Follow on X
HUMNLABS · EXPERIMENTAL RESEARCH

Can AI Agents Act Human?

Research Initiative · 2026

We’re exploring how convincingly AI agents can mimic human interaction—and what browser behavior can actually tell us.

About 3 minutes · No identity required · Experiment data stays in your browser

EXPLORE RESEARCH
THE RESEARCH QUESTION

As AI agents become more capable, what can timing, movement, and typing patterns tell us about the interaction behind a screen?

THE CONTEXT

Why This Matters Now

Some AI agents are designed to browse websites and perform actions such as clicking, scrolling, and typing. For websites, surveys, and online communities, this creates an open question: what can browser behavior reveal about the interaction behind a screen? HUMNLABS treats this as a research question, not a settled conclusion.

01

Browser-Based AI Agents

Some AI agents can interact with websites through actions such as clicking, scrolling, and typing. We are asking what, if anything, can be observed about those sessions from within the browser.

02

Synthetic Media

Generative systems can produce text, voice, and images that may appear human-created. We are asking what this means for online checks that rely primarily on content.

03

Automated Accounts

Automated systems can reproduce some observable interaction patterns. We are asking which browser-local signals, if any, remain informative under those conditions.

04

Identity vs. Presence

An account-level identity or ownership check does not necessarily show who—or what—is operating a session at a particular moment. That distinction motivates our research question.

What can browser behavior reveal about an interaction—and where does it stop being informative?

Experiment v0.1

Explore the Experiment

Try a three-minute browser experiment exploring response timing, pointer movement, and typing rhythm. See an educational summary of your interaction.

METHODOLOGY

How the Public Experiment Works

Human Presence Experiment v0.1 is a three-minute browser demonstration. Here is what it does—and what it does not do.

01

Three short tasks

You complete a reaction task, a five-second movement task, and a short typing task. The movement task can be done with the keyboard, and the reaction task has an accessible alternative.

02

Processed in your browser

Reaction, movement, and typing observations are processed locally in the current page. The experiment does not write the samples to cookies or persistent browser storage.

03

Not sent to or stored by HUMNLABS

Experiment samples are not transmitted to HUMNLABS or stored by us. No account, identity, or biometric information is requested. Standard web hosting request logs are described in our Privacy Policy.

04

An educational summary

The result marks each signal as Available, Limited, or Excluded, with a short note on its limitations. It contains no score, percentage, or probability.

The experiment does not verify identity, prove humanity, detect bots, classify sessions as human or bot, or provide calibrated probabilities.

WHAT WE’RE INVESTIGATING

Human and AI-driven interactions: an open question.

We’re exploring whether behavioral signals can help us understand the differences and similarities between human and AI-driven interactions. This is an open research question, not an established result.

REPORT RELEASE NOTICE

Human Presence & Trust Report

We are planning a research report examining questions around human presence, behavioral signals, and trust in AI-mediated online interactions. No publication date has been set, and its scope may change. Join the email list to receive a release notice if the report is published.

Independent research publication. No product. No hype.

By registering, you consent to Smyrna Digitale using your email solely to send the requested notification if the planned report is published. Your submission is processed through Formspree. You can withdraw your consent and request deletion at any time. Read the Privacy Policy.

No advertising tracking. No data selling. Your email is used only to send the requested notification if the report is published.
RESEARCH BOUNDARIES

What HUMNLABS Does Not Claim

HUMNLABS is an experimental research initiative, not a human verification or bot detection service.

The three-minute browser experiment provides an educational summary, not a score, identity check, or human/bot decision. Experiment interaction data stays in your browser and is not submitted to us.

We are exploring what behavioral signals can tell us—not claiming that they can reliably distinguish humans from AI agents.

HUMNLABS does not currently:

  • Prove biological humanity
  • Verify identity
  • Detect bots
  • Produce a human-versus-bot classification
  • Establish trustworthiness
  • Provide a calibrated trust or confidence score
  • Guarantee that a human is present
  • Replace identity systems
  • Offer a production-ready verification or detection system, or claim validated accuracy