Modern Slavery? A.I.'s Exploitation of Kenyan Workers

Send us a text

The dark underbelly of artificial intelligence development comes to light in this eye-opening exploration of how our most sophisticated AI tools are made "safe" for public consumption. When OpenAI was training GPT-3, the predecessor to ChatGPT, they discovered their model regurgitated the worst parts of the internet—racism, violence, abuse, and more. Their solution? Outsource the cleanup to Kenyan workers earning less than $2 per hour.

Episode Transcript

Speaker 1 (00:00):
Peace family.
You're tuned into the sound ofthe only one mic podcast, mic
drop, where we break down theday's biggest moves and culture
and community throughconversation that matters.
All right, once again, I saypeace to your family and I hope

(00:25):
that you're feeling well andblessed today.
I'm your host, carl Gerard, andI appreciate you locking in
with me.
Whether you've been followingevery episode, this is your
first time tuning in.
This is a space about truth,clarity and conversations that
truly matter, and I'm not hereto tell you what to think, folks
.
I'm just here to lay it out foryou.
All right, let's get into thisy'all.
The thing, folks, I'm just hereto lay it out for you.

(00:46):
All right, let's get into thisy'all.
It's a story that might be newto some people and might not be
to others, and I just wanted topull the curtain back on it.
It's a story that doesn't getenough spotlight, a story that
reminds us that even the mostpowerful tech in the world is
still built on the backs ofpeople.
I'm talking about a TimeMagazine investigation into open
AI.
Yes, the folks behind ChatGPT.

(01:08):
We all use it, we all love it Iguess Some of us to some
various degrees and it talksabout how they outsourced the
most traumatic, disturbing workto low paid workers in Kenya.
You heard me that AI magic weall use.
It was made safer by people whohad to process some of the

(01:34):
darkest parts of humanity forless than get this folks $2 an
hour.
So here's the breakdown.
Back when OpenAI was trainingits large language models like
chat GP I'm sorry, like GP GPT-3, pardon me folks GPT-3, which
that came before chat GPT theyrealized that the model could

(01:58):
generate toxic content, racism,sexism, violence and, straight
up, hateful stuff.
And that's because it wastrained on the internet, which
we all know isn't exactly thecleanest source.
So to clean it up, openai hireda third party company called
Sama Sama, set up shop inNairobi, kenya, and hired
workers to do what they calldata labeling.

(02:20):
And that sounds simple, right,but these folks weren't laboring
cat picture labeling like catpictures, and they were.
They were like reading graphiccontent and disturbing texts,
and the AI would learn how tolike filter it out.
One worker said he had to readdescriptions of child abuse and
murder daily, daily, and thiswas for a dollar and 32 cents an

(02:42):
hour.
Now, let that sink in for aminute, guys.
Now here's where it gets evendeeper.
Open AI pays someone about$12.50 per hour, but the workers
themselves, they receive lessthan two bucks of that, and this
isn't like fast food work.
This is psychologicallydamaging stuff, right?
Workers reported havingnightmares, flashbacks and

(03:04):
mental breakdowns.
I've seen one report where, youknow, a man was saying like
basically couldn't be intimatewith his wife because of the
stuff that he was looking at.
It was real sad stuff.
So SAMA claimed to offer mentalhealth counseling, but most of
it was in groups and I even seenin one report where they was
like they told them like formental health counseling
counseling, they told them, golook at funny cat videos on the

(03:25):
internet, which is crazy.
When the workers asked for moreone-on-one therapy, they were
denied.
Eventually, the contract wascut short in early 2022 and sama
backed out open ai moved on andthose workers you know many of
them were just left in the dustright.
So it doesn't end there.

(03:45):
From the pain came resistance.
In Nairobi, more than 150content moderators, including
some from the chat gpt project,came together to form Africa's
first content moderation unionand that's a real thing.
They've been calling, they'vebeen calling on kenyan
parliament, tech firms and eventhe us government to bring labor

(04:06):
protections to the digitalunderworld, because that's what
this is a digital underworld,the part of the internet no one
wants to see unless you're beingpaid to clean it up.
And this ain't just about openAI, it's about the global
machine.
The pattern is familiar Richtech giants in the West, they

(04:28):
outsource the dirtiest digitallabor to continue in the global
South for pennies, and it'salways under the radar, right?
But here's the thing, folks.
Here's the thing.
If these companies want to callthemselves ethical or
responsible AI leaders, thenthey got to take responsibility
for the humans in the loop.
This is about tech, yes, whichis true, but it's also about

(04:50):
labor.
It's also about dignity andtrauma and what we're willing to
ignore just to make our lives alittle more convenient.
So let me know what you thinkabout that.
One in the comments, folks.
So let me know what you thinkabout that.

(05:19):
One in the comments, folks.
And also, the next time you'rechatting with an AI, remember
there's a human story buried,but it's also just human
exploitation period.
Okay, the Only One Mike podcastis available on all major
platforms you stream yourpodcasts on.
So wherever you listen, that'swhere we are.
Be sure to check out ourYouTube channel to catch up on

(05:41):
the past and current episodesand don't forget to rate the
show and subscribe.
Hit the subscribe button, folks.
It really helps us out a lot,guys and I want to thank all of
the uh new subscribers.
Like I said before, the channelwas starting to grow
exponentially with a lot of umsubscribers in the last few days
and it helps more than you know.
It helps more than you know andI helps more than you know and

(06:01):
I truly appreciate it.
Guys, you can connect with uson Instagram and X at the Only
One Mike P1, facebook andLinkedIn at the Only One Mike
Podcast.
Email us at TheOnlyOneMike00 atgmailcom.
Call or text us at 302-367-7219.
Your voice, your questions,your take could be featured on

(06:23):
the show.
The only one.
My podcast mic drops.
Carl Gerard is out.
Peace.