Latest / The Human Factor: Exploring the Intersection of Humanity, Technology, and Transformation / Season 2 Episode 017: The Algorithmic Mirror – What AI Reveals About How We Actually Think, Decide, and Deny

Transcript

Season 2 Episode 017: The Algorithmic Mirror – What AI Reveals About How We Actually Think, Decide, and Deny · The Human Factor: Exploring the Intersection of Humanity, Technology, and Transformation

0:06
I want to start with something that happened at a client I was
0:09
advising last year. They had deployed an AI system
0:13
to assist with talent evaluation, promotions and
0:16
internal staffing decisions. Within three months, the head of
0:21
HR approached me while I was on site for a meeting.
0:25
She was visibly shaken, she said.
0:29
Kevin, the AI is biased. It keeps recommending the same
0:34
profiles, the same demographics, and the same employees for
0:40
promotions. I asked her to show me the
0:44
training data. We spent over 2 hours going
0:47
through it all and when we were done, she sat back and said
0:52
something that I'll never forget.
0:55
She said it's not biased, it's accurate.
1:00
It learned exactly what we've been doing for the last 15
1:04
years. You just never had a look at it
1:08
all at once before. That moment captures everything
1:13
we're going to explore today. The discomfort people feel
1:16
around AI is rarely about the technology.
1:21
It's about what the technology reveals about us.
1:27
Welcome to the Human Factor Podcast, the show that explores
1:31
the intersection of humanity, technology, and transformation,
1:37
along with the psychology behind transformation success.
1:42
In our last episode, we explored the middle management trap, the
1:47
five psychological burdens that organizations place on their
1:52
most critical change agents. Without acknowledgement or
1:56
support, we examined the translation burden, the
2:01
absorption burden, the identity burden, the loyalty burden,
2:05
along with the accountability burden.
2:08
One of the themes running through that episode was how
2:11
middle managers become mirrors for the gap between what
2:15
organizations say they value and what they actually reward.
2:19
Today we're going to take that mirror metaphor much further,
2:24
because there's another Mirror Now operating in virtually every
2:28
single organization, and it's far more unforgiving than any
2:33
manager's quiet frustration. That mirror is artificial
2:38
intelligence. I wrote about this extensively
2:41
in my recent newsletter, The Algorithmic Mirror.
2:45
What AI reveals about how we actually think and Decide.
2:49
So today I want to go deeper, drawing on research published
2:54
through early 2026, connecting this to the frameworks that
2:58
we've built across this season, as well exploring what I believe
3:02
is the most psychologically important question leaders face
3:05
right now. Not whether AI belongs in your
3:09
organization, but whether you are prepared to see what it
3:13
reflects back. AI is almost always framed as a
3:18
disruptive force, something that replaces human judgement,
3:22
automates your expertise, accelerates decisions beyond our
3:27
control. That framing dominates today's
3:31
boardrooms, media coverage, and vendor presentations, but it
3:36
misses what makes AI genuinely unsettling.
3:40
The discomfort most organizations and individuals
3:43
feel around AI is not really about technology or the fear of
3:49
being replaced by you. It is about the exposure that
3:54
results. The very core of human thinking,
3:58
human faults, human biases, and human interpretation.
4:03
All of this brought to the forefront, exposed and made
4:06
visible for everyone to see. AI does not just work
4:12
differently than humans. It reflects how we've always
4:16
worked, only more clearly, more consistently, and without the
4:22
social camouflage humans rely on to soften or hide uncomfortable
4:27
truths. Let me ground this in some
4:30
research. A 2025 study from the University
4:34
of Washington tested 528 participants across 16 different
4:40
job roles. What they found was striking.
4:45
When AI systems exhibited racial bias and hiring recommendations,
4:50
human decision makers mirrored those biases, selecting
4:55
candidates in line with the AI recommendations, but without AI
5:01
input. Choices were unbiased.
5:04
Or at least they were unbiased in the eyes of humans.
5:09
The lead researcher put it quite plainly.
5:12
Unless bias is obvious, people are perfectly willing to accept
5:17
the AI's biases as their own. Now think about what that means.
5:24
Humans feed AI the information on which it bases its decisions.
5:29
The AI learned human patterns. Then humans absorb those
5:35
reflected patterns and acted on them, often unconsciously.
5:41
This is not a story about flawed machines.
5:45
This is a story about human patterns being made visible at a
5:50
scale and a speed we were never prepared for, or perhaps never
5:55
really wanted to admit and reveal.
5:58
A comprehensive 2025 review published in Frontiers of Big
6:02
Data mapped the entire landscape of algorithmic bias and
6:06
categorized it into four different families.
6:09
The first, historical and representational bias, which
6:13
reflects the values embedded in the training data itself.
6:18
Second, selection and measurement bias, which reflects
6:22
what organizations choose to measure and also what they
6:25
choose to ignore. The third, algorithmic and
6:29
optimization bias, which reflects the objectives humans
6:34
define for the system. And finally, feedback and
6:38
emergent bias, which reflects the reinforcing loops between
6:41
human behavior and machine output.
6:44
Every single category traces back to human decisions, human
6:50
curated data, and human input. Not one of them is purely
6:55
technical, although we would like to think that is the case.
6:59
Here's what makes this so psychologically difficult.
7:03
For decades, organizations have relied on the opacity of human
7:07
judgement as a feature, not a bug.
7:10
Decisions could be explained through experience and tuition
7:13
relationships, all of the soft language that makes subjective
7:18
choices feel authoritative. AI removes that opacity.
7:24
It makes the logic visible. And when the logic is visible,
7:29
the inconsistencies become undeniable.
7:33
Research published in the Management Science in 2025 found
7:38
something both reassuring and deeply uncomfortable.
7:41
Even fairness aware machine learning algorithms, systems
7:45
never designed to be fair, can reduce bias in human decisions.
7:50
If an algorithm that was never designed to be fair produces
7:54
less bias outcomes than a human process, what does that tell us
7:59
about the human process? It tells us that the judgement
8:04
we trusted was never as objective as we believed.
8:08
We defined our own version of the truth and accepted
8:13
subjectivity as objectivity. This is precisely what I
8:18
explored in the revised version of my book The Truth About
8:21
Transformation Leading in the Age of AI.
8:25
When examining how humans shape and distort the systems that we
8:29
built. What we are seeing now in the
8:32
research confirms what that analysis anticipated.
8:36
The technology is not the problem the human system feeding
8:42
and using the technology, it is the actual problem.
8:46
So when organizations and individuals look honestly at
8:50
algorithm without outputs, 3 things consistently become
8:54
visible, and each one challenges a different aspect of
8:58
organizational self-image. The nearer I mentioned earlier
9:02
routinely reveals what gets rewards, not what is written in
9:07
mission statements, but what actually advances careers, earns
9:12
recognition, and avoids friction.
9:15
If an algorithm prioritizes volume over quality or speed
9:20
over reflection, it is because the system learned those
9:23
preferences from human behavior. Consider how we think and
9:28
construct the prompts we input, what language we use, how that
9:33
language can be evaluated, along with the inferences that AI
9:37
makes about what we are really seeking.
9:40
MI TS research from Sloan Management Review found that 26%
9:46
of respondents reported getting meaningful personal value from
9:51
AI while noting that their organizations derive little or
9:55
no value. That gap between individual
9:59
benefit and organizational return is not a technology
10:03
failure. It is a mirror that reflects
10:06
what organizations and their leaders actually reward versus
10:11
what they say they reward. When individuals extract value
10:16
but the organization does not, you are seeing the system where
10:20
personal optimization has been prioritized over collective
10:25
alignment. AI did not create that dynamic,
10:29
it made it measurable. Many roles assumed to require
10:33
deep expertise are in practice governed by heuristics,
10:36
shortcuts, and many unspoken roles.
10:40
AI does not eliminate judgement in these cases.
10:43
It reveals how thin it already was.
10:47
The California Management Review published a 2025 study that
10:52
should be deeply uncomfortable for anyone who believes human
10:56
judgement is inherently superior.
10:59
What they found was when AI was deliberately debiased, it
11:04
outperformed human decision making in board candidate
11:07
selection by enabling a shift from fast unconscious processing
11:13
to slow conscious evaluation. The researchers titled their
11:19
paper Slow Thinking Fast. It was a deliberate inversion of
11:24
Daniel Kinnaman's framework. What they demonstrated is that
11:29
AI can force the kind of deliberate reasoning that humans
11:33
claim to exercise but rarely do. It is not that human judgement
11:39
is bad. It's that human judgement is
11:42
often faster, more intuitive, and more socially acceptable.
11:47
And it is accurate. I wrote about this dynamic in my
11:51
newsletter on professional identity crisis.
11:54
When your expertise becomes obsolete, the paradox remains.
11:59
Here, the employees organizations rely on most the
12:04
go to problem solvers. The institutional knowledge
12:07
keepers often become the people most psychologically threatened
12:12
by change, not because they oppose progress, but because
12:18
systematic change threatens the very expertise that defines
12:22
their professional identities. This also connects directly to
12:27
what we explored in Season 2, Episode 1 on the identity crisis
12:31
of expertise. AI does not just threaten what
12:35
people do, it threatens the story people tell themselves and
12:39
others about who they are. Our new algorithmic mirror then
12:45
surfaces inconsistencies we likely didn't see before.
12:49
Humans routinely make exceptions, override rules, and
12:53
rationalize deviations. We are always seeking this way
12:58
others to our values and our ways of thinking.
13:02
When AI applies the same logic consistently, those
13:06
inconsistencies become visible, not subtly visible.
13:11
They become impossible to ignore.
13:15
Harvard Business Review reported in January 2026 that bias in AI
13:20
is not just baked in the training data.
13:23
It is shaped by the broader ecosystem of human AI
13:27
interaction. The way people engage with AI
13:31
through their thinking, their questions, their interpretation
13:34
and decisions significantly shape how these systems behave.
13:39
Organizations do not just train algorithms on their data, they
13:44
train them on their culture. And culture includes every
13:48
exception, every override, every unspoken compromise.
13:53
So the issue is not then that AI lacks judgment.
13:57
It is then that it forces organizations and individuals to
14:01
confront how selectively they have been using their own.
14:05
So why do we so very strongly resist the reflection?
14:10
Why do we not want to look into the mirror?
14:15
Resistance to AI is often framed as caution, ethics, risk
14:20
management, or even policy. Sometimes those concerns are
14:25
entirely legitimate. We know hallucinations occur.
14:29
We see deep fakes that appear disturbingly realistic.
14:34
We question whether our prompts actually produce what we
14:38
intended. But often the stated concerns
14:42
mask something much deeper Identity threat.
14:47
Let's take a deeper look at the psychology of AI resistance.
14:51
For many leaders and experts, authority has long rested on
14:55
opaque judgement, the ability to make decisions others can't
14:59
fully explain or replicate. AI challenges that mystique.
15:05
When an algorithm produces similar or better outcomes using
15:09
transparent logic, it raises an unsettling question.
15:14
What exactly has expertise been hiding?
15:18
Not hiding deliberately, but hiding structurally, because the
15:23
opacity of human judgement has never before been challenged at
15:28
this scale. A foundational study published
15:32
in Electronic Markets identified 3 central predictors of AI
15:36
identity threat in the workplace.
15:39
First, changes to work, the structural alteration of what
15:44
someone does day-to-day. Second, loss of status position,
15:49
the erosion of hierarchical authority.
15:53
And 3rd, what the researchers called AI identity, the degree
15:58
to which professionals see AI as challenging who they are, not
16:02
just what they do. That third dimension is the one
16:07
most organizations completely miss.
16:10
They address the functional threat, retraining, upskilling,
16:15
process, redesign. They mandate the use of AI
16:19
engage performance on such. However, they rarely address the
16:24
existential 1. A peer review 2025 study
16:28
published in NDPI Systems went further.
16:32
It identified 2 distinct dimensions of AI anxiety.
16:36
Anticipatory anxiety, which is driven by the fears of future
16:40
disruption, the worry about what might happen, and annihilation
16:46
Anxiety, reflecting existential concerns about human identity
16:51
and autonomy. Really, the fear that one's
16:54
professional self might cease to matter, The language itself,
16:59
aniliation, anxiety, captures something that leaders must take
17:04
seriously. This is not resistance born of
17:07
stubbornness. This is resistance born of an
17:10
existential threat to professional selfhood.
17:15
Think about what we had built across the season.
17:18
In episode 1, we explored the identity crisis of expertise,
17:23
how professionals whose value rested on specialized knowledge,
17:28
experience, AI is a fundamental threat to who they are.
17:33
In episode 2, we examined the contagion effect, how those
17:38
identity fears don't stay contained within individuals but
17:43
spread through emotional contagion, amplifying in each
17:47
organizational layer. In episode 3, we explored the
17:51
middle management, how people most responsible for translating
17:56
change in the reality are carrying psychological burdens
18:00
the organization refuses to acknowledge.
18:03
So now in this episode, we're seeing the mechanism that
18:07
accelerates all of those dynamics.
18:10
AI acts as a mirror that makes the gap between espouse values
18:15
and actual behavior visible at unprecedented speed and scale.
18:21
That visibility intensifies identity threat.
18:25
That identity threat feeds emotional contagion.
18:29
And the people caught in the middle of that cascade.
18:32
Your middle managers absorb the full force of organizational
18:36
anxiety without the authority or support to even address it.
18:40
This is not a series of separate problems.
18:43
It's one of interconnected psychological systems.
18:49
AI is the catalyst that makes the entire system visible and
18:53
very real simultaneously. In my newsletter on invisible
18:57
friction, I explored how strategies falter not because
19:01
people are obstinate, but because they fear loss.
19:05
They disengage not because they're lazy, but because they
19:09
feel disconnected from meaning. AI amplifies this dynamic
19:14
because it introduces a new form of loss, the loss of narrative
19:19
control. When an algorithm reveals that
19:22
your decision making was less subjective than you believed,
19:25
that your experience was thinner than you presented, that your
19:30
organization rewards behaviors it publicly condemns, you lose
19:35
the ability to maintain the story you've been telling
19:38
yourself and others. The most common response to that
19:42
loss is not adaptation, it's denial.
19:46
You're blaming the tool for revealing uncomfortable truths.
19:51
You're demanding human override without examining the reflected
19:55
human bias. You're rejecting outputs instead
20:00
of interrogating underlying assumptions.
20:04
In this type of situation, here's where AI bias becomes
20:07
genuinely dangerous. AI bias is not linear, it's
20:13
recursive. Organizations train AI on their
20:17
historical decisions. AI reflects those patterns back.
20:22
Humans absorb those reflections and then act on them, often
20:26
unconsciously. And the cycle repeats, each
20:30
iteration reinforcing what came before.
20:33
Research in Nature human behavior has shown that this
20:37
amplification is exponential rather than linear, with humans
20:42
often unaware of a is influence, making them then more
20:47
susceptible to the recursive pattern.
20:50
Meanwhile, transparency is declining.
20:54
As organizations become more dependent on AI systems, the
20:59
biases reflected back become harder to detect, harder to
21:03
attribute, and really harder to correct.
21:06
You end up in a situation where the organization is looking into
21:11
a mirror that is growing more distorted with each reflection,
21:16
and no one can see the distortion because they have
21:19
normalized the image. The scale of organizational
21:24
denial about this dynamic is completely staggering.
21:27
MI TS 2025 report the Gen. AI divide.
21:32
It was based on interviews with 150 leaders, surveys of a Hot
21:37
350 employees, and analysis of 300 AI deployments.
21:42
It found that 95% of corporate AI projects fail to create
21:47
measurable value, 95%. Meanwhile, 91% of CI OS cite
21:55
culture as the primary impediment to AI adoption,
21:59
versus only 9% citing the technology.
22:04
Read those numbers together. Leaders largely agree that the
22:07
technology works. The human systems don't.
22:12
Or more precisely, the human systems within organizations
22:17
refuse to engage with what the technology reveals, looking into
22:22
that algorithmic mirror, and they're choosing not to see.
22:26
Following on this, the McKinsey Global Survey on AI, which was
22:30
published in March 2025, found that the single biggest factor
22:35
in achieving measurable returns from AI is the redesign of
22:40
workflows, A fundamentally human and cultural undertaking.
22:46
High performing organizations were three times more likely to
22:49
have redesigned workflows around AI capabilities rather than
22:53
simply layering AI on top of existing processes.
22:58
The value of AI, then, comes from rewiring how organizations
23:03
run. That's not a technology project.
23:07
That is the transformation project.
23:10
And transformation, as I have argued consistently throughout
23:14
this podcast, my books, my newsletter, my research, lives
23:18
and dies in the human system. Most organizations,
23:22
unfortunately, approach AI implementation as a technology
23:26
deployment. They focus on selecting tools,
23:29
training users, building governance frameworks.
23:33
All that stuff is necessary, but it's all really inefficient
23:37
because they address the functional dimension while
23:41
ignoring the psychological one. Until implementation approaches
23:45
address the story people tell themselves and shape that
23:49
identity narrative, they will continue to fail at the rate MIT
23:53
document. I wrote about this cognitive
23:57
dimension in my newsletter on the mental overload of modern
24:00
leadership. Leaders today are already
24:03
operating at the limits of the cognitive architecture,
24:07
processing across multiple complex systems simultaneously
24:11
while their brains evolve for sequential processing.
24:16
Adding the eye to this environment without addressing
24:18
the fundamental mismatch between cognitive capacity and world
24:22
demands doesn't solve the overload problem, it simply
24:26
amplifies it. Our new mirror reflects faster
24:30
than leaders could process what they are seeing.
24:33
So what do organizations do with this?
24:35
How do leaders move from denial to engagement with what the
24:40
algorithmic mirror reveals? Let's reframe the question.
24:45
Most AI conversations focus on control.
24:49
How do we govern it? How do we constrain it?
24:52
How do we limit risk? Those are very necessary
24:56
questions, but they're very incomplete.
25:00
A more uncomfortable and ultimately more useful question
25:05
is this. What is AI revealing about who
25:08
we are and how we already operate?
25:12
That reframing changes everything?
25:14
Instead of asking how do we prevent AI bias, you ask what
25:19
human biases is AI making visible?
25:24
Instead of asking how do we maintain human override, you ask
25:29
what are we overriding and why? Instead of asking how do we
25:34
build trust in AI, you ask why did we trust our existing
25:39
processes when the evidence clearly suggests they were never
25:44
as objective as we believe them to be?
25:47
Let me outline 5 practices that leaders can begin implementing
25:51
to move from denial to productive engagement with what
25:55
AI reveals through that mirror. First, conduct an algorithmic
26:00
audit of values alignment. Take your organization's stated
26:04
values and compare them against what your AI systems have
26:08
learned from your actual behavior.
26:10
Where the algorithmics output diverges from your stated
26:14
values, you're looking at a gap between aspiration and reality.
26:20
That gap is not an AI problem, it's a leadership opportunity.
26:25
Second, create psychological safety around AI revealed
26:29
truths. When AI surfaces uncomfortable
26:32
patterns, the organizational response matters enormously.
26:36
If people are punished for what AI reveals, you're going to
26:40
guarantee deeper denial. If the revelations are treated
26:44
as organizational learning opportunities, you create the
26:47
conditions for genuine change. Third, address identity threat
26:53
directly. Stop pretending that AI anxiety
26:57
is simply a training problem. You have to acknowledge the
27:00
prophetic. For many professionals, AI
27:03
represents an existential challenge to their sense of
27:06
professional self. You have to create structured
27:09
conversations where people can express that threat without
27:13
being dismissed as resistant or told simply.
27:17
Get on board. 4th Prank the recursive loop.
27:21
Establish regular intervals where you examine what AI has
27:25
learned from your organization and ask whether those learned
27:29
patterns reflect the organization you want to be or
27:33
the unique organization that you've been.
27:35
This requires the kind of honest organizational self-assessment
27:39
that most organizations avoid because it's psychologically
27:43
uncomfortable. But avoiding it guarantees that
27:47
the bias feedback loop continues to compound. 5th, redesign
27:52
workflows before deploying AI. The McKenzie data is clear.
27:58
Organizations that redesign workflows around AI achieve
28:02
measurably better outcomes. But workflow redesign is not a
28:07
technical exercise. It is a psychological one.
28:11
It requires people to let go of processes that have defined
28:15
their professional identity, to accept that the way things have
28:19
always been may not be the way they should continue.
28:24
That is transformation work, and it requires the kind of human
28:28
centered approach that the human factor method of mine was built
28:32
to address. Let me bring this together.
28:36
AI is not forcing organizations to change who they are, it is
28:40
forcing them to see who they had been.
28:43
For leaders willing to look honestly, that reflection can be
28:46
truly transformative. For those who refuse, AI is
28:51
going to remain an external threat, something to resist,
28:54
regulate, and even blame. The technology here is not the
28:58
mirror, it is your light. And once something becomes
29:02
visible, it can't be on scene. Across the season, we've been
29:08
building a comprehensive picture of the human psychology of
29:12
transformation. The identity crisis.
29:15
The change triggers the emotional contagion that spreads
29:19
that crisis through organizations.
29:22
The structural trap, the crushes, the people in the
29:26
middle, and now the algorithmic mirror that makes all of it
29:30
visible at a scale and speed that organizational denial can't
29:35
contain. In our next episode, we're going
29:39
to explore what happens when all of these psychological dynamics
29:42
collide with something even more deeply embedded, organizational
29:46
culture. I'll be joined by James Elliott
29:50
from James Elliott and Company, who has been involved with his
29:53
clients with transformation for over 40 years.
29:56
Together, we're going to examine what I'm calling the
29:59
organizational immune system, the collection of cultural
30:03
antibodies that organizations develop over years of success
30:07
that attack anything perceived as foreign to the established
30:10
order. We're going to draw on Edgar
30:13
Shine's foundational work on organizational culture.
30:17
We're going to connect to what we explored in Season 1, Episode
30:21
11 on organizational drift. Because understanding why
30:25
organizations can intellectually grasp everything we've discussed
30:29
this season and still fail to change requires understanding
30:33
how culture operates below conscious awareness, how it
30:38
creates invisible resistance, then of strategy deck or AI
30:42
implementation can overcome without first being diagnosed.
30:47
If you have not yet subscribed to the Human Factor Podcast,
30:51
find us on Apple Podcast, Spotify, YouTube, Amazon Music,
30:55
or anywhere you watch or listen the podcast.
30:58
You can explore more at 2040 digital.com.
31:01
I encourage you to read the full newsletter that inspired today's
31:04
episode for the complete research and analysis.
31:08
Until next time, remember, the question is not whether AI
31:12
belongs in your organization. The question is whether you're
31:16
prepared to see what it reflects back and if you are, what you
31:20
are willing to do about what you see.
31:22
This is the Human Factor podcast.
31:25
I'm Kevin Novak, thanks for watching or listening.