[Ana]

Our upcoming webinar Towards Transparent and Explainable AI: The Current Standardisation Landscape will provide an overview of the most prominent international standardisation initiatives on AI transparency and explainability, including IEEE 7001 (published) as well as ISO/IEC AWI 12792 and ISO/IEC AWI TS 6254 (standards under development).

The event will provide a unique opportunity to hear directly from working group chairs and editors for all three standards and will include a discussion segment focused on questions from AI Standards Hub community members. If you have questions for our speakers, please submit them in advance of the event by posting them in this thread.

• This topic was modified 1 year, 4 months ago by Ana.

[BEBen]

Okay, here’s a question, to prompt some discussion…

The technology development process is constantly accelerating. Standards take time, and a lot of deliberation and consensus, to produce. How do you think the standards development process (and perhaps the subsequent laws) could or should accelerate to keep up with the rapid pace of technology such as AI.

[Jon]

Good question Ben. Standards such as BSI or ISO are powerful tools where there’s a high degree of certainty or maturity in the sector. However, they aren’t the only way in which standards can be developed.

BSI also offers fast-track standards such as Publicly Available Specifications (PAS) or BSI Flex. Flex is the faster of the two as it can enable initial content (a version 1 standard) to be released in as little as six to seven weeks, with the content, scope, and type of standard able to be updated iteratively over subsequent versions. This makes Flex ideally suited as an up-to-date reference point when the subject matter is constantly accelerating.

A BSI PAS may be a suitable option where speed is important (they take around 12 months to develop), but there is also more certainty in the subject matter. Both PAS and Flexes can also be used as the basis for future BS and ISO standards. More information on each of these can be found on BSI’s Fast Track Standards page.

[CACarl]

Testing of AI based systems seem to be a non-existent topic in the AI standards hub, unless I have missed it in which case could you point to where it is covered. Transparent and explainable AI will require some transparency on the testing. It does not seem to be clear what standards we have for actual testing of the AI in the AI based systems. Again it may be that I have missed these or do not understand where it has been covered so would appreciate some pointers towards info on this. (Note when I cover testing I mean the software engineering perspectives on testing (not just testing an AI new algorithm performs better than other algorithms), such as testing performed on safety and mission critical systems … an issue that will become prominent as AI is used on such systems.)

[Author]

Posts