Transparency in Artificial Intelligence

Transparency in Artificial Intelligence

On 8 June 2021, the CUHK Fintech Webinar gathered over 150 participants to introduce a new tool, AI FactSheet to enhance AI transparency and governance.

Watch the video extracted from the webinar:

Transparency in Artificial Intelligence
Artificial Intelligence might sound abstract and technical, especially for those without relevant background. Aiming to resolve the issue, IBM has developed a new tool, AI FactSheet to enhance its transparency and governance. In this seminar, our speaker, John Richards from IBM Research has introduced their powerful tool, and explained on how it works and how it can help.

What is a FactSheet?
FactSheet is a tool to document all necessary information when building an AI model or service, similar to the nutrition facts label on packaged food or energy label on the electrical appliance.

It can be regarded as a detailed FAQ guidebook, which is created from Model Facts based on the schema from FactSheet Template. The template defines the questions in the guidebook, like what information or metrics are to be collected, such the producers can retrieve the answers from the Model Facts.

At the moment, the FactSheet is still a concept, but IBM has announced to productize the revolutionary technology within this year, it is believed that it would bring huge benefits to businesses that possess AI models.

Why do we need a FactSheet?
FactSheet provides thorough, simple and straightforward information for AI consumers, enhancing the transparency for model or service, allowing them to better understand or trust the AI technology.

With FactSheet, AI governance can also be enabled via extensive information coverage. Automate documentation of AI lifecycle can be applied, where most useful information can be captured, without changing the existing procedures. FaceSheet also specifies and enforces actionable AI lifecycle policies for sharpening regulators or enterprise governance. Last but not least, FactSheet makes information assessable to all stakeholders, which can further encourage the collaborations between the parties.

How to create a FactSheet?
To begin, it is crucial to understand the FactSheet consumers, knowing what and why do they need. Then, it is also required to understand the FactSheet producers, anticipate what information they might generate.

After figuring out the problem statement, then producers can proceed to create a FactSheet template to define the topics and questions required. When the template is completed, producers should try to fill in the template once to assess if the framework is corrected designed. Afterwards, stakeholders can start filling it in with the information retrieved from the Model Facts; resultantly, the FactSheet is completed.

With the actual FaceSheet, stakeholders and consumers should evaluate its quality once again, to check whether any enhancement is needed; otherwise, we can confirm that the FactSheet is ready for use.

For the launching of future AI models or services, their templates can be built based on any existing templates, which can act as references to speed up the progress.

Speaker :

John Richards
Distinguished Research Staff Member, Research,
IBM US

Date :   8 June 2021 (Tue)

Time :  5:30  –  6:30 p.m. (Hong Kong Time : GMT+8)

Language:  English

Venue:        Live virtual class (Zoom)

Fee: Free Registration. A processing fee applies for proof of attendance

As AI models and services are used in a growing number of high-stakes areas, a consensus is forming around the need for a clearer record of how they are created to increase trust.  Several proposals for higher quality and more consistent AI documentation have emerged, addressing technical details along with ethical concerns and potential social impacts.  This talk will provide an overview of FactSheets, IBM’s approach to AI documentation.  FactSheets will typically contain model purpose, performance, safety, security, and fairness information collected at various points in the AI lifecycle.  We will review some of these common elements and discuss a methodology for creating useful FactSheets.

BIOGRAPHY

John Richards
Distinguished Research Staff Member, Research,
IBM US

John Richards is a Distinguished Research Staff Member in the IBM Research AI department. He is a Fellow of both ACM and the British Computer Society and has received numerous awards including the Lifetime Practice Award from ACM SIGCHI. His recent research in human-centered AI has focused on the quality of chatbots and techniques for documenting AI models and services.

FINTECH SERIES

The CUHK FinTech Seminar Series is co-organized by the CEFAR Academy and the Asia-Pacific Institute of Business (APIB). Industry practitioners are invited to campus to share their areas of expertise.  We see the advantage of bringing together the academia and industry experts, providing a platform to carry out fruitful and productive discussions, encourage the exchange of ideas and promote future collaboration opportunities.  CUHK students may also widen their exposure and receive some enlightenment in their career development.