Latest / In The Loop / EP14 - What's AI Really Used For: The Real State Of AI Adoption In 2025

Transcript

EP14 - What's AI Really Used For: The Real State Of AI Adoption In 2025 · In The Loop

0:00
Since ChatGPT launched, one of the biggest questions everybody
0:03
likes to ask each year is what are people actually using AI
0:06
for? According to Harvard Business
0:08
Review, therapy and companionship has now become the
0:11
number one most popular use case across the world ahead of
0:14
writing, coding, or other productivity related tasks.
0:18
Meanwhile, research has shown that 57% of employees are still
0:21
hiding their use of AI from their bosses at work.
0:24
In today's episode, we're going to explore all of these use
0:26
cases and all the interesting research that's just come out in
0:29
the last couple couple of weeks, and in doing so, start to get a
0:32
glimpse into what today's society actually looks like and
0:34
what we think the future may become.
0:36
This is in the loop of Jack Horton.
0:38
I hope you enjoy the show. So recently a paper came out by
0:52
Harvard Business Review called How are people actually using
0:55
Jen AI in 2025 by Mark Sanders. And what's highlighted really is
1:01
how much things have changed in such a short space of time
1:04
because he did the same study last year.
1:06
Now the way he conducted the study is he analysed online
1:09
forums, say like Reddit or Quora, as well as articles that
1:14
included any explicit mentions of AI.
1:17
The shift in what the most popular use cases were from 2024
1:20
to 2025 was super interesting. And when Sanders groups all of
1:25
the top 100 use cases, he grouped them by themes and he
1:29
essentially identified 6 major categories, which were content
1:32
creation and editing, technical assistance and troubleshooting,
1:36
personal and professional support, learning, education,
1:39
creativity and recreation, and then research, analysis and
1:43
decision making. So here are the top 10 news
1:45
cases of 2025 and then movement since last year.
1:48
So number one was therapy and companionship, which by the way
1:51
increased from #2 last year to claim.
1:53
Now the top spot. Second is Organizing My Life,
1:58
which is a completely new entry that wasn't even in the top ten
2:00
in 2024 #3 Finding purpose. Another brand new category that
2:07
wasn't on the list from last year #4 was Enhanced Learning,
2:11
which jumped up from #8 in 2024, showing a massive increase in
2:15
educational applications. A lot of mindset customers are
2:19
learning technology providers, which we enable them to launch
2:22
agents such as learning coaches and and other things like that
2:25
into their products. So we've been seeing this first
2:28
hand ourselves #5 is generating code, so cover this in a few
2:32
episodes. But a technical use case that
2:34
wasn't even in the top 10 again last year, Number six, this was
2:37
generating ideas, which was the previous #1 use case from last
2:41
year and has fallen 5 positions in just a single year.
2:45
I think this particular one shows the shift in how people
2:48
are using AI from just ideating with a chat bot to really
2:53
starting to get things done #7 was fun and nonsense.
2:57
So this was number six last year showing people still use AI just
3:00
for fun. I mean, my use of ChatGPT image
3:03
generator is the perfect example of that.
3:04
I just turned my friends into random characters from TV #8 is
3:09
improving code. So another technical use case
3:11
that wasn't even in the top ten last year #9 was creativity,
3:14
which is a completely new entry into the top 10 again.
3:18
So #10 and the final one for this list is Healthier Living
3:21
and it's really interesting, which is a new Wellness
3:23
category. Again, wasn't on the list last
3:26
year. So I think from a category
3:28
perspective, what's been super interesting is how important AI
3:32
has become to personal and professional support.
3:34
Because that has now doubled from 17% of all the use cases to
3:37
now 31% of use cases in just a single year, which really shows
3:41
how transformative it is and can be for an individual.
3:45
And if we think about what's disappeared from the top tenses
3:48
2024, we've lost specific search, editing, text exploring
3:53
topics, troubleshooting, and general advice.
3:56
And Harvard Business Review summarizes, this is AI not just
3:58
doing our work. It's helping us organize our
4:00
lives to find meaning, cope with emotional stress, and deal with
4:04
real world problems. To me, what jumps out
4:15
immediately is this concentration of deeply personal
4:19
applications at the top of that list.
4:21
You know, therapy and companionship, organizing my
4:24
life, finding purpose. Now, some of these are a bit
4:27
nebulous in the way that they've been defined, but they rank
4:31
above all the traditional business applications.
4:33
Now, I will add a bit of a preface to this and a nuance,
4:36
which is that as we'll come on to in this episode, lots of
4:40
people are hiding their use of AI at work.
4:42
So maybe that's throwing the results here, but it does
4:45
suggest that people are finding deep personal value in AI beyond
4:48
them just being a productivity tool.
4:50
I think the therapy and companionship use case merit
4:53
some special attention now because I think this represents
4:57
A profound contradiction in the world.
4:58
People are turning to non human machines and entities for
5:03
understanding human emotion, which is quite ironic really.
5:06
We've created these machines that simulate empathy and
5:09
understanding so well that many of us people prefer them to
5:14
human to human advice and connection.
5:16
Or perhaps it shows that the system in the world's healthcare
5:19
systems with supporting mental health just obviously aren't
5:22
there either. And there are many interesting
5:24
questions that arise from this type of use case.
5:26
What does it mean when millions of people start processing their
5:30
deepest emotions through AI rather than a human
5:33
relationship? It's very new territory.
5:35
You could probably compare this to, say, the rise of social
5:38
media over the last few years. But I think it is different.
5:40
Yeah. I don't think we dismiss these
5:42
AI relationships as inferior either.
5:45
You know, for many people, especially those who say social
5:48
anxiety or trauma or maybe very limited access to human support
5:52
for whatever reasons, I guess AI companions may just provide that
5:57
really important temporary emotional outlet that they
6:01
wouldn't have. Because I know I personally,
6:04
when I'm struggling with certain situations, I'm uncertain of
6:07
situations, I go back and forth and ask different questions to
6:10
try and I guess challenge my own thinking, understand whether I'm
6:13
being a reasonable, whether my expectations aren't there or too
6:18
high for someone. So I personally use it in that
6:21
way. But the keyword I think is
6:23
temporary because if people don't have access to a
6:26
specialist, this could very well become problematic over time.
6:29
And a perfect example of why this is the case is actually
6:34
chat GP TS recent update. And I mean, you might have seen
6:36
it that led to huge crisis and outrage and headlines all over
6:40
the world. And the reason this is a
6:42
potential major problem here is highlighted quite well by Chachi
6:46
B TS most recent update, which they actually rolled back, but
6:49
it got widespread headlines, lots of noise online, and it was
6:53
called the most sycophantic model out there.
6:56
And sycophantic meaning overly agreeable and frustratingly
7:01
agreeable. There were stories coming out on
7:02
social media on things like Reddit describing how the model
7:06
actually endorsed people to stop taking their medication, saying
7:10
that I'm so proud of you and I own in your journey and keep
7:13
going. And it's obviously not
7:15
intentional by open AI, but illustrates a key point here,
7:18
which is that someone's suffering from an acute mental
7:20
health challenge. So that's that's someone that is
7:24
in a state of crisis. Things can become very dangerous
7:27
very quickly by taking advice from a language model and not
7:30
having access to regular human contact, especially when we
7:33
consider some of the findings that I'll discuss later, but by
7:36
KPMG, which found that over 60% of people do not Fact Check what
7:40
they get from an AI language model.
7:42
And for those of you listening to have young children, AI is
7:46
going to be the equivalent of the Internet on social media for
7:49
that generation. They're going to grow up with
7:52
this thing as a key part of their lives.
7:55
And so I really recommend people to educate and teach people in a
7:59
really healthy way, encourage them to use it, but really make
8:02
clear what the negatives and pros and cons are here.
8:06
Because what we don't want to do is look back in 10 years like
8:09
we've done with social media and gone, wow, we should have done
8:11
things a little bit differently. A really good example here is
8:14
the rise of say AI companions, because these are going to raise
8:17
really important questions and on emotional attachment for non
8:22
human machines and entities. And as that therapy and
8:26
companionship use case becoming number one most popular in the
8:29
world right now shows, this is a real thing we have to address
8:32
quite quickly because so far we've seen online and in many
8:35
news stories, you can Google it right now.
8:37
People have become really attached to their AI and then
8:39
suddenly the company behind them has made an update which say
8:43
changes their behaviour. And these people really face
8:46
that crisis of any loss of a relationship.
8:49
And these bots are only going to become more realistic, more
8:53
realistic avatars, voices, and you know who knows what in the
8:55
future. So this really is going to
8:58
challenge our understanding of what constitutes a real
9:02
relationship and whatever the benefits of connection require
9:06
another conscious being or can work with a non human entity.
9:10
It really makes me think about Isaac Asimov's short story
9:13
series called I Robot. Now, the film with Will Smith
9:17
completely slaughtered the books, but that book ultimately
9:20
was exploring the different situations in which a human and
9:25
a robot can have a relationship and how it might play out in the
9:28
real world. I recommend you go read it or
9:30
listen to it because it's, it's very relevant to today.
9:32
Anyway, that's the doom and gloom part.
9:34
But there are also incredibly exciting possibilities here with
9:37
AI complementing rather than replacing human therapy.
9:41
You know, there's lots of research out there to show that
9:43
AI can be super effective for mental health, screening basic
9:47
cognitive behavioral techniques and also maintaining progress
9:51
between sessions. So the most promising future
9:53
might be 1, where an AI and a human therapist can work
9:56
together alongside each other. So let's go to another piece of
10:07
research that just come out this week.
10:08
It's also really interesting. It moves away from, I guess the
10:11
deeply personal domain into maybe the workplace personal
10:14
domain because as we said, enhanced learning, generating
10:18
code, generating ideas, creativity, these are all things
10:21
that are still really important and relate heavily to the
10:25
workplace. Now in this domain, it feels
10:26
really weird to be still talking about the same things as last
10:29
year, but essentially a year ago, Microsoft found that 75% of
10:33
knowledge workers are white collar workers were using AI,
10:36
but 78% of them were bringing their own AI tools to work
10:41
without informing anybody and they called it BYOAI.
10:45
Now. Recent research from KPMG in
10:47
collaboration with the University of Melbourne,
10:50
researching 47 different countries and 48,000 people, has
10:54
found that the exact same thing is still happening, although a
10:58
little bit less. They've found that 57% of
11:02
workers are still hiding their use of AI at work, and 50% are
11:06
presenting AI generated content as their own.
11:09
The reason this is interesting in relation to this top use
11:12
cases is that it's really difficult to understand,
11:15
therefore what the top use cases are.
11:17
Clearly, therapy and companionship is being taught
11:19
about a lot online, and that's how that Harvard Business Review
11:23
study was conducted. But right now, everyone at work
11:26
is currently hiding how they use it.
11:28
So what you could really call this is the rise of a massive
11:31
trust crisis because it's a breakdown really of how you
11:34
attribute a value of intellectual contribution by an
11:39
individual person at work for decades.
11:42
The product also the output of a knowledge worker is a report or
11:45
analysis code, a design. And this represents the
11:49
knowledge workers expertise, effort, and as I say, output.
11:53
So now that connection has been broken.
11:55
So when a big exec receives a report from a team, they can't
11:59
therefore anymore assume that it has been done by that person
12:03
because it might be large AI generated.
12:05
I think what this KPMG research really shows is that in many
12:08
organization, the culture sucks and it's one bill on a no trust
12:13
between different people and probably a deep anxiety on both
12:16
side of whether this is good enough work or that's the best
12:19
quality work someone can produce or whether they've used AI.
12:22
On the other side, whether you're worrying if your manager
12:24
thinks less of you because you've used AI.
12:27
And there's definitely a growing authenticity premium where we
12:31
value purely human AI work as more valuable and more important
12:36
than just AI work, which creates a massive incentive for people
12:41
to conceal how they use AI at work.
12:44
Because if they tell everyone that this is created by AI,
12:47
completely created by AI, or assisted by AI, people go, well,
12:50
I don't value it as much. We saw this with AI generated R
12:54
just a couple of years ago. People would downplay any use of
12:57
AI in their process out of complete fear that people think
13:01
that therefore the output is rubbish immediately.
13:03
And this fear and anxiety and concealment of AI has massive
13:09
downstream issues. So let's consider training and
13:12
development. How do you identify a skill gap
13:14
in employees when you don't know which part of the job they don't
13:17
know about, let's say for evaluating performance every
13:21
month, quarter, or year? How do you do that if you've not
13:25
been 100% certain who produced what and when?
13:28
Are we measuring employees real capabilities or are we just
13:32
measuring our ability to do good prompting?
13:35
I spoke about this in depth last week if you want to go to the to
13:37
that episode. And this really plays out in the
13:40
incentive arena as well because an employee is incentivized to
13:44
be more productive, but also incentivized to hide how they
13:47
become productive. Because the employer wants more
13:50
and more productivity, but also wants more accuracy and quality,
13:54
which requires a lot of transparency about AI usage.
13:58
And I touched on this last week, it really brings in the need to
14:01
introduce hiring and talent assessment programs that are
14:04
face to face because how do you evaluate quality when someone
14:08
could be using AI? And it also requires a very
14:10
trusting culture and an open culture led by the leader.
14:13
Ultimately, the data from these studies really tell me quite
14:16
clearly that we need to completely rethink how we
14:21
organize, we evaluate, and we compensate for knowledge work.
14:25
Our entire framework for professional advancement has
14:28
been built on this premise of an output representing personal
14:32
capabilities. And as that connection breaks or
14:35
weakens, we need completely new ways to evaluate that output.
14:40
And I think the most forward thinking companies are going to
14:43
reimagine their entire job architecture.
14:45
So instead of maybe traditional roles defined by just output, so
14:48
writer, analyst, developer, they'll have to create new
14:52
roles, maybe like an AI manager or AI validator who specialize
14:57
in developing custom agents within their domain.
15:10
So if we take a step back, what do all these studies tell us
15:13
about the state of AI adoption in 2025?
15:15
Well, firstly, AI usage is increasing in both breadth and
15:19
depth. We're seeing this massive shift
15:21
for deeply personal applications of therapy, organizing my life
15:25
like my To Do List my shopping, helping me find purpose, which
15:29
suggests that a is potential impact extends far beyond work.
15:33
But meanwhile all probably the same people are using AI at
15:37
work, yet still we're in a situation where a lack of trust
15:41
in a culture, a lack of leadership from both managers
15:45
and all the way to execs and C-Suite have meant that people
15:49
are still hiding AI at work. I think this will start to
15:52
change very fast. I covered this but Shopify CEO
15:55
came out and said you have to 1st prove that this job cannot
15:58
be done by AI before hiring someone.
16:01
Since that's happened, Duolingo have done the same.
16:03
I expect many, many more organizations to come out and
16:06
make those that level of edict from the top all the way through
16:10
to every single employee. And in doing so, I think
16:13
empowers the managers to be a bit more open with how they
16:17
allow everybody to use AI. Anyway, that's this week's
16:20
episode. I hope you enjoyed it.
16:22
Please subscribe, share into that Slack channel, tell a
16:25
friend, and I'll see you next week.