AI and Data Privacy in Slack
TL;DR - Slack AI is a feature within Slack that enhances productivity through AI-powered search, summarization, and recaps directly within Slack. Due to product accessibility challenges, my team and I decided to take the speculative road and solved for potential scenarios where Slack AI could mess up. This included activities like 'Black Mirror' brainstorming, competitor analysis, paper prototyping and testing, and A LOT of iterations.
01
Overview
Timeline
13 weeks - Jan 2024 - Apr 2024
Target Audience
Students who use Slack in the educational context
Team
Maliha Hashmi, Snehashish De, Aniruddha Murthy, Kaushik JV, Biheng Gao, Li-Hsuan, and Harshika Rawal
My Contributions
Helped narrow down the project scope to one product - Slack, contributed to secondary research, led and facilitated team meetings, contributed to core AI scenarios as well as the UI Design & Prototyping of the same. Lastly, contributed to the handover document.
02
Problem Statement
How do we enable user trust in AI w.r.t. data privacy
Why do we need to solve for this though?
Meet De,
A jolly, hard-working grad student.
Can you guess what happened when he engaged in a fun banter with a professor?
AI presented him in the wrong light
When asked to summarize the chat, AI made it seem like De had hurt a prof with remarks about their old age.
AI Summary
Narrowing Project Scope
Product Choice
We chose to work on Slack AI as our choice of product to narrow down the scope of the project and based it in the educational context as it was an accessible user group for us. Given the nature of Salesforce's usual user base - B2B - it would be a big challenge to conduct primary research, hence a product with a more reachable user base.
by
Salesforce
Product Choice
About Slack AI
In Context
AI Concerns
But
Challenge
Product Access
Slack AI was the latest feature launched by Slack and due to the nature of the project we did not have access to the product just yet since it was also not available to all Slack users in the market, only to limited users with specific plans and subscriptions.
IDEATION
'Black-mirror' Brainstorming
What did this lead to?
Low-Fi Concept Prototypes
On the basis of these scenarios we sketched out certain interactions within slack
Wizard of Oz testing
Our goal was to evaluate user trust and gather insights for improvement with paper prototypes. We focused on potential Slack users who have not extensively interacted with Slack AI.
Major Findings
We finally narrowed down to
We didn't solve for the positive scenario. So why is it listed? We wanted to emphasize on the fact that AI is not all bad, in our "what if" scenarios, there's room for the good that AI brings
In-Line Privacy Alert
AI would recognize any P.I.I violation and have a pop up in-line alert
Being Transparent About AI's Bias Moderation
AI might flag political comments/conversations and there is a need for transparency in doing so. The "view details" section talks about it in detail
Misinterpretation of Tone in Conversation Summaries
AI might misinterpret tone of conversation and give a whole new meaning to the converastion when summarizing it for another team member. A disclaimer about this is important. "View Details" elaborates on that
Solution and Final Screens
04
Result
The concept solutions are designed in a way that tackles major user issues in trusting AI with conversations or interactions with other teammates. It was designed using the Slack design system.