AI MUST READS — W25 2018, by City AI
Artificial Intelligence, Machine Learning and related fields are in a constant state of change. We want to inform but also encourage discussions on well presented topics we think are necessary in the context of putting AI into production. Every week we’re picking applied AI’s best articles plus adding a discussion starter
1. Unmasking A.I.’s Bias Problem
Machine-learning algorithms “haven’t been optimized for any definition of fairness, they have been optimized to do a task.”
This is one of the most comprehensive articles on the current issues regarding Bias within A.I. and manages to stand out amongst the others by drawing from the opinions, viewpoints and experiences of stand-out experts within the field.
Its refreshing (IMO) to find that actually, the publication itself doesn’t seem to offer their own opinion but instead relies upon the view points of the genuine authorities, view-points that if genuine should be encouraging in their honesty about the current issues that the technology faces and the realisation that proactivity in this situation is important in order to allow Artificial Technology to reach its potential.
Author — Johnathan Vanian
2. OpenAI Five
OpenAI Five plays 180 years worth of games against itself every day, learning via self-play.
In the past OpenAI has beaten top DOTA players in the world at the 1v1 game mode, they have now been able to progress and develop their Artificial Intelligence and beat a team of the top 1% Amateurs in a 5v5 match. It should be noted that there are limitations placed upon the matches which leads to a limited experience for the human players.
However, the most interesting and fascinating part of this development is the level of reinforcement training that took place, these systems played the equivalent of 180 years worth of games against itself. What limitations can hardware place upon the potential of reinforcment learning?
3. Adobe is using machine learning to make it easier to spot Photoshopped images
The development of technologies like this that are built in direct opposition to the development of technologies like DeepFake are crucially important in the modern age of media, where it’s already hard to discern between legitimate and false imagery without the involvement of Artificial Technology.
I’m predicting that soon we will be in an Artificial technology based arms race across multiple sectors from cyber-security to image/video editing. There will be a running trend in the coming years that technology will be developed that that will incorporate Artificial Intelligence simply to combat other forms of software that already use A.I.
Author — James Vincent
Join 6,000+ AI practitioners from over 100 countries at WorldSummit.AI this October!