Guess who - AI version

From Competendo - Digital Toolbox
Jump to: navigation, search
Guess Who is a popular game where opponents attempt to guess which character out of 24 possibilities their opponent has picked. This adaptation can be used in educational settings to reflect on how liable to error facial recognition systems can be.

Goals

  • Understand AI biases by playing with human ones

Context

Any face recognition system uses biometrics to map facial features from a photograph or video, comparing the information with a database of known faces to find a match. Let’s pretend that the participants are detecting machines: how rich in information are their “internal databases”? Celebrity pictures can be used in order to play with real face features.


Steps

1. The participants, split into two teams, and play the game, following the provided instructions:

  • each team picks a “mystery person” that the other team has to guess;
  • each team is allowed to ask one question per turn, answering either with “yes” or “no”;
  • if one guesses the mystery person wrong, they lose. At least three game rounds must be played.
  • Here you find a set of celebrity cards ready to test: Template: Celebrities in A4

2. When the game ends, the educator prompts a reflection session:

  • What is the process behind your attempts to guess the mystery person?
  • How many questions did you need to guess?
  • What was easy and what wasn’t so?
  • How did you feel while playing the game?
  • The questions that participants used to guess can be discussed specifically

3. Transfer to facial recognition: the educator explains how facial recognition systems work. This can be supported by the Algorithmic Justice League (AJL) materials (see above), for instance, the “Gender Shades” video explainer. Thoughts about AI accuracy, inclusiveness and fairness can be collected on a poster.


Reflection

  • How does the game relate to face recognition technology?
  • Did you experience bias and stereotyping during the game?
  • How can we fight stereotypes and simplification?

Valentina Vivona

Researcher at the Osservatorio Balcani Caucaso Transeuropa (OBCT), a think tank focused on South-East Europe, Turkey and the Caucasus located in Trento (Italy).


Variation

If you create your own set of cards, make sure that the characters have some small peculiarities, but also that they share enough common attributes. If only one character wears a hat, for example, it would be too easy to detect!

Creating the set of cards can be part of the educational activity. The educator can prompt the participants to look for “very difficult people to guess” and their choices can be discussed consequently.

Pictures can be picked and downloaded from Humanæ’s photographic work by artist Angélica Dass. The project aims “to document humanity’s true colors rather than the untrue labels “white”, “red”, “black” and “yellow” associated with race [...] The background for each portrait is tinted with a colour tone identical to a sample of 11 x 11 pixels taken from the nose of the subject and matched with the industrial pallet Pantone®, which, in its neutrality, calls into question the contradictions and stereotypes related to the race issue”.

Experience

  • The educator can forbid certain questions such as, “Is it a woman/Is it a man?”.
  • Skin types are likely to stir a debate which is part of the game; the educator can prevent stigmatization, for example prompting the participants to find other words than “White” and “Black” and emphasize instead other descriptors.
  • To bring out both the cognitive and the discriminatory mechanisms, it could be useful to devise at least one game round with cards all representing celebrities of the same ethnicity.



Time 60-120 minutes

Material 3 printed sets of cards, 1 poster

Group Size 2-20 people

Keywords discriminatory bias; identity; facial recognition systems


From:

CCI.png

Related:


Learning the Digital

Digital-book-cover.png A Competendo Handbook

Read more