# Processing with AI
## Exploration : IA & Ethics

Name:
> Iman DAKHLAOUI
>
Subject:
> Detect student attention in class using computer vision

>[TOC]

## Design brief

### Bias

If we don't source our dataset with enough rigor, the following bias might appear:

>1. Only recognizing one gender (males), and not being able to recognize females, therefore, only part of the students would be detected by our computer vision
>2. Only recognizing one skin color type (light skin), therefore, only part of the students would be detected by our computer vision
>3. Not recognizing transgender people, or man with long hair, women with short hair, man with wigs, man with make-up, women with hijab... Therefore, only part of the students would be detected by our computer vision

We will ensure that our model is not biased by:

>1. Sourcing our data from more than only one skin color, one gender... Taking a wide range of different people to make our tool able to recognize a wide range of student, all different
>2. Making sure our data take into account different lightings of the room. For instance, I am darkskin, and when I use snapchat filter in a bad lightning, it odes not recognize my face, while it may recognize the faces of my lightskin friends
>3. Making sure the recognition doesn't fail when there's more than one person in front

### Overfitting

We will make sure our model does not overfit by males and white people, as Joy Buolamwini explains it is often the case.
> Checking the accuracy of our model on people of all gender and independently of physical criterias such as the lenght of the hair, the use of scarves on hair, the use of make-up, of all skin colors.

### Misuse

>We have to remind ourselves that our application could be misused by parents, professors to track their children/ students, in order to know what everyone is doing at all time in school time. This kind of misuse would clearly be privacy invasion.

### Data leakage

>In a catastrophic scenario, where all of our training dataset were stolen or recovered from our model, the risk would be that ill-intentionned people could recreate the schedules of each and everyone of the students, and know exactly when and where they are (it can lead to kidnappings, racketeerings), people could also make data on attendance depending on student's ethnicity (which is strictly forbideen in France)...

### Hacking

> If someone found a way to "cheat" our model and make it make any prediction that it want instead of the real one, the risk would be that anyone could pretend to be any student, and our device would still mark them as present instead of absent, pretend to be attentive while doing absolutely something else...