Hetan's right. #Transparency alone is insufficient, and his combined proposals provide a far better framing for how to achieve #AlgorithmicAccountability.
A few thoughts:
On #piloting, or what in medicine would be called #ClinicalTrials, to check solely for #bias is missing...
...the point. #Bias is just one side-effect; trials must (in stages) test for any and all - so the general approach must attend first to #safety (including 'rights impacts'), then to #efficacy, then #effectiveness.
While I agree publishing models may be of limited use, it is...
...still important that they are - if only to establish the principle that they can be forensically analysed when something has gone wrong, not hidden behind '#CommercialConfidentiality', etc.
Publishing the #provenance of the training dataset, and assumptions, is vital however!
Ongoing #monitoring (in medical terms, Phase 4 trialling / #pharmacovigilance) is a must-have.
As recent events have shown, it's really only once we see how #algorithms play out on millions of people's lives that we truly understand them. And in this we should be guided by...
...#MarthasTalisman* - counterfactuals too.
The weaknesses of #GDPR on both "solely" automated and #redress (not to mention a supine #regulator) may be fixable, but we can't afford to wait for the legislators on this one. We'll see you in Court...
__
*medconfidential.org/2017/remember-…
Finally #governance. Some would be nice! But, like '#ethics' in AI, it's still sorely lacking for all that it's talked about.
Maybe all the failed efforts should be examined for common characteristics* that should be avoided next time?
_
*or characters: news.sky.com/story/chair-of…
Share this Scrolly Tale with your friends.
A Scrolly Tale is a new way to read Twitter threads with a more visually immersive experience.
Discover more beautiful Scrolly Tales like this.