Why IBM’s AI truth sheets must be the business customary


Every now and then an idea comes up that is so good that you wonder why it took so long for someone to think about it. IBM’s AI fact sheets are one of those ideas.

AI fact sheets are similar to heavily packaged nutritional labels for food. They contain information about the development, capabilities, benchmark performance, and much more of an AI model.

Big Blue today announced plans to “commercialize key automated documentation capabilities from IBM Research’s AI Factsheets methodology in Watson Studio in Cloud Pak for Data during 2021”.

In other words, companies and developers using Watson Studio in Cloud Pak for Data will soon have access to an automated AI fact sheets tool to generate transparency and information reports. The tool automatically generates most, if not all, of the information in the AI ​​fact sheet.

Via IBM:

The FactSheet project aims to build confidence in AI by increasing transparency, improving understanding of how AI is created and deployed, and empowering governance to control how AI is created and deployed . Increased transparency provides information for AI consumers to better understand how the AI ​​models a program component that is generated through learning patterns in training data to make predictions about new data, e.g. B. a loan application. or service an executable program that is provided behind an API and makes it possible to respond to program requests from other programs or services has been created. In this way, a consumer of the model can determine whether it is suitable for his situation.

Take quickly: We love this idea. While it’s a little more complex than we can describe in this article (research report here), the bottom line is that anything that standardizes transparency in machine learning models is a good thing.

This is part of the IBM “AI Governance” initiative and new AI advisory services. The company also announced some interesting new Watson features.

Today we announced Watson Discovery’s new reading comprehension feature in beta. It can identify precise answers to natural language questions from a wide variety of complex corporate documents. https://t.co/Hunb0jPtBr #ai #ml #nlp

– Fredrik Tunvall (@tunvall) December 9, 2020

Final thoughts: We’ve heard a lot about the “black box” over the years, but the truth is that the black box doesn’t hide much. The real problem with AI in large business is that there is no standard to explain what an AI model does specifically and how well it can do it.

Realistically, the model with the best marketing team is likely to beat the best model if no one can objectively explain which one is best for a particular application.

AI Fact Sheets is a blueprint for an industry standard to address these issues head on.

Published on December 9, 2020 – 19:26 UTC

Leave A Reply

Your email address will not be published.