what do they hear with
50:50when you're when you put out your next
50:52AI announcement? And um
50:54I'll stop there. Um
50:56we'll see I know there's a couple
50:57questions and we'll address those, um
50:59but if you have any questions, just pipe
51:02those through into the Q&A the chat and
51:04we can start addressing those.
51:07Um, I don't know, why don't we start
51:08with I think
51:09>> Yeah. I'll I'll take the one that that's
51:10in there and maybe we can pull the deck
51:12down. Um,
51:14so Peter and Della, thanks for the
51:15questions about supreme intelligence or
51:18sorry super intelligence.
51:20I voted for supreme intelligence
51:22in the president's poll. Um, but and the
51:26question of
51:28what people think of it and and whether
51:29it will get used which I think are two
51:30very different questions in this
51:31context.
51:33In the we've done a little bit of
51:34testing on it. And you know, the notion
51:38of taking a term that already scares
51:40people in artificial intelligence and
51:43then trying to elevate the power of that
51:46technology by calling it super
51:47intelligence. Um,
51:50would seem to be a backwards strategy in
51:53terms of trying to calm people's nerves
51:55about it. Obviously that was not
51:57necessarily the president's goal in
51:59renaming it, but I think
52:01it does not serve the purpose of making
52:04people feel more comfortable
52:07with the technology by calling it super.
52:09If anything it does the opposite and
52:12which is a very different
52:14I think question than whether or not
52:16companies will start to adopt it because
52:18they feel like they have to because they
52:20work with the government
52:23either because they they
52:25they feel like they have to or they
52:26actually have to in order to work with
52:28the government. I suspect you know, just
52:30based on this article
52:32that we will see some of that. I just
52:36I don't think that it helps
52:38move the conversation forward because I
52:40really do think that it makes this
52:42technology and then from what we've
52:44heard it makes this technology sound
52:45scarier rather than
52:48less so.
52:49Um,
52:51let's see.
52:53Hi Katie, do we ask anything about
52:55messengers are and who the most
52:57effective messengers are? We did not
52:59cover that.
53:00We did not cover that here. I think the
53:03Um,
53:04the question is well to some extent
53:06depend on what it is that we are uh that
53:09we are talking about. If we're talking
53:11about the
53:13uh um
53:15you know, if we're talking about
53:16specific you know, member benefits or
53:19customer benefits, then somebody closer
53:22to them that can actually make those
53:24benefits real, you know, if it's a uh
53:27uh uh a care provider who, you know, in
53:29the case of health care who can talk
53:31about how the technology is helping make
53:34their jobs easier so that they can serve
53:36members better. Uh I think that's a
53:38great messenger. I think if you're uh
53:41but bringing it closer to the
53:42beneficiary would be my general
53:44perspective, but we did not um we did
53:47not test it here.
53:50Um I don't know if there are any other
53:51questions.
53:53But um
53:55uh thanks everybody for joining. Thanks
53:56to our team, Jason, Z Lee, Randy for
53:59helping put this together. And um and
54:02let us know if you have other questions
54:03or want to explore different areas.

About this transcript

This page contains the full transcript of [Webinar] The AI Meaning Gap: New Research on Trust and Hidden Risks in AI Communications by maslansky + partners, generated from the public captions YouTube serves with the video. The transcript has 8,623 words across 1,380 segments, with the original timestamps preserved so you can click any line to jump to that moment in the embedded player.

What you can do with it

Use the transcript to take notes, quote the speaker, build a study guide, generate a summary with ChatGPT or Claude via the YouTube Summary tool, or export it as a timed subtitle file with YouTube to SRT. You can also re-open it in the transcriber to choose a caption track available on the video.

Free YouTube transcript tool

YouTube2Text is a free YouTube transcript generator — no signup, no daily limit. Paste any YouTube link and get the full transcript instantly, with timestamps, click-to-jump, caption language selection, AI prompts for ChatGPT, Claude, and Gemini, and exports to TXT, SRT, VTT, or Markdown.