This is a great set of sources and synthesis. In case it's of interest I wanted to point you to another batch of relevant papers that I wrote up here: https://www.ai-accountability-review.com/p/closing-information-gaps-via-ai-transparency -- accessibility and understandability are important, but there are other factors that also contribute to quality transparency info.
Hi Nick, thanks for sharing! Always great to connect with people interested in transparency.
I should clarify this article doesn't cover the full thesis nor the full transparency question. What my research was focused on was perceived trust by the end user, which isn't the same as actual technical accuray of either the prediction or the explanation. Much of the transparency documentation is reserved for the technical teams deploying and auditing the systems. What I'm looking at are the people using the system - particularly non-technical end users. How can they make sense of what they're doing?
Really interesting, thanks for sharing. fully agree more ai literacy is needed, and not just promting. but what does training for nontechies look like? knowledge levels are miles apart
I think training for end users doesn't need to be complicated or technical. It should be focused on the overall logic and reasoning of the system. I think this would reduce many of the misunderstandings and improper use scenarios we are seeing today
I read your full thesis a few months ago and this is so interesting. I remember you also wrote about how explanations can be be used to give credibility for incorrect predicitons, will you be covering this in future articles?
I am very interested in the best way to present information. What does the middle ground look like? Would love to have a chat with you about this.
Thank you again for your interest and support. I may write a separate article about the halo effect and white box effects later, if that sounds interesting. I'm trying to stick to my own rule of own concept per article to keep things digestible.
As for what the middle ground looks like, this is a great question. It's context dependent, but what is missing in the cases I've seen is the global explanation (on system level). Too much focus on local explanations (decision level) means that it still doesn't always make sense to a user how the system reached that conclusion.
This is a great set of sources and synthesis. In case it's of interest I wanted to point you to another batch of relevant papers that I wrote up here: https://www.ai-accountability-review.com/p/closing-information-gaps-via-ai-transparency -- accessibility and understandability are important, but there are other factors that also contribute to quality transparency info.
Hi Nick, thanks for sharing! Always great to connect with people interested in transparency.
I should clarify this article doesn't cover the full thesis nor the full transparency question. What my research was focused on was perceived trust by the end user, which isn't the same as actual technical accuray of either the prediction or the explanation. Much of the transparency documentation is reserved for the technical teams deploying and auditing the systems. What I'm looking at are the people using the system - particularly non-technical end users. How can they make sense of what they're doing?
Really interesting, thanks for sharing. fully agree more ai literacy is needed, and not just promting. but what does training for nontechies look like? knowledge levels are miles apart
Thank you so much!
I think training for end users doesn't need to be complicated or technical. It should be focused on the overall logic and reasoning of the system. I think this would reduce many of the misunderstandings and improper use scenarios we are seeing today
I read your full thesis a few months ago and this is so interesting. I remember you also wrote about how explanations can be be used to give credibility for incorrect predicitons, will you be covering this in future articles?
I am very interested in the best way to present information. What does the middle ground look like? Would love to have a chat with you about this.
Hi Angie,
Thank you again for your interest and support. I may write a separate article about the halo effect and white box effects later, if that sounds interesting. I'm trying to stick to my own rule of own concept per article to keep things digestible.
As for what the middle ground looks like, this is a great question. It's context dependent, but what is missing in the cases I've seen is the global explanation (on system level). Too much focus on local explanations (decision level) means that it still doesn't always make sense to a user how the system reached that conclusion.