Data Science Investment Counter —Funding raised by UK data science companies in 2018.
£ 5.640 Million

Google tackles AI bias with open-source tech

Google Chief Executive Officer Sundar Pichai announced the company is working towards making its artificial intelligence and machine learning models more transparent in order to tackle bias.

Writing on Google’s blog in the occasion of the company’s flagship event I/O ’19 on Tuesday, Pichai described a new list of AI improvements and said Google is moving to implement them on more devices.

“Building a more helpful Google for everyone means addressing bias. You need to know how a model works and how there may be bias. We will improve transparency,” said Pichai.

Google CEO also revealed a new machine learning method called TCAV, or Testing with Concept Activation Vectors. TCAV is an interpretability method able to understand what signals your neural network models employ for prediction.

In theory, TCAV’s ability to understand such signals could locate bias because it would highlight, for example, whether males were a signal over females and other discriminating issues such as race, income and location. Using TCAV, computer scientists will be able to see how high-value concepts are valued. 

Bias is a substantial issue in artificial intelligence. Because AI applications don’t always reveal the logic behind their decisions, Pichai suggested, tackling bias should be introduced in training machine learning logic as well as data, thus reducing and eventually eradicating bias-related errors.

Pichai said he is aware that, because of the limited clarity behind its judgement, many enterprises are still hesitant to trust AI when it comes to making important decisions. He hopes that, through TCAV, companies may start re-evaluating the role of AI in their future developments. To this end, Google made TCAV open source.

 “We’ve open-sourced TCAV so everyone can make their AI systems fairer and more interpretable, and we’ll be releasing more tools and open datasets soon”, Pichai explained.

“There’s a lot more to do, but we are committed to building AI in a way that works for everyone.”

Image via PXhere and Google.


Co-working space and blog dedicated to all things data science.

Subscribe to our newsletter