Skip to content
IA pra Gente
PT
Aa

Where does AI get so much information?

· 1 min read · by Stephanie Ferreira dos Santos

Every AI needs to "eat" before it works. Its dish is called data.

Data is simple stuff: a sentence on a website, a photo of a dog, a restaurant review, a voice message. Put billions of these little pieces together and you have the raw material to train an AI.

Where do so many pieces come from?

  1. The open internet. Websites, encyclopedias, forums, scanned books.
  2. Companies. Purchases, recorded calls, customer service histories.
  3. You. Every click, like, uploaded photo and search can become an example.

The detail nobody mentions

Not every piece of data was collected with clear permission. Sometimes the "permission" was in that 40-page contract nobody reads. That is why there are laws about it (we cover them in the edition on your rights).

And if the dish is spoiled?

If the AI learned from prejudiced texts, it repeats prejudice. If it saw few photos of older people, it recognizes older faces poorly. In short: bad data, bad AI. People who work in AI governance spend a good part of their time on exactly this.

Keep this

AI does not invent knowledge out of thin air. It learns from what already exists, with all the qualities and flaws that already exist.

A question for Sunday lunch: do you know which apps on your phone use your photos to train some AI?

Found it useful? Pass it on: WhatsApp

Get the next edition by e-mail

One edition every two weeks. Plain language, illustrated, no ads. Cancel anytime.

This helps us write better for you