Back to topics

What did Tay reveal about testing an AI system in public?

Microsoft’s 2016 account says Tay underwent testing before release, then encountered malicious use the team had not adequately anticipated. The gap between an intended interaction and an adversarial one makes deployment part of the story. The developer’s account is valuable evidence of its response, but it is also the company’s own explanation.

3–7 min

Questions you could explore

  • What kinds of behavior could a friendly test group fail to reveal?
  • Who should be able to stop a deployment when harmful behavior appears?

Start reading

Ready to try your ideas aloud?

Your topic carries over. Record up to 2½ minutes, depending on your plan.

Computers and AI

Why can a damaged QR code still work?

A scratch or stain does not always stop a QR code from being read. Its stored information includes redundancy, but that does not mean any part of the image can be covered without consequence.

3–5 min
Computers and AI

How does a website name become an internet address?

You remember a website by its name, while the network needs an address for the connection. DNS supplies that translation, sometimes from a saved answer and sometimes by asking other servers.

4–5 min
Computers and AI

Same emoji, different picture

The same emoji can arrive with a different expression or style from the one you saw when sending it. Unicode identifies characters and sequences, while software platforms choose artwork and decide what their versions support.

2–3 min
What did Tay reveal about testing an AI system in public? | Speech Topic | speaking.app