An Algorithm For De-Biasing AI Systems

A fundamental truth about AI systems is that training the system with biased data creates biased results. This can be especially dangerous when the systems are being used to predict crime or select sentences for criminals, since they can hinge on unrelated traits such as race or gender to make determinations.

A group of researchers from the Massachusetts Institute of Technology (MIT) CSAIL is working on a solution to “de-bias” data by resampling it to be more balanced. The paper published by PhD students [Alexander Amini] and [Ava Soleimany] describes an algorithm that can learn a specific task – such as facial recognition – as well as the structure of the training data, which allows it to identify and minimize any hidden biases.

Testing showed that the algorithm minimized “categorical bias” by over 60% compared against other widely cited facial detection models, all while maintaining the same precision of detection. This figure was maintained when the team evaluated a facial-image dataset from the Algorithmic Justice League, a spin-off group from the MIT Media Lab.

The team says that their algorithm would be particularly relevant for large datasets that can’t easily be vetted by a human, and can potentially rectify algorithms used in security, law enforcement, and other domains beyond facial detection.

How To Telepathically Tell A Robot It Screwed Up

Training machines to effectively complete tasks is an ongoing area of research. This can be done in a variety of ways, from complex programming interfaces, to systems that understand commands in natural langauge. A team from MIT’s Computer Science and Artificial Intelligence Laboratory (CSAIL) wanted to see if it was possible for humans to communicate more directly when training a robot. Their system allows a user to correct a robot’s actions using only their brain.

The concept is simple – using an EEG cap to detect brainwaves, the system measures a special type of brain signals called “error-related potentials”. Simply noticing the robot making a mistake allows the robot to correct itself, and for a nice extra touch – blush in embarassment.

This interface allows for a very intuitive way of working with a robot – upon noticing a mistake, the robot is able to automatically stop or correct its behaviour. Currently the system is only capable of being used for very simple tasks – the video shows the robot sorting objects of two types into corresponding bins. The robot knows that if the human has detected an error, it must simply place the object in the other bin. Further research seeks to expand the possibilities of using this automatic brainwave feedback to train robots for more complex tasks. You can read the research paper here.

MIT’s CSAIL work on lots of exciting projects – their video microphone technology is truly astounding.

[Thanks to Adam Connor-Simmons for the tip!]