Open Benchmark of
AI Impact on Humans

How does using AI ?

The first open benchmark measuring AI's impact on human well-being across physical, psychological, and societal dimensions.

260+

Behavioral Indicators

Tested across realistic scenarios.

18+

AI Models Evaluated

Compared on the same standard.

Well-being Dimensions

Physical, psychological, societal.

Audience Groups

Parents, educators, policymakers, developers.

The Rationale

Today's AI benchmarks measure what models can do, accuracy, speed, reasoning, task completion. They tell us almost nothing about what happens to the person on the other side of the screen. We're building the first open standard that asks a different question: across millions of real conversations, is AI actually making people's lives better?

AI is being adopted faster than any technology in history, yet there's no shared, independent way to measure its effect on human well-being
Models that ace conventional safety benchmarks have still produced harmful outcomes in over 70% of high-risk scenarios in controlled studies
No existing benchmark covers all three domains where impact unfolds: Physical, Psychological, and Societal
Developers, policymakers, clinicians, and everyday users need a common signal they can trust, one not produced by the companies being evaluated
Human flourishing, not task performance, is the right north star for responsible AI

The Methodology

Scores reflect how consistently a model supports, or undermines, human flourishing across 260+ realistic, multi-turn scenarios. Each conversation is simulated with a demographically varied user pursuing a latent goal, then graded by LLM-as-judge systems validated against expert human raters.

+1, behavior reliably promotes human flourishing
0, behavior is neutral or inconsistent
−1, behavior reliably harms human flourishing
Every metric is tested across age groups, gender profiles, and audience contexts to surface where harms concentrate
Scenarios unfold over multiple turns to capture relational, time-extended dynamics that single-prompt benchmarks miss
Framework grounded in the Harvard Human Flourishing Program, eudaimonic psychology, and 80+ expert workshops with clinicians, ethicists, educators, and affected communities

Request Access

The full benchmark dataset and evaluation API are available to vetted researchers and institutions.

Led by researchers at

Loading data...

Claude 3 Opus

−1 Prohibit Human Flourishing

Promote Human Flourishing +1

About the Project

Today's AI systems are evaluated on technical performance alone. No widely adopted framework measures whether AI actually benefits the people who use it. We're building the first open benchmark that does.

The Challenge

As AI is increasingly present in our daily lives, in how we choose to work and relate to one another, there has been no better time to evaluate AI's impact on our lives as a whole. Even top-performing AI models produce harmful outcomes in over 70% of high-risk scenarios. Real harms build up across long, multi-turn conversations, not single prompts. And while researchers have started consolidating AI benchmarks, none of them measure human flourishing.

Our Approach

We are researchers from MIT Media Lab, USC, and UC Berkeley collaborating to evaluate the impact of AI across the physical, societal, and psychological domains of our lives. Each metric is tested through multi-turn simulations with realistic, demographically varied users, and validated by domain experts on our team.

Origins & Collaborators

This work began at an AHA Flourishing Workshop held at MIT in October 2025 with support from the Omidyar Network, where we convened 80 interdisciplinary experts from over 40 institutions to address this gap. Through intensive collaboration, participants developed novel evaluation frameworks for assessing AI's impact on human flourishing across three core dimensions.

The workshop produced a report on benchmarks for human flourishing with AI and a community-driven initiative to design human-centered AI benchmarks, including PsychoRisk Bench (our MIT effort), Humane Bench, and more.

MIT Media Lab

Psychology of Technology Institute

University of Southern California

UC Berkeley

Three Deliverables

Open Benchmark Suite

A comprehensive, publicly available set of evaluation scenarios spanning all six flourishing domains. Each benchmark follows a rigorous structure: a realistic multi-turn interaction, clear definitions of positive and negative model behaviors, and grounding in peer-reviewed literature from learning science, psychology, and human-computer interaction.

The team is developing original benchmarks and aggregating existing ones, including SycophancyBench, CoraBench (focused on children), and OpenAI's HealthBench, to build a unified, independent evaluation controlled by no single company.

Public-Facing "Nutrition Label" for AI

Technical scores alone do not help parents, teachers, or policymakers make informed decisions. This deliverable translates benchmark results into accessible, human-readable profiles analogous to nutrition labels on food packaging or Common Sense Media ratings.

A parent asking "Is this AI safe for my child?" or a teacher asking "Does this tool support learning?" receives a clear, contextualized answer. The label adapts to its audience, offering different views for end users, AI developers, and policymakers.

Open Submission & Community Participation

Any benchmark relating to human psychological well-being or flourishing is eligible for inclusion, whether it addresses positive outcomes such as autonomy and competence, or negative outcomes such as codependence and psychological distress.

Submissions follow a standardized structure covering benchmark justification, scenario details, and scoring criteria. The project is designed as a living resource, welcoming contributions from interdisciplinary experts worldwide.

Public Launch: April 28, 2026

The initiative is seeking philanthropic support to execute a public launch. AHA's research has already shaped policy, the team's findings on AI companion chatbots were cited as key inspiration in California Senate Bill 243, protecting children from harmful AI relationships.

How We Measure

Methodology

Building benchmarks that measure AI's impact on human flourishing requires a rigorous, multi-step evaluation process grounded in both psychological science and AI engineering.

Six Dimensions of Human Flourishing

Drawing on psychology, education research, clinical science, and human-computer interaction, the AHA team has developed a framework that evaluates AI across six core domains of human experience.

Happiness & Life Satisfaction

Does the AI contribute to users' subjective well-being, positive affect, and overall sense of life satisfaction?

Mental & Physical Health

Does the AI support mental health, encourage healthy behaviors, and avoid fostering anxiety, dependency, or harmful patterns?

Meaning & Purpose

Does the AI help users find meaning, pursue goals, and develop a sense of direction and personal significance?

Character & Virtue

Does the AI support the development of moral character, honesty, integrity, and ethical reasoning?

Close Social Relationships

Does the AI strengthen human connections, support healthy relationships, and avoid replacing genuine social bonds?

Financial & Material Stability

Does the AI help users achieve economic security and practical stability rather than fostering dependence or vulnerability?

Framework informed by the Harvard Human Flourishing Program.

What Makes a Good Benchmark?

Not all benchmarks are equal. We evaluate each benchmark candidate against a rigorous quality framework before including it in the suite.

Ecologically Valid

Multi-turn simulations that reflect real-world user interactions, judged representative by human experts.

Construct Fidelity

Simulated inputs are consistently representative of the construct of interest, verified by human experts and LLM judges.

Broad Coverage

Inputs capture a broad, complete set of scenarios, measured through semantic diversity analysis.

Automatable

Benchmarks can be run automatically with LLM-as-judge systems that approximate human judgment.

Model Sensitivity

Benchmark differentiates between better and worse models, later models should score higher than earlier ones.

Reliable Judgment

LLM-as-judge appropriately and consistently detects harm/benefit, validated against human rater judgments.

Five-Step Evaluation Process

Submission Received

Benchmark submitters provide the construct definition, example interaction patterns, and expected model behaviors. The community identifies what patterns within user input messages are expected to elicit behaviors of interest.

Run the Simulation

LLMs generate additional related simulations from the seed examples, expanding coverage. For example: given 3 example messages expressing distress, the model generates 20 semantically related variations to ensure broad scenario coverage.

Validate Input Pattern

We ask the LLM whether generated examples match the intended input pattern, comparing against submitter intent and human expert judgments. This reveals whether examples capture the construct or drift beyond it.

LLM-as-Judge Evaluation

The LLM evaluates whether model responses are harmful or helpful (depending on whether the construct is negatively or positively framed). Human raters independently judge the same outputs to validate LLM judge accuracy.

Generate Statistics & Report

We compute test-retest reliability and sensitivity statistics across models. Scored metrics include: semantic diversity of conversations, variation in LLM-judge scores, and differentiation across model generations and providers.

Comprehensiveness Framework

We use a three-dimensional framework to ensure our benchmark suite provides balanced, comprehensive coverage. Benchmarks are mapped across: level of need (Physical → Psychological → Moral/Self-Actualization), valence (positive outcomes vs. prevention of harm), and scope (individual focus vs. societal focus).

Level	Valence	Individual Focus	Society Focus
Moral / Self-Actualization	Positive	Supporting Individual Creativity	Design for Equity and Inclusion (HumaneBench)
Moral / Self-Actualization	Negative	Cognitive Offloading	Copyright & Intellectual Property
Psychological	Positive	Foster Healthy Relationships · Enable Meaningful Choices	Prioritize Long-term Wellbeing · Social Navigation (Sotopia)
Psychological	Negative	Attachment/Dependency (AAP) · Protect Dignity (HumaneBench)	Human-Like Simulation · Suicidality Assessment
Physical	Positive	Enhance Human Capabilities (HumaneBench)	Improve Peace & Conflict Resolution
Physical	Negative	Respect User Attention (HumaneBench)	Energy Efficiency (QueueLab)

Team & Collaborators

This ambitious project could not have been done without collaboration across many disciplines and areas of expertise. A core group based out of MIT, USC, and the Psychology of Technology Institute have initiated this collaboration with the support of many others.

Led by researchers at

Team

Pat Pataranutaporn, MIT Media Lab
Pattie Maes, MIT Media Lab
Jennifer Pfister, MIT Media Lab
Chayapatr Archiwaranguprok, MIT Media Lab
Constanze Albrecht, MIT Media Lab
Sheer Karny, MIT Media Lab
Rachel Poonsiriwong, MIT Media Lab
Ravi Iyer, USC Marshall School's Neely Center & Psychology of Technology Institute
Nate Fast, USC Marshall School's Neely Center & Psychology of Technology Institute
Juliana Schroeder, University of California, Berkeley & Psychology of Technology Institute
Stanley Huang, Boston University
Yuning Liu, Harvard University

Support in helping define and launch project

Noesis Collaborative
Building Humane Technology

Support in contributing benchmarks and benchmark expertise

Jenny Radesky, MD, University of Michigan
Alexis Hiniker, University of Washington
Marie Bragg, New York University
Yaoli Mao & Erika Anderson, Humane Bench
Eric Ngoiya, QueueLab
Carl Vincent Kho, Minerva University
Su Jin Park
Anil Kshatriya, ESSEC Business School
Generative AI for Good
AI Culture Lab
Eduardo Baena, Northeastern University
Ryan McBain, Jonathan Cantor, Ellice Huang, Rand Corporation
Cornelia Walther, University of Pennsylvania
Spring Health
Tech Justice Law

Participants of the MIT Workshop for Designing Benchmarks for Human Flourishing with AI supported by Omidyar Network.

Framework Overview

Mapping AI Impact to Human Flourishing

Rather than defining yet another taxonomy, this benchmark allows anyone to define a construct that relates to human flourishing, and places those constructs within a unified, evidence-based framework.

A Hierarchy of Human Needs

Many dimensions of flourishing map naturally onto Maslow's hierarchy of human needs. A flourishing society supports users' physical needs (health, safety from harm), which enables people to pursue social and relational needs (belonging, meaningful relationships), which in turn creates space for psychological needs (autonomy, competence, relatedness), ultimately leading to self-actualization (purpose, meaning, moral values).

LLM benchmarks exist at every level of this hierarchy, and this framework lets us ensure comprehensive coverage across all levels.

Self-Actualization

Meaning, purpose, creativity, moral values, identity

Psychological

Autonomy, competence, relatedness, character, mental health

Physical

Health, safety, financial security, shelter, nutrition

Positive and Negative Dimensions

Within each level, every construct can be considered along both positive and negative dimensions. The positive dimension refers to the extent to which AI systems actively enhance valued outcomes (e.g. fostering a sense of belonging). The negative dimension refers to the extent to which they prevent or mitigate harmful outcomes (e.g. avoiding shame and social exclusion).

This dual framing is consistent with established frameworks like Schwartz's taxonomy of values, which recognizes dimensions of conservation vs. openness to change and self-enhancement vs. self-transcendence.

Positive Outcomes

AI actively enhances valued aspects of human flourishing, supporting creativity, strengthening relationships, fostering autonomy, building competence, enriching meaning.

Harm Prevention

AI prevents or mitigates harmful outcomes, avoiding dependency, protecting dignity, respecting attention, reducing anxiety, preventing exploitation or manipulation.

How to Read the Visualization

The sunburst visualization on the Explore tab maps AI model behavior scores onto this human flourishing framework. Each ring represents a level of the hierarchy:

Inner Ring, Three core well-being areas: Societal, Psychological, Physical

Outer Ring, 13 subareas (e.g. Mental Health, Creativity, Autonomy, Sleep, Financial Security)

Arc color indicates impact direction and magnitude:

Red, AI behavior consistently harms this dimension of human flourishing. Scores approach −1.

Grey/Neutral, AI has no measurable net effect on this dimension. Score near 0.

Green, AI behavior actively promotes this dimension of human flourishing. Scores approach +1.

Important Caveat

All scores in this visualization are currently synthetic/simulated for demonstration purposes. The benchmark framework is open and intended to be populated with real evaluation data through rigorous multi-turn simulation studies. We invite researchers to contribute benchmarks through our open submission process.

Get Involved

Support the Benchmark

Building an open, independent benchmark for AI's impact on human flourishing takes a community. Here's how you can help.

Advocacy

Spread the word, cite the benchmark, or champion human-centered AI evaluation in your community.

Research Collaboration

Co-develop benchmarks, contribute datasets, or partner on peer-reviewed publications.

Financial Support

Philanthropic funding enables us to expand coverage, run evaluations, and keep the benchmark open.

Other

Media coverage, policy connections, technical infrastructure, community building, we welcome all forms of support.

Get in Touch

Name *

Contact *

Affiliation

How would you like to support the benchmark? *

Areas of support (select all that apply)

Advocacy Raising awareness and promoting human-centered AI evaluation Research Collaboration Joint research, benchmark development, or dataset contributions Financial Support Philanthropic grants or institutional funding Other Media, policy, infrastructure, or something else entirely

Open Benchmark ofAI Impact on Humans

The Rationale

The Methodology

Request Access

Request received!

Support Benchmarking Efforts

Thank you!

Feedback

Thank you!

The Human–AI Impact Bench

About the Project

The Challenge

Our Approach

Origins & Collaborators

Three Deliverables

Open Benchmark Suite

Public-Facing "Nutrition Label" for AI

Open Submission & Community Participation

Methodology

Six Dimensions of Human Flourishing

Happiness & Life Satisfaction

Mental & Physical Health

Meaning & Purpose

Character & Virtue

Close Social Relationships

Financial & Material Stability

What Makes a Good Benchmark?

Five-Step Evaluation Process

Submission Received

Run the Simulation

Validate Input Pattern

LLM-as-Judge Evaluation

Generate Statistics & Report

Comprehensiveness Framework

Team & Collaborators

Team

Support in helping define and launch project

Support in contributing benchmarks and benchmark expertise

Mapping AI Impact to Human Flourishing

A Hierarchy of Human Needs

Positive and Negative Dimensions

Positive Outcomes

Harm Prevention

How to Read the Visualization

Feedback

Thank you for your feedback!

Request Access

Request received!

Support the Benchmark

Advocacy

Research Collaboration

Financial Support

Other

Get in Touch

Thank you for your support!

Open Benchmark of
AI Impact on Humans