# STA's Guide ## Welcome aboard Congrats on your STA position and welcome to CSCI 1430 (Computer Vision) and to Brown's growing SRC program! We hope you learn a lot this semester and wish you the best of luck in teaching and guiding students with SRC perspectives. Spring 2023 -> Spring 2024 (General Advice) ------------------------ **Fiona:** 1. Be part of the discussion. Don’t spoon-feed students your beliefs - prompt them to challenge their own and each other’s beliefs, and make them feel comfortable enough to come to terms with their own stances. 2. Don’t be afraid to think outside of the box. Students actually seem to respond better through creative engagements, so find ways that can keep them talking! 3. Enjoy the work you do! It sounds extremely cliche, but you applied and were chosen because you care; your role stands at the frontline of students’ exposure to these topics. It can be intimidating for students and for yourself because these have some weighted implications, but your openmindedness passion for the work you’re doing is going to lead by example. **Melvin:** 1. Communication is key. Don't be afraid to speak up and ask for help. Your co-STA is your best friend. Your fellow HTAs, UTAs, and the course instructor are all there to support you. You got this! 2. Engage with and take advantage of Brown's SRC program. You will have an excellent opportunity to collaborate with the SRC department and be able to work personally with HSTAs and the professor leading the program. Be sure to listen and share your ideas. You may find great content to add or know what to avoid as an STA! 3. Have fun in discussion sections and try to get students thinking. You may have to adapt on the fly, but this is a great way to boost your confidence and public speaking skills. **James:** 1. The class ethos is big and broad (but not deep). We have students from sophomores to Masters, and we cover everything from image filtering to deep nets. The class tries to be welcoming without pandering --- it is still a 1000-level class. Many of the perceived problems or tensions come from these core constraints: people don't know the prereqs (math or a 1000-level familiarity with general CS systems), and they need them because we move fast. 2. 1430 is now a split two-semester course, which requires care and communication between the teams due to changes, as there's an institutional knowledge gap. Keep everything on Github; no Google Docs, as these are lost when students leave. Do not put dates on anything; it just adds work to update them. I/we may fail to document things; if so, I am here to let you know _why_ things were done like that, so that we don't repeat mistakes or undo progress. Please ask me. 3. A course is like a ship in the ocean---changing direction too fast against the waves risks capsizing. The waves are student expectation and TA knowledge, which are set from the previous year's course. So, from semester to semester, pick _one big thing only_ to change and make better (or even none!). It takes about 3 course iterations to make any one change good. That means changes to infrastructure, environments, questions, code, etc. Everything you change affects something you didn't anticipate, and causes bumps for students that harm learning. The default mode should be 'if it ain't broke, don't fix it'. 4. Minimize all busywork. Think carefully about how everything will be communicated, put out, handed in, and assessed to streamline all processes. Whenever anyone (TA, instructor) proposes something new, your goal is to consider whether it creates _work_ and, if it does, what the value of that work is wrt. learning objectives. If it's low, don't do it. 5. Fight bloat. Every course 'aspect' is another thing to maintain and keep up to date (and cause bugs). A better course _reduces its surface area while keeping the same learning content_ <- INTERNALIZE THIS. Do not add anything without taking something away. The temptation is to ask TAs to _add_ all sorts of things, but a better activity is to critically assess what is there, study why students tripped up, and remove/restructure things to clarify and simplify the task through exposing only what is essential. 6. SRC content: Make sure that the course staff does not impose its values on the students. No question should lead students in any way; there is no expected answer. The goals are awareness and critical reasoning, not projecting our own opinions. Pose genuine dilemmas and reject inauthentic, trivial, or obvious situations. ## Discussion Sections ### Overview Discussion Sections will likely be the bulk of your work as an STA for Computer Vision. This semester, we created three new dynamic 30 minute discussion sections worth 5% of students' grades. Each were led by the course instructor and one STA. We scheduled them from 6-9 pm on Mondays and Tuesdays in the CIT. This amounted to 6 thirty minute back-to-back sections on each day, for a total of 12 of these. We split the workload so that one TA would lead 6 on Monday and the other would lead the other 6 on Tuesday. You can change the dates to work better for your time schedule if needed, but this general structure was very effective. Holding back-to-back discussion sections is tiring work! If the professor teaching the current iteration of the course isn't available to help you out during the sections, have an HTA or UTA help assist you for crowd control each section. If there is only one STA in the course somehow, have multiple assist you. **Don't overwork yourselves. Ask for help.** ### Past materials Past discussion sections slides, alternate assignment forms (3 of them), and an example attendance sheet are all available to you within the folder CS143 under Course-Specific Materials inside an STAs drive that should be available to you at SRC camp. If this isn't available to you or you don't have access, please reach out to the HSTAs, Professor Julia Netter, or whoever is running the SRC department at the time. An SRC_PLAN.md file consisting of some of the ideas we brainstormed at the beginning of the semester should also be available to you under CourseOrganization in this github repo. We also tried using Qualtrics for poll feedback visuals this semester. The visuals were nice, but sometimes loading took a while in the middle of the discussion section. If you want to use Qualtrics software, be sure to familiarize yourself with it to minimize this latency. Otherwise, something like Google Forms could get the job done as well. ### An Effective Example [Example Discussion Section Link](https://docs.google.com/presentation/d/1fHH19V1R9FaWWxPvvpY6Ok9qlR3fV3_kDz2liKkwSLo/edit#slide=id.p) In one of the three interactive discussion sections we designed and led this semester, students were initially tasked with a scenario that places them directly into the role of working as a CV engineer. In this initial scenario, the supposed engineer discovers a revealing shot of a person on a toilet captured by the home robotics company (e.g., iRobot) despite the fact that the company claims that cameras used don’t collect PII (personally identifiable information) data. Students were then tasked to discuss what courses of action they would take should they find illegal/unethical data in industry (talk to team, report internally, delete the data personally, talk to lawyer, tell the media, quit the company, take collective action). Following this elaborate discussion, we add a layer of complexity to the scenario: “It is discovered that an opt-in beta development program with home testers had robot images sent to a third-party AI labeling company. They employed gig-workers in an emerging economy country. How did this happen, and what would you do?” This added detail then leads to a stakeholder discussion in which each student has to then defend their roles (argue why they are not responsible for the situation) with the roles being: Computer Vision engineer, Third-party AI labeling company, Gig labeling worker, Robot vacuum manufacturing, Social Media moderation system (human + AI), and the home beta tester. Following this dynamic argument, students were then polled anonymously on which stakeholder they actually feel is more responsible (ordering each by % of accountability). Afterwards, students were shown visually the variability in perspectives amongst their own 12-person discussion section and amongst the whole 150+ CV course. Next, we add an additional layer of complexity. This happened in real-life: “iRobot manufactures vacuum cleaners that use computer vision for navigation. iRobot claims the cameras used do not collect personal identifying information (PII). However, a revealing shot of a young woman in a lavender T-shirt sitting on a toilet was captured by the development version of iRobot’s Roomba J7 series. Images were sent to Scale AI, a startup that crowdsources workers to label audio, photo, and video data to train AI. Low-paid gig worker in Venezuela broke Scale AI’s policy and posted the images online. The photos were posted in a private group on Facebook; these photos were then sent to MIT Technology Review.” Finally, we inform an ultimate layer of complexity: “Amazon just announced in 2022 that it will acquire robot vacuum maker iRobot for $1.7 billion. The deal is currently [1 month ago] under review & investigation by the FTC and EU Antitrust…” Students responded really well to the carefully designed shift towards engaging SRC content that we incorporated throughout the semester. #### TLDR: Note in the example above: - Students’ time is valued. No busy work. Crisp, beefy 30 minute sections (3 of them). - In-person interactive discussions in a small classroom setting (12-14 person) - Students placed in scenarios they could picture themselves in (Computer Vision engineer, Amazon SWE, AI algorithm designer, beta tester, etc) - Group-think is avoided (students given specific role to defend; sometimes when you throw a question in the crowd, one person comes up with an answer in the air and everyone just agrees with them even if they wouldn't have individually) - Diversity of perspectives are shown visually (Qualtrics anonymous polling) - Novel and relevant materials are introduced: PII, ethics reporting, journalism, C2PA, beta-testing, crowdsourced data labeling, B2B, M&As, FTC, EU Antitrust, robotic surveillance, digital privacy, and accountability. ### Improving Discussion Sections Material: Feel free to touch up or revamp the existing material. Reread advice 3, 4, 5, 6 from **James** at the top of this page. You may have to improvise during discussion sections to keep students engaged. Also an additional note, from student feedback during our course, they generally ranked Discussion 2 > Discussion 3 > Discussion 1. ## Written SRC Assignments We only made a few minor changes here this semester. You shouldn't have to do much additional work other than small tweaks or at most, change one or two questions/articles. ## Grading Grading for this course is primarily based on completion and effort. Occasionaly, you can flag a response if it's particularly good or leave feedback on Gradescope. HTAs and the instructor may give you more instructions, but this should be straightforward. ## STA Proposal Swap Critique The content is already made. You don't need to change anything. The other UTAs will likely handle the swapping portion of emailing back the corresponding proposal. Grading the SRC content on the final project should be essentially the same as for the homework.