404 Media Podcast: What It’s Like to Be a Data Labeler Training AI

Date: February 16, 2026
Host: Jason Kebler (404 Media)
Guest: Michael Jeffrey Asia, General Secretary of the Data Labelers Association, Kenya

Episode Overview

This episode provides a rare, in-depth look at the hidden labor powering artificial intelligence systems. Host Jason Kebler interviews Michael Jeffrey Asia, a veteran Kenyan data labeler and the General Secretary of the Data Labelers Association. Michael draws on his personal and collective experience to illuminate the emotional, psychological, and economic costs of data labeling—particularly in Kenya, a global hub for such work. The conversation covers the grueling realities of annotating everything from self-driving car images to pornography and pretending to be AI companions, the lack of worker protections, and efforts to organize data labelers globally for better conditions.

Key Discussion Points & Insights

1. What Is Data Labeling? (05:30)

• Michael explains data labeling as the process of teaching AI to recognize and categorize information, using self-driving cars as an example.
  • Quote (Michael Asia, 05:30):

    "This vehicle has to be taught a lot of things ... It has to be given this information like a small child."

• Data labelers annotate millions of images—identifying objects, people, vehicles, etc.—so AI learns to operate safely and reliably.

2. Why Kenya Is a Global Data Labeling Hub (07:23)

• Michael attributes Kenya's prominence to:
  • Strong internet penetration
  • A well-educated, tech-savvy population
  • A culture of embracing and challenging technology
    • Quote (Michael Asia, 07:23):

      “Kenyans are tech savvy, we love challenges... That is why everyone feels so comfortable bringing their jobs here in Kenya.”

3. Michael’s Personal Path into Data Labeling (08:33)

• Michael’s journey:
  • Lost his job, faced family health and financial crises during COVID-19.
  • Entered data labeling through the company Sama, earning around $240/month—insufficient but necessary.
    • Quote (Michael Asia, 09:45):

      "I had to pick the job because I didn’t have an option."

4. The Brutality and Complexity of the Work (10:16)

• Strict quotas and monitoring: Constant activity required, with accounts deactivated after 8 minutes of inactivity.
  • Quote (Michael Asia, 10:31):

    "Most of the tools...could deactivate your account or put you on an available mode if you go eight minutes without touching the mouse."

• Types of tasks:
  • Image/video labeling, tagging, and annotation.
  • Some projects involved labeling pornography for up to eight hours a day—leading to deep emotional and psychological effects.
    • Quote (Michael Asia, 13:29):

      "You get to a point where your body can't function...you can't [be intimate]...it fractured a lot of things that time."

    • Michael underwent therapy for 6 months due to the toll of the work.

5. Labeling Pornography and Graphic Content (11:26–14:53, 41:56–42:11)

• Annotators had to tag extremely graphic and even unlawful material frame by frame for global porn platforms, with no real avenues to report illegal content.
  • Quote (Michael Asia, 41:56):

    "I viewed a 13 year old [in a graphic sexual] session and it has never left my mind...and you're supposed to put tags on that."

• Lack of specific or effective psychological support from employers.

6. Chat Moderation and Pretending to Be AI Sexbots (15:40–19:43)

• Another role involved pretending to be various "AI companions" or sexbots in intimate conversations—often with elderly, vulnerable users from Europe or the US.
• Required operating multiple personas, improvising responses, and rapidly typing long, believable messages.
  • Quote (Michael Asia, 18:49):

    "This job requires a lot of creativity and fast thinking...If I’m talking to a man, I’m supposed to act with a woman."

• These conversations blurred lines between reality and the personas, causing emotional confusion and burnout.

7. Grueling Schedules, Low Pay, and Health Impacts (19:50–21:41)

• Many worked multiple jobs—data labeling during the day, chat moderation at night—leaving no time for sleep or recovery.
  • Quote (Michael Asia, 20:10):

    "We used to work for at least 18 hours a day ... most of the guys have been in this space have a problem with sleep."

• Companies rarely provided meaningful mental health support, and therapists often couldn't relate or help effectively.

8. The Global Labor Inequality in the AI Supply Chain (33:28–36:22)

• Stark differences in pay and conditions between data labelers in Kenya and those in the US/Philippines.
  • US/Philippines: up to $50/hour and paid for training
  • Kenya: as little as $0.01 per task, no pay for training
  • Quote (Michael Asia, 33:28):

    “Is it a crime for example in the US to do the right thing? No. Then why are they doing this in Kenya? Why in Africa?”

9. Hidden Structure and Clients: Who Are Data Labelers Really Working For? (37:59–40:29)

• Kenyan labelers are subcontracted through companies like Sama, but ultimately work for US tech giants (e.g., Meta), who avoid legal responsibility via intermediaries.
  • Quote (Michael Asia, 39:00):

    "I worked for Meta...for three years ... I can say this more than 100 times, I can prove this more than 100 times...we, we need to understand."

• This abstraction protects big tech from liability and public scrutiny.

10. Organizing and Reform: The Data Labelers Association (30:03–32:37, 47:19–48:31)

• Michael and colleagues founded the Data Labelers Association in response to widespread violations and lack of protections (Dec 2023).
• Pushes for:
  • Legal reforms to include digital workers
  • Model contracts and codes of conduct to protect workers now
  • Fair pay and real mental health support
  • Greater worker representation in policy-making
• The association has over 870 members and aims for global expansion to advocate for data labelers worldwide.
  • Quote (Michael Asia, 47:19):

    "We need to address these are global problem. ... If this change has to be effected. It has to be across the globe, everywhere."

11. The Irony and Absurdity of “Training Your Replacement” (43:43–46:46)

• Michael reflects on the paradox of teaching AI tools that may eventually replace his own job, or the jobs of others like him.
  • Quote (Michael Asia, 45:21):

    "AI can never be AI without humans. ... It's not artificial intelligence. It's Africa intelligence. Most of these dirty jobs and most of these jobs have been done here in Africa."

• The influence of Kenyan data labelers is so great that even ChatGPT is said to “write like a Kenyan.” Yet workers receive scant recognition.

Notable Quotes & Memorable Moments

• “I felt like I was losing myself in the role. ... It became harder to separate the act from reality. The lines blurred. I began questioning if I was acting or if I was truly becoming the Persona I was forced to embody. I was losing touch with who I really was, a feeling that has never left me.”
  — Michael Jeffrey Asia, quoted by Jason Kebler (03:48)

• "It was something I never want to talk about sometimes because...it wasn’t an easy thing watching pornography for eight hours. And for eight months ... I went for therapy for six months." (Michael, 13:24)

• "If this job was done in the U.S. ... would they still give the pay they're giving? ... Why this discrimination? If they can pay people in the U.S. and in Philippines well, that means they can pay people in Kenya well." (Michael, 33:28)

• "We must be part of the policy making process. ... Because I went through that mess, I understand what it is and I understand what can be done to find a solution to this problem." (Michael, 23:20)

• "AI can never be AI without humans. ... It's not artificial intelligence. It's Africa intelligence." (Michael, 45:21)

Important Timestamps

• 05:30 – Michael’s introduction and explanation of data labeling
• 07:23 – Why Kenya is a hub for data labeling work
• 08:33 – Michael’s personal story entering the industry
• 10:16 – Pressures of the job: KPIs, quotas, and monitoring
• 13:24 – Labeling pornography and its mental toll
• 15:40 – Chat moderation and pretending to be AI companions/sexbots
• 19:50–21:41 – The impact of overwork and insomnia
• 30:03 – Founding the Data Labelers Association
• 33:28 – Global labor inequality in AI
• 39:00 – Working for Meta via intermediaries
• 41:56 – Encountering and tagging illegal/violent content
• 43:43 – Training AI that replaces humans; “Africa intelligence”
• 47:19 – The association’s push to go global

Conclusion

This episode exposes the reality of data labelers at the heart of AI development, focusing on the human and social cost behind smart algorithms. Michael Jeffrey Asia’s testimony is a powerful reminder of the urgent need for legal protection, fair pay, and mental health support for digital workers—especially those in the Global South. Through organizing and advocacy, data labelers seek to claim their rightful place as central contributors to the AI revolution.