|
|
Do you read the news the way our AI does?
Join a short academic study with UNESP. No experience required
|
|
Hi there,
|
|
For the past few months our researchers at AskNews have been
collaborating with researchers at the Graduate Program in
Computer Science (PPGCC) at São Paulo State University (UNESP)
to build a bias extraction model. Today, we'd like to invite
you to spend 10 to 15 minutes participating in the next stage
of our research!
|
WHAT WE BUILT
|
|
BiasExpert is a small language model we trained to spot bias
in news coverage. It recognizes 18 kinds of bias, the patterns
that shape how a story reads: loaded word choice, sensational
framing, opinions presented as fact, and more. You can read
our white paper
here. You can also explore the model on
Hugging Face
and read our
blog post.
|
|
The training labels were created using a geographically
diverse consensus of several state of the art reasoning
models. That leaves a question we can't answer on our own: how
does a model built this way compare with actual human readers?
|
WHY WE'RE ASKING YOU
|
|
That's what this study is for. Your answers show us, bias type
by bias type, where the model reads text the way people do,
and where it doesn't. Both results are useful. Honestly, the
disagreements might be the most interesting part.
|
HOW IT WORKS
|
-
You read 6 short news paragraphs. We use paragraphs rather
than full articles to keep it light, and we hide the
headline and the outlet so your reading is about the text
alone.
-
Each paragraph comes with two yes/no questions about a
specific kind of bias. "Not sure" is always an option.
-
The questions cover eight of the model's 18 bias types, the
ones a reader can judge fairly from a paragraph on its own.
Each is explained with examples right before you're asked
about it. No preparation, no jargon.
-
At the end you see one number: how often you and the model
agreed.
-
The whole thing takes about 10 to 15 minutes.
|
|
Nothing is graded. There are no wrong answers, and where you
disagree with the model is precisely what we want to learn.
|
|
A quick practical note: the study runs on AskNews accounts.
That's how we keep one session per person, so you'll log in
with your AskNews account or create a free one.
|
|
And the fine print, up front: this is an academic study,
taking part is voluntary, you can stop at any time, and
nothing personal about you will ever be published.
|
|
Take part in the study
|
|
Thanks for reading, and for helping if you can.
|
|
The AskNews team
|
