Sentiment .99

A ml5.js interactive project on ethics of A.I.

December, 2021


Sentiment .99 is built using p5.js, ml5.js and TensorFlow.js and it addresses two topics: text sentiment analysis and data annotation. It uses the dataset ETHICS introduced by Hendrycks et al. in Aligning AI With Shared Human Values. And through looking at those texts in the dataset that were given 0.99 sentiment scores, almost a perfect score regarding sentiment, it is concluded that SENTIMENT DOES NOT MEAN EVERYTHING. By putting the user in the shoes of a data labeler, user are asked to go through a subset of 15 samples from the processed dataset and they are forced answer in voice: “yes” or “no”, corresponding to whether each sentence that pops up on the screen is ethical or not. It not only wants to give an immersive experience of interactive with the datasets, but also points out the various issues hidden behind the AI data preparation. All in all, it wishes to open up the “black box” of AI for the user to take a peek into certain issues.

Please see [https://github.com/MstXy/Sentiment.99] for more details.

Thank you!