How clicking "I am not a robot" could be training AI

Summary

---

Marketplace Morning Report: "How Clicking 'I am not a Robot' Could Be Training AI"

Date: May 11, 2026
Host: Sabri Ben-Ashur
Featured Guest: Panos Ipeirotis, Professor of Technology and Business at NYU Stern School

---

Episode Overview

This episode explores the often-invisible ways everyday digital behaviors—like clicking “I am not a robot”—are used to train artificial intelligence (AI) systems. The discussion focuses on the broader trend of companies leveraging user data and activity to improve AI models, the lack of transparency and compensation involved, and the implications for privacy and digital rights. The conversation centers on a recent revelation that Meta is installing software to monitor employees’ computer usage, ostensibly for AI development.

---

Key Discussion Points

1. Meta’s New Software and Employee Privacy (05:29-06:07)

• Meta has begun installing software on employees’ computers that tracks keystrokes and mouse movements.
• Official Purpose: Meta claims it’s for training AI, not for employee productivity monitoring.
• Controversy: Raised internal privacy concerns but highlights a common practice among tech companies.

“Meta…announced it’s installing software on employees computers that tracks their keystrokes and mouse movements…to help train its artificial intelligence. That’s apparently sparked some backlash within the company over privacy concerns, but it is far from the only controversial way companies have trained their AIs.”
—Sabri Ben-Ashur (05:29)

2. The Breadth of Data Sources Used in AI Training (06:07-07:06)

• Panos Ipeirotis places Meta’s move in context, noting that user and employee monitoring for various outcomes (including AI) has occurred for over 20 years.
• Similar practices: Early platforms like Upwork (formerly Autodesk) recorded employees’ screens and actions to build trust between parties.

“This thing exists for almost 20 years.”
—Panos Ipeirotis (06:16)

3. Everyday Actions as Unwitting AI Training (06:57-07:41)

• Captchas: Clicking items (“bicycles,” “buses”) to prove "I am not a robot" trains visual recognition AIs by labeling images.
• Reddit and Other Forums: User votes, discussions, and interactions are harvested for language model training without contributors’ knowledge or compensation.

“Recognizing different objects, you are providing labels to a confused AI system…So all this recaptcha is providing directly training data.”
—Panos Ipeirotis (07:06)

“When people participate on Reddit, people ask questions, there is a lot of discussion back there. People vote on the answers. This is very useful and extremely often leveraged by the large language models when they are trying to get knowledge about obscure topics.”
—Panos Ipeirotis (07:24)

4. Users Provide Value, But Don’t Share in It (07:41-08:25)

• AI companies are incentivized to find ever-new methods and sources for model training.
• Monetization Discrepancy: Platforms like Reddit profit from selling user-generated content to AI firms; users rarely receive any compensation or even awareness.

“You, as a Reddit participant, don’t get paid anything. The LLM companies, they train their models, they get value out of that. So the only person that doesn’t get value out of it is the people who contribute the data.”
—Panos Ipeirotis (08:10)

5. Legal and Privacy Frameworks—Or Lack Thereof (08:25-09:03)

• Most companies’ terms of service grant them extensive rights to use user data (often irrevocably).
• Users often can’t opt out or remove their data once it has been used to train a model.
• Missing Framework: There’s little regulatory clarity or protection for data contributors, especially when data is ingested into machine learning systems.

“You don’t have the right to opt out often, but this is literally just deleting your data from someone’s database. This doesn’t work well when we’re talking about trained models. We don’t really have much of a framework around such scenarios.”
—Panos Ipeirotis (08:48)

---

Notable Quotes & Memorable Moments

• On the hidden labor behind AI:
  “Recognizing different objects, you are providing labels to a confused AI system…all this recaptcha is providing directly training data.”
  —Panos Ipeirotis (07:06)

• On who profits from user-generated data:
  “The only person that doesn’t get value out of it is the people who contribute the data.”
  —Panos Ipeirotis (08:10)

• On regulatory gaps:
  “We don’t really have much of a framework around such scenarios.”
  —Panos Ipeirotis (08:52)

---

Timestamps for Key Segments

| Segment | Timestamp | |-------------------------------------------|--------------| | Meta’s employee monitoring for AI | 05:29–06:07 | | Early history of digital monitoring | 06:07–06:43 | | Everyday actions train AI (captchas, etc) | 06:57–07:41 | | Users’ role and lack of compensation | 07:41–08:25 | | Legal/regulatory landscape | 08:25–09:03 |

---

Tone & Language

• The tone was informative, slightly incredulous about the extent of “hidden” labor being provided by unwitting users.
• Panos Ipeirotis brought expert gravitas while expressing concern about user rights and regulatory gaps.
• The conversation was straightforward, accessible, and concerned with business, technology, and consumer fairness.

---

Summary Takeaway

The everyday digital activity of users—from clicking captchas to posting on forums—is an essential but invisible component in the development of modern AI. While companies and platforms profit and advance their technologies with this crowd-sourced "data labor," users themselves typically remain unaware and uncompensated, all within a landscape lacking meaningful legal protections or opt-out mechanisms. The episode offers an eye-opening glimpse into the unseen ways we all contribute to—and are affected by—the rise of artificial intelligence.

Transcript

[Marketplace Host (Sabri Ben-Ashur or Bridget)]

Marketplace podcasts are supported by Viking, committed to exploring the world in comfort. Journey through the heart of Europe on an elegant Viking longship with thoughtful service, destination focused dining and cultural enrichment on board and onshore. And every Viking voyage is all inclusive with no children and no casinos. Discover [e-mail] this Marketplace podcast is supported

[Freedom From Religion Foundation Representative]

by the Freedom From Religion Foundation. America is an aspirational idea, one we're still working to fulfill, shaped by Enlightenment ideals, reason, liberty and freedom of conscience, the belief that power comes from we the people, not a divinely appointed ruler. And while these ideals have not yet been fully perfected, they created something powerful, a framework that has expanded rights and freedoms over time. Now those freedoms and your rights are under attack. We're seeing growing efforts to blur the line between church and state, public funds to promote religion, 10 Commandments in classrooms, and pushing a version of America that leaves out too many people. As our Nation approaches its 250th anniversary, the Freedom Freedom From Religion foundation is working to protect the Constitution, defend secular government, and ensure that freedom continues to expand for everyone. Because America isn't just where we started, it's what we choose next. Go to FFRF US America or text America to 511-511-America to 511-51-511 text fees may apply.

[Sabri Ben-Ashur]

You know, it's kind of hard to just quit Gas from Marketplace I'm Sabri Ben ashore in New York, a gallon of gas will set you back $4.52 on average in the US $6.15 in California. A mere eternity ago, in February, before President Trump launched the war on Iran, gas was at just $2.98 a gallon. Energy Secretary Chris Wright over the weekend floated the idea of pausing the federal gas tax to give Americans some relief. As you can imagine, those prices are going to be harder on some people than others. That is what a study by the Federal Reserve bank of New York found. Upper income households barely cut back on gas at all. They just paid it. Lower income households, though, not only pulled back on gas, but they still ended up spending more money on gas overall because the prices are so high. Marketplaces Nova Sapphoff reports.

[Marketplace Reporter (Nova Sapphoff or Ryan)]

At a gas station down the block From Marketplace in LA, the price board read $6.98 and 910 for a gallon of regular. When Joanne Lee was filling up her Cadillac suv, it's doubled.

[Marketplace Host (Sabri Ben-Ashur or Bridget)]

Everyone feels it, right?

[Marketplace Reporter (Nova Sapphoff or Ryan)]

Lee says. One of her children commutes to school by bus and in the before times

[Marketplace Host (Sabri Ben-Ashur or Bridget)]

we used to go and pick him up more frequently. Probably are not doing that as much.

[Marketplace Reporter (Nova Sapphoff or Ryan)]

If gas prices stay high months from

[Marketplace Host (Sabri Ben-Ashur or Bridget)]

now, I think it will have implications on, you know, how much maybe we donate to the certain causes.

[Marketplace Reporter (Nova Sapphoff or Ryan)]

Lease cuts so far have been minor, but the New York Fed found that for lower income households, the cuts have been more dramatic. Gasoline use declined 7% in March. Higher income households cut back just 1%. It really does seem that the lower 20 or 30% of the income distribution

[Panos Ipeirotis]

is really feeling this most acutely.

[Marketplace Reporter (Nova Sapphoff or Ryan)]

Ted Rossman of Bankrate says that's consistent with the K shaped economy. Lower income households struggling, upper income prospering. But the New York Fed says the disparities are growing even wider this time than in 2022 when gas prices spiked because Russia invaded Ukraine. Benga Agilori is chief economist at the center for Budget and Policy priorities.

[Panos Ipeirotis]

So in 2022 this was right after the pandemic. So we had the expanded unemployment insurance

[Marketplace Reporter (Nova Sapphoff or Ryan)]

benefits, also expanded tax credits, rebate checks, higher savings, stronger wage gains. So we had people at the lower

[Panos Ipeirotis]

middle income of the distribution having more money so that they could weather these price shocks.

[Marketplace Reporter (Nova Sapphoff or Ryan)]

This time around, Agiloria expects more people to cut costs as gas prices remain elevated. I'm NovaSafo for Marketplace.

[Puerto Rico Investment Representative]

Reshoring to the US Is a strategic decision and where you land matters. In Puerto Rico, Companies like Amgen, AbbVie and Boston Scientific run advanced manufacturing and life science operations serving global markets. As a US Jurisdiction, Puerto Rico offers regulatory alignment, strong IP protections, seamless access to the mainland, competitive tax incentives and a deep bench of STEM trained bilingual talent that helps build resilient suppliers supply chains. See why industry leaders are already here? It's not what's next, it's where puerto Rico. Visit investpr.org Marketplace podcasts are supported by

[Marketplace Host (Sabri Ben-Ashur or Bridget)]

Viking, committed to exploring the world in comfort. Journey through the heart of Europe on an elegant Viking longship with thoughtful service, destination focused dining and cultural enrichment on board and onshore. And every Viking voyage is all inclusive with with no children and no casinos. Discover [e-mail] not too long ago, Meta

[Sabri Ben-Ashur]

announced it's installing software on employees computers that tracks their keystrokes and mouse movements. Not for like performance reviews, Meta says, but rather to help train its artificial intelligence. That's apparently sparked some backlash within the company over privacy concerns, but it is far from the only controversial way companies have trained their AIs. The game Pokemon Go used billions of images uploaded by users to train a navigation aid. AI Uber uses its drivers to train AI by uploading photos or voice clips. It does pay them to do that, though. Panos Epirotis is a professor of technology and business at NYU Stern School and is here to talk about Meta's plan and more. Good morning.

[Panos Ipeirotis]

Good morning.

[Sabri Ben-Ashur]

Is this just something freaky that Meta cooked up, or is this part of some larger trend on how we train our AI's?

[Panos Ipeirotis]

15 years back I was working with Autodesk, which is now called Upwork, which was allowing people to hire contractors anywhere in the world and to establish trust. One of the things that they were doing was they were recording the screen of the employees and the keystrokes and the mouse movements, et cetera. So this thing exists for almost 20 years.

[Sabri Ben-Ashur]

What's new about it, I guess, is the fact that it's done for purposes of AI training. So I just wonder if there are ever new ways that we are training AI and may not know about it.

[Panos Ipeirotis]

Well, this goes back 20 years back, you know. Do you remember all these captchas?

[Sabri Ben-Ashur]

Yes. Like click here to prove you're not a robot.

[Panos Ipeirotis]

Yeah. And you need to click on a bicycle, on a bus. Recognizing different objects, you are providing labels to a confused AI system that tries to figure out is there really a bus here? And so on. So all this recaptcha is providing directly training data, but also when people participate on Reddit, people ask questions, there is a lot of discussion back there. People vote on the answers. This is very useful and extremely often leveraged by the LLMs when they are trying to get knowledge about obscure topics.

[Sabri Ben-Ashur]

Are AI firms constantly looking for new and creative ways to train their models?

[Panos Ipeirotis]

100%. There is a general policy question, who gets the benefit of this data? For example, you go on Reddit, you participate in a discussion, and Reddit afterwards sells this data to LLM companies, top and AI, anthropic, Google, etc. It's actually one of the biggest sources of revenue. You, as a Reddit participant, don't get paid anything. The LLM companies, they train their models, they get value out of that. So the only person that doesn't get value out of it is the people who contribute the data.

[Sabri Ben-Ashur]

Are there any guardrails or laws that would regulate like the privacy aspect of this or the people kind of unwittingly training models?

[Panos Ipeirotis]

Part of this companies typically do have the right terms of service that give them the ability to mine the data, train on the data, and so on. You don't have the right to opt out often, but this is literally just deleting your data from someone's database. This doesn't work well when we're talking about trained models. We don't really have much of a framework around such scenarios.

[Sabri Ben-Ashur]

Panos Iperotis is a professor of technology and business at NYU Stern School. Thank you so much.

[Panos Ipeirotis]

Thank you.

[Sabri Ben-Ashur]

In New York, I'm Sabri Ben, ashore with the Marketplace morning Report. From apm, American Public Media.

[Marketplace Reporter (Nova Sapphoff or Ryan)]

There's so much happening in the world, and if you have particularly, shall we say, inquisitive kids, it can be hard to answer their questions. Hi, I'm Ryan.

[Marketplace Host (Sabri Ben-Ashur or Bridget)]

And I'm Bridget. And we host Million Bazillion, a podcast from Marketplace about money for kids and their families. We help your little ones think big about important but tricky topics like taxes, gas prices, and even what a cashless society might be like.

[Marketplace Reporter (Nova Sapphoff or Ryan)]

There's a bunch of new episodes out now, so go listen to Million Bazillion on your favorite podcast. Apparently.