A data-driven study of 894 occupations using ONET 30.3 to identify the jobs where AI assistance and human expertise work best together.
The most interesting question about AI and employment may not be which jobs AI will replace.
A better question is: Which jobs are likely to benefit most from working with AI, while still relying heavily on human expertise?
That question sits at the heart of the Aiwork AI Collaboration Index.
Instead of looking at AI exposure as a measure of automation risk, this study takes a different approach. It looks at three parts of an occupation at the same time:
- how much of the work involves activities that AI could potentially support;
- how much the job relies on human judgment, expertise, and problem-solving; and
- how much it depends on collaboration, interpersonal interaction and decisions that require human involvement.
Using data from O*NET 30.3, the team at Aiwork, an AI workspace platform for businesses and investors, analyzed 894 occupations with complete data across these three dimensions. The goal was to see which occupations have the strongest combination of work that AI could assist with and responsibilities that still depend heavily on people.
The results are not a prediction of which jobs AI can perform, and a high score does not mean an occupation is protected from automation. Instead, the index highlights jobs where AI assistance and human expertise may complement each other particularly well.
In other words, this is less about asking whether AI can replace the person, and more about asking what happens when the person has AI on their side.
Let’s dive in.
TL;DR – Key Takeaway
- Eight of the top 10 occupations are healthcare professions.
- High AI potential does not necessarily mean a job is easy to replace.
- AI collaboration is not limited to technical or healthcare roles
- Management has the highest average AI Collaboration Index at 76.8.
- The findings suggest that AI and human expertise can often work together rather than simply compete.
Recommended For You: AI Job Exposure: Which Jobs Are Most Exposed to AI
What the AI Collaboration Index measures
The AI Collaboration Index is a proprietary measure developed by Aiwork to evaluate which occupations may be best positioned to benefit from working with AI.
It is important to clarify that the index is not an existing O*NET metric. O*NET provides the underlying occupational data, while Aiwork developed the framework and scoring method used in this study.
O*NET contains detailed information about hundreds of occupations, covering areas such as work activities, skills, abilities, tasks, and work context. The data come from several sources, including workers, occupational experts, and analysts.
For this study, we used O*NET 30.3 as secondary data and built an index around three dimensions that matter when people and AI work together:
- AI-Compatible Work: How much of an occupation involves activities where AI could potentially help with information-intensive work.
- Human Judgment & Expertise: How much the job depends on knowledge, skills, abilities, problem-solving, and professional judgment that remain important to the work.
- Human Collaboration: How much the job involves working with other people, communication, interpersonal interaction, and decisions that carry human responsibility.
The research team combines these three dimensions into a single score, making it possible to compare and rank occupations based on their relative potential for human-AI collaboration.
A high score does not mean that AI can do the job, or that the occupation is protected from automation. It simply means that the occupation has a particular combination of AI-assistable work and continued human involvement.
1. AI-Compatible Work
This dimension captures the extent to which an occupation involves information-intensive, analytical, decision-support, planning, documentation, and knowledge-processing activities.
We selected 12 O*NET Work Activities:
- Getting Information
- Processing Information
- Analyzing Data or Information
- Making Decisions and Solving Problems
- Thinking Creatively
- Updating and Using Relevant Knowledge
- Developing Objectives and Strategies
- Organizing, Planning, and Prioritizing Work
- Working with Computers
- Documenting/Recording Information
- Interpreting the Meaning of Information for Others
- Providing Consultation and Advice to Others
O*NET defines Work Activities as general types of job behavior that occur across multiple occupations. Its 30.3 Work Activities file contains 73,308 occupation-level records across 894 occupations. (O*NET Resource Center)
Why these activities matter: each describes work involving information, knowledge, analysis, planning, interpretation, or structured decision processes. Those are areas where AI systems can plausibly assist, although O*NET itself does not establish that a particular AI system can perform them.
That caveat is fundamental.
For example, “Making Decisions and Solving Problems” should not be interpreted as “AI can make the decisions.” Instead, it identifies an occupational activity in which AI might provide information retrieval, analysis, scenario comparison or decision support while a human retains responsibility for the decision.
2. Human Judgment & Expertise
The second dimension captures capabilities that can make human oversight, interpretation, and professional expertise particularly important.
This dimension combines:
Seven O*NET Transferable Skills
- Complex Problem Solving
- Operations Analysis
- Technology Design
- Social Perceptiveness
- Judgment and Decision Making
- Systems Analysis
- Systems Evaluation
and eight O*NET Abilities
- Deductive Reasoning
- Inductive Reasoning
- Problem Sensitivity
- Originality
- Fluency of Ideas
- Information Ordering
- Written Comprehension
- Oral Comprehension
One thing to note: The O*NET 30.3 file we used in the analysis is the Transferable Skills dataset, and not the generic Skills dataset. The 30.3 release includes 25 transferable-skill elements covering 894 occupations.
Transferable skills describe capabilities that can be useful across different types of work, including problem-solving, communication, and working with information.
Our analysis also includes ONET Abilities. ONET treats abilities as relatively enduring capacities that influence how people perform work.
Why combine skills and abilities?
We combined both skills and abilities together to give the index a broader view of what human expertise looks like within an occupation since they capture different sides of the same picture.
Skills describe capabilities used in performing work, while abilities reflect underlying capacities that support that performance.
Using both allows the index to identify occupations where AI may have plenty of room to assist, but where human reasoning, interpretation, and expertise remain an important part of getting the work done.
3. Human Collaboration
The third dimension measures the social and decision-making context surrounding the work.
We use nine O*NET Work Context variables:
- Face-to-Face Discussions with Individuals and Within Teams
- Contact With Others
- Work With or Contribute to a Work Group or Team
- Deal With External Customers or the Public in General
- Coordinate or Lead Others in Accomplishing Work Activities
- Frequency of Decision Making
- Freedom to Make Decisions
- Impact of Decisions on Co-workers or Company Results
- Determine Tasks, Priorities and Goals
O*NET defines Work Context as the physical and social factors that influence the nature of work. The 30.3 Work Context dataset contains 297,676 records covering 57 context elements and 894 occupations.
Why does this dimension matter?
Because two occupations can contain similarly high levels of information processing but have completely different human requirements.
A programmer and a physician, for example, may both process information, solve problems, and use computers. But the surrounding work context differs substantially.
The physician may need to:
- communicate with patients;
- coordinate with other professionals;
- make consequential decisions;
- interpret ambiguous information; and
- take responsibility for outcomes.
The collaboration index is designed to capture that difference.
How the scoring works
The O*NET variables used in the study are measured on different scales.
For example:
- Importance: 1 to 5
- Level: 0 to 7
- Context: 1 to 5
O*NET itself standardizes Importance and Level ratings to a 0–100 scale using:
S = ((O − L) / (H − L)) × 100
where O is the observed rating and L and H are the minimum and maximum values of the relevant scale. (O*NET OnLine)
Aiwork follows that standardization logic.
For each selected descriptor:
- Importance is converted to 0–100.
- Level is converted to 0–100.
- The two standardized values are averaged for each selected Work Activity, Transferable Skill or Ability.
- The relevant descriptors are then averaged to produce the dimension’s raw occupation-level score.
For Work Context, the selected CX ratings use the 1–5 Context scale and are standardized to 0–100.
This produces three raw scores for every occupation.
Why not simply average the original O*NET scores?
Because the O*NET measures do not all use the same scale. Importance is rated from 1 to 5, while Level is rated from 0 to 7, so taking the raw numbers and averaging them would give the two measures different meanings within the same calculation.
To make the measures comparable, the research team converts each rating to a common 0–100 scale before combining them. This follows the same basic standardization approach O*NET uses when presenting its own standardized scores.
This step does not change the underlying O*NET ratings. It simply puts measures that start on different scales onto a common basis, making it more reasonable to combine them in the index.
Why the final index uses percentiles
This is one of the key methodological choices in the study.
The three dimensions do not have identical statistical distributions, so combining their raw scores directly could cause differences in scale and distribution to influence the final ranking. To make the three measures more comparable, the research team converts each dimension into a percentile rank across the 894 occupations included in the analysis.
The final score is then calculated by taking the average of those three percentile ranks:
AI Collaboration Index = (AI-Compatible Work Percentile + Human Judgment & Expertise Percentile + Human Collaboration Percentile) / 3
This gives each dimension the same weight. No single dimension is deliberately given more influence than the others.
The percentile approach also makes the index easier to interpret as a relative measure. An occupation with a score of 98.7 is not being assigned a 98.7% chance of benefiting from AI. Rather, its average position across the three dimensions is at the 98.7th percentile within the study’s 894-occupation universe.
That distinction is important. The index is a ranking of relative structural potential for human-AI collaboration, not a probability, forecast, or measure of automation risk.
The 10 Occupations With the Highest AI Collaboration Scores
Our study finds that the highest-ranking occupations combine strong potential for AI assistance with skills that still rely heavily on human expertise and interaction. Here are the top 10.
| Rank | Occupation | AI-Compatible Work | Human Judgment | Collaboration | Index |
| 1 | Neurologists | 99.8 | 97.5 | 98.9 | 98.7 |
| 2 | Emergency Medicine Physicians | 95.2 | 96.9 | 99.0 | 97.1 |
| 3 | Chief Executives | 92.1 | 100.0 | 97.1 | 96.4 |
| 4 | Ophthalmologists, Except Pediatric | 97.9 | 90.9 | 99.6 | 96.1 |
| 5 | Hospitalists | 96.6 | 94.0 | 96.8 | 95.8 |
| 6 | Obstetricians and Gynecologists | 95.0 | 92.2 | 99.7 | 95.6 |
| 7 | Sports Medicine Physicians | 93.7 | 94.9 | 97.3 | 95.3 |
| 8 | Preventive Medicine Physicians | 96.9 | 99.8 | 89.1 | 95.2 |
| 9 | Urologists | 99.9 | 86.7 | 97.7 | 94.8 |
| 10 | Physical Medicine and Rehabilitation Physicians | 92.3 | 91.3 | 99.2 | 94.3 |
Free Download: AI Collaboration Index Full Ranking of 894 Occupations
Finding 1: Healthcare dominates the top of the index
Eight of the 10 highest-ranking occupations in our study are healthcare professions. That makes sense when you look at what the index is actually measuring.
Take neurologists, which rank first overall. The occupation scores:
- 99.8 for AI-Compatible Work
- 97.5 for Human Judgment & Expertise
- 98.9 for Human Collaboration.
Averaged together, those scores produce an overall AI Collaboration Index score of 98.7.
The reason neurology ranks so highly is the combination of these three factors. Neurologists work with large amounts of clinical information, but their work also requires interpretation, complex reasoning, and direct engagement with patients. AI may be able to help with parts of the information-heavy workload, but the role still depends heavily on clinical expertise and human judgment.
The O*NET task data provides useful context for this result.
Neurologists’ high-importance tasks include examining patients, discussing symptoms and medical histories with patients, and ordering or interpreting laboratory results. These tasks involve more than processing information. They require gathering information, making sense of it, and applying professional judgment in a human setting.
Key takeaway: Neurology ranks highly because it combines substantial opportunities for AI assistance with work that continues to rely heavily on human expertise, judgment, and patient interaction.
Finding 2: High AI Potential Does Not Mean Low Human Involvement
Our study found that just because a job has high AI exposure does not mean AI can easily replace the person doing it.
Take Emergency Medicine Physicians, who rank second in the index. Their scores are:
- AI-Compatible Work: 95.2
- Human Judgment & Expertise: 96.9
- Human Collaboration: 99.0
Together, these produce an overall score of 97.1.
What stands out is the exceptionally high collaboration score. Emergency physicians work with large amounts of clinical information, but they also have to interpret that information, communicate with patients and colleagues, and make decisions in situations where the consequences can be significant.
This is where the distinction between AI assistance and AI replacement becomes important.
Some parts of the work are well suited to computational support. AI could potentially help a physician retrieve information, summarize a patient’s history, organize clinical data, identify unusual results or compare possible diagnoses.
But those capabilities do not remove the need for a physician to assess the situation, communicate with the patient, weigh the available evidence, and take responsibility for the decision.
Key Takeaway: an occupation can have substantial potential for AI assistance while still being deeply dependent on human interaction and judgment. In some professions, those human elements are not peripheral to the work. They are the work.
Finding 3: AI Collaboration is not Limited to Technical or Healthcare Jobs
Chief executives rank third in the index, showing that the strongest potential for AI collaboration can also be found in roles built around strategy, leadership, and decision-making.
- AI-Compatible Work: 92.1
- Human Judgment & Expertise: 100.0
- Human Collaboration: 97.1
- Overall Index: 96.4
Their work involves processing large amounts of information, but it also requires people skills, judgment and responsibility that cannot simply be separated from the role.
According to their O*NET task profile, Chief executives are expected to direct and implement organizational policies and objectives, review reports from staff, and analyze operations to identify areas for improvement. Much of this work involves gathering, organizing, and interpreting information before deciding what the organization should do.
That creates several opportunities for AI to support the role.
AI could help an executive research markets and competitors, summarize internal reports, analyze financial or operational data, compare different scenarios, or bring together information needed for a strategic decision.
But there is an important difference between helping with the analysis and making the decision. The executive still has to decide which information matters, weigh competing priorities, consider the impact on the organization and its people, and take responsibility for the outcome.
The high score for Human Judgment & Expertise reflects this part of the role.
Chief executives scored 100.0 on that dimension, the highest possible percentile in the study, while also scoring 97.1 for Human Collaboration. This combination helps explain why the occupation ranks so highly despite not being a technical or healthcare profession.
Finding 4: Management has the highest average collaboration index
At the major occupational-group level, Management ranks first, with a mean AI Collaboration Index of approximately 76.8.
Healthcare Practitioners and Technical occupations rank second at approximately 72.9, followed closely by Community and Social Service occupations at approximately 72.7.
| Major occupational group | Mean AI Collaboration Index |
| Management | 76.8 |
| Healthcare Practitioners & Technical | 72.9 |
| Community & Social Service | 72.7 |
| Architecture & Engineering | 65.8 |
| Life, Physical & Social Science | 64.2 |
| Computer & Mathematical | 63.6 |
| Business & Financial Operations | 63.3 |
| Education, Training & Library | 63.1 |
| Legal | 62.4 |
| Protective Service | 60.0 |
Management scores highly because the group combines substantial information and decision work with human coordination and consequential decision-making.
What the Index Tells Us About the Future of Professional Work
The way we talk about AI and jobs is often too simple. A job is either described as something AI will replace or something AI cannot do. In reality, most professional work is likely to be more complicated than that.
AI can take over some tasks, support others, and have little impact on the parts of a job that depend on experience, judgment, or working with people. The more useful question is therefore not simply whether a job is “automatable,” but how AI might fit into the way that work is actually done.
For this study, it helps to think about three broad forms of AI involvement:
- AI substitution: AI takes over a significant part of the work, with relatively little human involvement required.
- AI assistance: AI handles specific parts of the workflow, while a person reviews the output, applies their expertise, and decides what to do with it.
- Human-led AI collaboration: AI becomes embedded in a professional workflow where humans retain substantial responsibility for judgment, coordination, communication, and decisions.
The AI Collaboration Index is designed to identify occupations that structurally resemble the third category. It does not predict which category a specific occupation will ultimately fall into.
So the index should be viewed as a starting point for understanding where human-AI collaboration may have strong potential, rather than a forecast of which jobs will or will not be automated.
What the AI Collaboration Index Means for Businesses Adopting AI
For businesses, the findings suggest that AI adoption should not start with a simple question of which jobs can be automated. A better starting point is to understand how AI could fit into the workflows people already use.
An occupational workflow audit can help organizations identify where AI is most likely to add value without removing the expertise that makes the role effective.
For each important role, businesses can ask:
Which information-heavy tasks take up the most time?
Look at activities such as research, documentation, reporting, data analysis and information retrieval. These are often the first areas worth examining for potential AI support.
Which parts of the job depend on human judgment?
These areas may be better suited to AI decision support than full automation. AI can help organize information or highlight patterns, while the professional remains responsible for interpreting the results and making the decision.
Which tasks depend on working with people?
Jobs involving patients, customers, colleagues, clients, or teams may benefit from AI working in the background rather than trying to replace human interaction itself.
Where could an AI error have serious consequences?
The higher the cost of getting something wrong, the more important human review, verification and accountability become.
Where can AI remove administrative work?
Reducing repetitive documentation, information gathering and other time-consuming tasks may offer a practical way to improve productivity while allowing professionals to focus on higher-value work.
The goal is not to add AI for the sake of using AI. It is to find the parts of a workflow where the technology can make people more effective without taking away the human responsibility the job requires.
What the Findings Mean for Professionals
The findings also suggest that AI literacy needs to go beyond knowing how to use a chatbot or write a good prompt.
Professionals increasingly need to understand how AI fits into their particular field and workflow. That includes knowing:
- which tasks AI can realistically assist with;
- how to check and evaluate AI-generated information;
- when human review is necessary;
- how to combine AI output with professional knowledge;
- how to use AI without losing sight of the wider context; and
- who remains responsible for the final decision or outcome.
The advantage may increasingly belong to professionals who can use AI effectively within their area of expertise, rather than simply those who know how to use AI tools.
The important skill is not just knowing what AI can do. It is knowing when to use it, how much to rely on it, and when human judgment needs to take over.
AI Collaboration Index: Final Thoughts
The AI Collaboration Index 2026 – 2027 offers a different way to look at AI and work. Instead of asking which jobs AI will replace, it looks at where AI assistance and human expertise can work together.
Neurologists, emergency physicians and chief executives top the ranking because their work combines information-intensive tasks with significant human judgment, expertise and interaction.
These occupations are not necessarily protected from automation. Rather, they show how AI can support parts of a job while people remain responsible for the decisions that matter.
The future of work may be less about AI replacing people and more about understanding where AI can help people do their jobs better.
The Research Methodology
Dataset: O*NET 30.3 Database, May 2026. (O*NET Resource Center)
Analytical universe: 894 occupations with complete data across the three index dimensions.
Primary datasets:
- Work Activities: 73,308 records, 894 occupations. (O*NET Resource Center)
- Transferable Skills: 44,700 records, 894 occupations. (O*NET Resource Center)
- Abilities: 92,976 records. (O*NET Resource Center)
- Work Context: 297,676 records, 894 occupations. (O*NET Resource Center)
- Task Ratings: 161,559 records, used for validation rather than index scoring. (O*NET Resource Center)
Dimensions: AI-Compatible Work, Human Judgment & Expertise, Human Collaboration.
Normalization: ONET Importance and Level ratings standardized to 0–100; Context ratings standardized from the 1–5 CX scale. ONET documents the same 0–100 standardization framework for its Importance and Level scores. (O*NET OnLine)
Aggregation: Each dimension is converted to an occupation-level percentile, then equally weighted.
Formula:
AI Collaboration Index = (AI-Compatible Work Percentile + Human Judgment & Expertise Percentile + Human Collaboration Percentile) ÷ 3
Sensitivity: Alternative 40% weighting models retained 95%, 95%, and 94% of the equal-weight top 100, respectively.
Interpretation: Higher scores indicate greater structural potential for human-AI collaboration relative to other occupations in the study, not a probability of automation, AI adoption, or productivity gain.
Data attribution: ONET 30.3 Database, National Center for ONET Development, sponsored by the U.S. Department of Labor, Employment and Training Administration. O*NET’s database is available under a Creative Commons Attribution 4.0 International license, with attribution and change-notification requirements. (O*NET Resource Center)
