AI at Work Changes More Than Productivity: New Research Links Human-AI Interaction to Role Identity

Imagine you are excellent at your job.
For ten years, people come to you because:
you know what to do.
Then artificial intelligence arrives.
Suddenly it can perform:
some of your analysis;
some writing;
some research;
some planning.
Your company celebrates.
Productivity increases.
There is just one psychological question nobody asked.
What happens to your understanding of your own value?
A peer-reviewed study published in Humanities and Social Sciences Communications on 9 September 2026 examined how human-AI interaction relates to employee task performance.
The researchers found that workplace human-AI interaction was positively associated with task performance, but the relationship was partly explained by two psychological factors:
role identity
and:
self-efficacy. �
Nature
That finding matters because organisations often treat AI adoption as:
a technology problem.
Psychology says:
not so fast.
What Is Role Identity?
Role identity is, broadly:
who I understand myself to be within this work.
A nurse may think:
I care for patients.
An engineer:
I solve technical problems.
A journalist:
I investigate and explain.
A creator:
I produce original work.
These aren’t merely:
job descriptions.
They become:
identity.
Then AI Enters the Role
Suppose AI begins performing:
half the tasks associated with that identity.
Now the employee may ask:
What exactly am I here for?
This does not automatically produce insecurity.
It can also create:
growth.
But the transition needs psychological sense-making.
The New Study Identified Three Dimensions
The researchers conceptualised workplace human-AI interaction around three dimensions:
anthropomorphic tool, adaptive trust and unidirectional emotional connection. �
Nature
That wording is fascinating.
Because it acknowledges that humans do not relate to AI only as:
software.
We may interact with it as:
tool;
adaptive collaborator;
social-like presence.
Different psychological layers can coexist.
Anthropomorphic Tool
The AI remains:
a tool.
But behaves in ways that resemble:
human conversation;
reasoning;
adaptation.
That can make interaction easier.
It can also create:
over-attribution.
The system sounds:
confident.
Therefore:
we assume competence.
Not always wise.
Adaptive Trust
This may be more important.
Good human-AI collaboration should not mean:
trust AI.
It should mean:
learn when AI deserves trust.
You use the system.
Notice:
where it succeeds.
Where it fails.
Your trust becomes:
calibrated.
That is adaptive.
Emotional Connection
The researchers also identify a one-way emotional component in human-AI interaction. �
Nature
That does not require assuming the AI itself has human emotions.
The relevant psychological event occurs in:
the employee.
An assistant can feel:
supportive;
frustrating;
reassuring;
annoying.
That changes the work experience.
Humans Already Have Emotional Relationships With Tools
Ask anyone whose laptop crashed ten minutes before a deadline.
🤣
But conversational technology intensifies this because:
the tool replies.
It has a style.
It reacts.
That makes human-machine interaction:
socially richer.
Now Add Self-Efficacy
Self-efficacy means:
I believe I can successfully perform this task.
That belief matters enormously.
People with stronger self-efficacy often:
persist longer;
approach difficult work differently;
recover more effectively after setbacks.
AI can increase self-efficacy.
Or undermine it.
AI Can Make Someone Feel More Capable
Imagine employee thinking:
“I’ve never analysed this kind of dataset.”
AI helps:
structure;
explain;
suggest.
Employee completes the task.
Next time:
confidence increases.
That is:
augmentation.
But AI Can Also Produce the Opposite Feeling
Employee:
“I used to be good at this.”
AI performs it instantly.
Employee begins thinking:
“Perhaps the machine is the competent one.”
Now productivity may rise
while:
self-efficacy falls.
That is a dangerous organisational trade.
Companies Need Both Performance and Human Capability
Imagine productivity increases:
20%.
Wonderful.
But employees gradually become:
less confident;
less independent;
less willing to make decisions without AI.
You may have built:
short-term efficiency
and:
long-term fragility.
The Research Suggests Identity Matters
The study found role identity and self-efficacy partially mediated the positive relationship between human-AI interaction and task performance. �
Nature
In plain language:
how employees understand:
their role
and:
their own capability
appears connected to whether human-AI interaction translates into stronger performance.
That gives leaders a practical clue.
Do Not Introduce AI Without Redefining the Human Role
Bad rollout:
AI now writes your reports.
Employee:
“So what do I do?”
Management:
“Be more strategic.”
What does that mean?
🤣
Better rollout:
AI drafts standard report sections.
Human responsibilities shift toward:
verification;
interpretation;
client judgement;
decision-making.
Now the human role remains visible.
AI Should Remove Tasks, Not Meaning
This may become one of the most important leadership principles of the AI era.
Every role contains:
tasks
and:
meaning.
They are not identical.
A therapist writes notes.
But being therapist does not mean:
note writer.
A teacher marks homework.
But being teacher does not mean:
marker.
AI may remove:
task.
Leadership must preserve or redesign:
meaning.
Work Identity Can Expand
Imagine accountant once defines role as:
“produce financial reports.”
AI does much of that.
New identity:
“I help leaders understand financial reality.”
That is a broader role.
And potentially:
more valuable.
This Requires Upskilling
You cannot tell someone:
be strategic
without helping them develop:
strategic skills.
If AI automates the lower layer, employees need opportunities to move:
up the value chain.
Otherwise automation simply creates:
role ambiguity.
Role Ambiguity Creates Anxiety
Human beings like understanding:
what success looks like.
If AI changes the work faster than organisations change:
expectations;
promotion criteria;
responsibilities,
employees become uncertain.
What am I evaluated on?
What does excellent work look like now?
Who owns the decision:
me
or:
AI?
Leaders Need to Answer Those Questions
Explicitly.
Not through motivational posters about:
“embracing transformation.”
🤣
Human-AI Performance Is a System
We should stop measuring:
AI productivity
as though the human is irrelevant.
The unit of analysis increasingly needs to become:
human + AI + workflow + organisation.
A fantastic model inside a terrible workflow can reduce performance.
A good model inside:
clear roles;
trust;
training;
feedback
may improve it significantly.
Trust Should Be Calibrated Through Experience
Employees need:
safe opportunities
to discover AI limitations.
If the organisation presents the system as:
magical,
workers either:
overtrust
or:
quietly distrust.
Neither is good.
Show Failure Cases
During training:
Here is where AI performed well.
Here is where it hallucinated.
Here is where human judgement corrected it.
That teaches:
appropriate reliance.
AI Literacy Is Partly Psychological
AI literacy is not only:
how prompting works.
It includes:
When should I trust this?
When should I verify?
How do I notice when I am outsourcing too much judgement?
That is metacognition.
The Study’s Emotional Dimension Is Especially Interesting
Work software traditionally aims to be:
functional.
Conversational AI may create:
a felt interaction.
Suppose employee experiences AI as:
encouraging.
The system:
helps;
explains;
does not punish mistakes.
That may change:
confidence.
AI Could Become a Confidence Scaffold
Beginner:
asks question.
AI explains privately.
No embarrassment.
Employee practises.
Then contributes in meeting.
Useful.
This could particularly help people who hesitate to ask:
basic questions
in public.
But There Is a Risk
If all confidence comes from:
the AI being present,
what happens without it?
Strong scaffolding eventually needs:
removal.
Education understands this.
Training wheels help.
At some point:
you ride.
We Need AI That Builds Human Capability
Not merely:
performs around it.
Imagine two assistants.
Assistant A:
always gives answer.
Assistant B:
sometimes asks:
“What do you think?”
Then provides:
feedback.
B may feel slower.
But it can build:
competence.
Businesses Need to Decide Which Objective They Want
Immediate output?
Use execution mode.
Employee development?
Use teaching mode.
Both are valid.
Different goals.
AI Could Have Modes
Do it for me.
Do it with me.
Teach me to do it.
That would be psychologically intelligent design.
Not Every Task Deserves Development
Employee:
“Please alphabetise these filenames.”
AI:
“What do you think the first step is?”
No.
🤣
Automate trivial work.
Preserve learning where:
judgement matters.
Human Roles May Become More Relational
As machines become better at:
information production,
humans may become more valuable in:
trust;
leadership;
negotiation;
ethics;
relationship;
judgement.
That could reshape professional identity.
Some Workers May Welcome This
Less:
admin.
More:
meaningful work.
Wonderful.
Others may discover that the automated tasks were actually the parts they:
enjoyed;
felt competent doing.
Different people need different transitions.
This Is Why AI Adoption Cannot Be One-Size-Fits-All
A senior expert and graduate employee may use the same system very differently.
Senior expert:
delegates execution.
Graduate:
still needs execution experience in order to build expertise.
Automating all junior work creates another problem:
Where do future experts come from?
This Is a Serious Future-of-Work Question
Senior lawyer gained expertise by:
reviewing documents.
Junior lawyer’s AI reviews them.
How does junior develop:
pattern recognition?
The profession must redesign:
learning pathways.
AI Can Compress Apprenticeship—but Should Not Erase It
We need deliberate opportunities for:
practice;
feedback;
exposure.
Otherwise organisations may enjoy:
productive juniors
and later discover:
nobody developed into senior experts.
The Research Has Limitations
The authors used workplace surveys and interviews rather than controlled causal experiments. �
Nature
Therefore:
association does not automatically prove causation.
People who interact effectively with AI may also differ in other ways.
The study nevertheless provides a useful framework and validated measurement scale for studying human-AI interaction in workplaces.
The Framework Is Valuable Because Work Is Becoming Socially Hybrid
We increasingly collaborate with:
humans
and:
machines
during the same task.
That changes:
trust;
identity;
confidence.
Psychology belongs inside the deployment conversation.
Final Thoughts
Companies will spend enormous amounts of money trying to answer:
How much productivity does AI create?
They should also ask:
What kind of worker does this workflow create?
More capable?
More confident?
More dependent?
More uncertain about their role?
Those outcomes matter.
The new research suggests AI performance does not occur independently of human identity.
People work better when they still understand:
who they are;
what they contribute;
what they can do.
That means the strongest AI workplace may not be the one where machines perform the largest percentage of tasks.
It may be the one where:
the machine does what it does best,
while the human becomes increasingly clear about:
what remains meaningfully theirs.
Research note: Wu and colleagues published Dancing with AI: how human-AI interaction affects employee task performance in Humanities and Social Sciences Communications on 9 September 2026. The research conceptualises human-AI interaction through anthropomorphic-tool, adaptive-trust and unidirectional-emotional-connection dimensions, and reports a positive association with task performance partly mediated by role identity and self-efficacy. Because the work relies on surveys and interviews, causal claims should be made cautiously. �
Nature
Read the peer-reviewed study⁠�