@MSFTResearch Great question! For me, challenges are about balancing the need to be transparent about the data presented and yet making it clear and actionable. Should one hide uncertainty if one wants to engender action?

@ploy_sundays_ I am not a Weka user, but it seems your question is about how to use a model generated by Weka, yes? Perhaps this post can help -

@rupeshmalpani @SaleemaAmershi You are very welcome!

@rupeshmalpani @SaleemaAmershi My colleague is doing great work in this space. I encourage you to take a look.

#ChatWithMSFTResearch Signing off from live-tweeting! I will continue the conversation offline - Hoping all stay well and healthy!

@mihaela_v This is one of the ways in which AI/ML is a different material than traditional ones. One recommendation is this: we can create systems that minimize the cost of errors. Also, this opens the door for systems were design is always ongoing.

#ChatWithMSFTResearch 6/? - There are many guidelines there. - provides a great resource for UX designers to think about the right questions to ask during designs.

#ChatWithMSFTResearch 5/? As UX practitioner we need to understand why working with ML/AI is different than working with past tools/materials/technologies. Where do our traditional design principles break? Where do they hold?

#ChatWithMSFTResearch 4/? - If I were to pick just one thing, I would say that as UX practitioners we need to educate ourselves about what AI/ML "interfaces" - what they can do, what inputs it needs, and the kind of outputs it provides.

#ChatWithMSFTResearch 3/? - It is a perspectives issue - As UX designers, are we using AI as a material, a tool, or a medium? Depending on the answer, one needs to learn/know different things.

#ChatWithMSFTResearch 2/? It is a literacy issue - What does one need to know to effectively communicate with my co-collaborators? Do we speak the same language, use the same terms?

#ChatWithMSFTResearch 1/? - What resources and tools should be in any UX designer’s toolkit for working on HCAI? Working on HCAI should not be a solitary task. It is best done in concert with AI, Data, Social Sciences, Law, etc. Specialists.

@MSFTResearch Pro: systems that are as easy to interact with as it is to interact with a person. Con: systems that are as hard to interact with as it is to interact with a person. The interfaces we build set up expectations. What expectations are we creating if systems are human-like?

@PaSe_Lp It depends on the type of solution / contribution. Wobbrock and Kientz have an excellent article that unpacks the same question for HCI - I think it still applies:

@SaleemaAmershi Indeed. This is one of the open challenges of HCAI.

@SaleemaAmershi Absolutely! One of the big things HCAI makes us think about is if we *should* solve a problem with AI. Or why.

#ChatWithMSFTResearch 4/? - It then follows that a significant part of research in AI/ML *has* to be Human-Centered.

#ChatWithMSFTResearch 3/? - As researchers like Amershi et al have articulated, ML/AI is a “people problem”: People are involved at every stage of the ML pipeline, from problem definition to data collection, labeling, system evaluation, and usage.

#ChatWithMSFTResearch 2/? - When we talk about intelligent systems, or behavior, we are talking about systems that have human-like behavior. It is not only about intelligence as a human metric...