From the Understandably newsletter:
I prompted ChatGPT for a definition. Here's what it came up with. (Hold onto this if you can, because we're going to come back to it):
Artificial Intelligence (AI): the capability of a machine or computer program to mimic human intelligence processes, learn from experiences, adapt to new inputs, and perform tasks that typically require human intellect.
I have to say: Not a bad definition. (Of course, if there's one thing AI should get right, it's probably the definition of AI).
I was prompted to use that prompt after reading an insightful, fun and marginally dystopian article on New York magazine's website, about the thousands of people toiling away at highly robotic tasks, in order to train AI robots.
Reason: If AI is supposed to learn from experiences and adapt to new inputs, then somebody has to provide the initial experiences and inputs.
This is a real job (maybe more of a "gig") to be exact. The people who do it are called "annotators" (or "taskers"). They're all over the world—primarily in economically challenged areas—for example, in small towns in the USA, sprawling African cities, and remote areas of South Asia.
If you're familiar with Mechanical Turk, aka MTurk, that's the idea: "a crowdsourcing website ... to perform discrete on-demand tasks. In fact, former Princeton professor Fei-Fei Li used MTurk 15 years ago to come up with people who were willing to label millions of images for AI purposes.
Now, there are tons of companies doing this, and their founders and stakeholders are getting rich, even while the annotators make roughly "the hourly living wage" for their location or more, according to a subsidiary of Google involved in the effort. The jobs are often mind-numbingly boring, requiring "learning to think like a robot" and follow extremely specific instructions.
But, they're better than other jobs the taskers are likely to find where they live. Among the typical tasks that the annotators and taskers do, as cited in the article:
- video calls,
- identifying corn in images and videos (apparently, "for baffled autonomous tractors"),
- looking at hundreds or thousands of photos of crowds, and identifying all the knees and elbows, since computers can't do that until we teach them,
- classifying American Reddit posts by emotion (there were noteworthy errors with this one, since the task was outsourced to countries where annotators were unfamiliar with U.S. slang), and
- simply making conversation with an AI chatbot for hours at a time, in order to teach it how to sound more human.
That last task earned an American worker $14 an hour. "It definitely beats getting paid $10 an hour at the local Dollar General store,” she told New York. The hardest challenge? Running out of things to talk about. "I just Google interesting topics."
There are also higher-paying tasks requiring specialized knowledge, like identifying specious or dangerous legal advice (J.D. preferred), or rating translations. (I found a few gigs, in case anyone happens to be fluent in Igbo or Sesotho sa Leboa, and would like to make a few bucks).
Anyway, I recommend reading the whole article if you can, but just to highlight a few key things:
First, for both corporate secrecy reasons, and maybe just to avoid a revolt, the annotators and taskers aren't told who they're actually working for, or even why they're doing it. However, they do sometimes figure it out. (“I Googled and found I am working for a 25-year-old billionaire,” said one worker in Africa. “I really am wasting my life.”)
Second, it's worth noting that if annotators are providing the experiential building blocks of AI, then at least some AI will start out by seeing the world as early 21st century humans with low economic prospects do, since they're the annotators.
Finally, companies employing annotators are always trying to find ways to cut costs, for example taking the jobs that were for Kenyans and moving them to Nepal. So, how do the workers react? Well, to copy the AI definition above, they "learn from experiences, adapt to new inputs, and improve their performances.
For example, they form networks on WhatsApp to share info about the best gigs. They get VPNs to suggest their in Missouri or Texas instead of Nigeria (to qualify for higher pay). And maybe most fascinating:
A Kenyan annotator said ... now, he runs multiple accounts in multiple countries, tasking wherever the pay is best. He works fast and gets high marks for quality, he said, thanks to ChatGPT.
The bot is wonderful, he said, letting him speed through $10 tasks in a matter of minutes.
So, we wind up with billion-dollar companies paying low wages to people to teach AI technology, but then the people making the low wages figure out how to use the same AI technology to trick the billionaires. Sounds peachy. Can anyone identify if there might be a problem with that?
Original article:
https://nymag.com/intelligencer/article ... ctory.html
I just raised it as food for thought that things may not always be as they appear at first sniff. The very fact that some of these companies seem keen on legislation is a cause for at least a raised eyebrow or two).
