https://mahimapushkarna.com Mahima Pushkarna Fri, 30 Nov 2018 16:43:40 +0000 https://mahimapushkarna.com en https://mahimapushkarna.com/Slideshow Wed, 20 Jun 2018 14:44:08 +0000 Mahima Pushkarna https://mahimapushkarna.com/Slideshow I’m a designer, researcher, and technologist. I create people-first AI systems at Google DeepMind. https://mahimapushkarna.com/Featured-Projects Wed, 20 Jun 2018 14:44:09 +0000 Mahima Pushkarna https://mahimapushkarna.com/Featured-Projects What-If Tool&nbsp;︎Probe trained machine learning models interactively, with minimal code <img width="640" height="360" width_o="640" height_o="360" data-src="https://freight.cargo.site/t/original/i/bad8d0520d42dc6472304738562f42c9dd28aa924f94f63e1706fc1ec943d034/CompareCounterfactuals_720HD_lowres.gif" data-mid="41861681" border="0" src="https://freight.cargo.site/w/640/i/bad8d0520d42dc6472304738562f42c9dd28aa924f94f63e1706fc1ec943d034/CompareCounterfactuals_720HD_lowres.gif" /> Facets: Visualize large ML data&nbsp;︎Debug your data by visualizing training and testing datasets <img width="3840" height="2160" width_o="3840" height_o="2160" data-src="https://freight.cargo.site/t/original/i/879e8de03f8963864e41c49315757ad28e96de283d217553ff66b3d9c1626495/Facets_Header_Illustration2x.png" data-mid="31768335" border="0" src="https://freight.cargo.site/w/1000/i/879e8de03f8963864e41c49315757ad28e96de283d217553ff66b3d9c1626495/Facets_Header_Illustration2x.png" /> Waterfall of Meaning︎&nbsp; At AI: More than Human, the Barbican Center︎ <img width="1200" height="855" width_o="1200" height_o="855" data-src="https://freight.cargo.site/t/original/i/aca9125be76e29f0250ef1aaec17bf34a2938fa5bd03820a19533ad2e0582524/waterfall-of-meaning-google-pair-from-ai-more-than-human-barbican-centre-until-august-26-2019.jpeg" data-mid="48030899" border="0" data-scale="77" src="https://freight.cargo.site/w/1000/i/aca9125be76e29f0250ef1aaec17bf34a2938fa5bd03820a19533ad2e0582524/waterfall-of-meaning-google-pair-from-ai-more-than-human-barbican-centre-until-august-26-2019.jpeg" /> PreexperiencesDesign frameworks for timely decision-making support <img width="1280" height="800" width_o="1280" height_o="800" data-src="https://freight.cargo.site/t/original/i/81249266952690231918addd349b642a5d91b75b701082fa85f84ff43b58798a/iPhone6-White-Car-2.jpg" data-mid="31768312" border="0" src="https://freight.cargo.site/w/1000/i/81249266952690231918addd349b642a5d91b75b701082fa85f84ff43b58798a/iPhone6-White-Car-2.jpg" /> ForemostLeveraging human judgement to prioritize your pet’s safety <img width="800" height="600" width_o="800" height_o="600" data-src="https://freight.cargo.site/t/original/i/e37164904c7452ccc401fd7cd46041122b5bec486fe7ddc349794e65dff290b9/Foremost_Intro_Slide.jpg" data-mid="31768089" border="0" data-scale="89" src="https://freight.cargo.site/w/800/i/e37164904c7452ccc401fd7cd46041122b5bec486fe7ddc349794e65dff290b9/Foremost_Intro_Slide.jpg" /> https://mahimapushkarna.com/foremost Mon, 11 Jan 2016 13:54:19 +0000 Mahima Pushkarna https://mahimapushkarna.com/foremost FOREMOST is a concept for a platform that prioritizes your pet’s safety. It works as a preventative measure, that is driven by human judgement from personal networks. <img width="800" height="600" width_o="800" height_o="600" data-src="https://freight.cargo.site/t/original/i/f7ba8585a8ec12dd25e0aae85723d7d6e813355f9e149035ae91a4682aa98282/Foremost_Intro_Slide.jpg" data-mid="18768729" border="0" src="https://freight.cargo.site/w/800/i/f7ba8585a8ec12dd25e0aae85723d7d6e813355f9e149035ae91a4682aa98282/Foremost_Intro_Slide.jpg" /> Foremost aids the inclusion of unattended pets in emergency evacuations at the first instance of an emergency, such as a fire or flooding. It is centered on two pillars: relationships of trust in your close, personal circle of friends and neighbors, and&nbsp; the networks embedded within local communities. By integrating dynamic mobile technologies with emergency detection, Foremost is a solution to offer a pet parent on-ground, first-incident support in emergencies. <img width="800" height="600" width_o="800" height_o="600" data-src="https://freight.cargo.site/t/original/i/e7bf7c9a77ba87a390a10cb6d34e91d32d0ea46398daedd131e2a9def1a94f2d/Foremost_Slide2_Map.jpg" data-mid="18768730" border="0" src="https://freight.cargo.site/w/800/i/e7bf7c9a77ba87a390a10cb6d34e91d32d0ea46398daedd131e2a9def1a94f2d/Foremost_Slide2_Map.jpg" />Above: When emergency strikes, quickly identify who in your personal network is closest to your pet. <img width="800" height="600" width_o="800" height_o="600" data-src="https://freight.cargo.site/t/original/i/899511168163c34a6892851ed34e07a0809724cfd88dd3006d7da40a8ba742da/Foremost_Slide3_NewLearning.jpg" data-mid="18768731" border="0" src="https://freight.cargo.site/w/800/i/899511168163c34a6892851ed34e07a0809724cfd88dd3006d7da40a8ba742da/Foremost_Slide3_NewLearning.jpg" /> Above: Create custom profiles for each pet that can be shared with your trusted network - be it your partner, vetenarian, dog walker or neighbor. <img width="800" height="600" width_o="800" height_o="600" data-src="https://freight.cargo.site/t/original/i/963dab9da5ddd5151d10c1494b6cbb266210bb241af080152562e7e2e0d6754a/Foremost_Slide4_Alert.jpg" data-mid="18768732" border="0" src="https://freight.cargo.site/w/800/i/963dab9da5ddd5151d10c1494b6cbb266210bb241af080152562e7e2e0d6754a/Foremost_Slide4_Alert.jpg" /> Above: After making sure that it isn’t a false alarm (such as a pre-planned fire drill), activate your buddy network to ask for help. Quickly and mutually designate a responsible buddy for your pet to get him or her out of your home, and to safety.&nbsp; <img width="800" height="600" width_o="800" height_o="600" data-src="https://freight.cargo.site/t/original/i/0eacd02a62861038e25d5d8aa77f0444a36fb99ae3048ae4f7fbfa16a993bd39/Foremost_Slide5_SignUp.jpg" data-mid="18768733" border="0" src="https://freight.cargo.site/w/800/i/0eacd02a62861038e25d5d8aa77f0444a36fb99ae3048ae4f7fbfa16a993bd39/Foremost_Slide5_SignUp.jpg" /> Foremost is... +&nbsp;A preventative measure, based on human judgement +&nbsp;A first incident emergency support &amp; action system + Driven by trust in personal networks + Scalable to other emergency scenarios + Aimed at increasing success rates of evacuation &amp; drills + Aims to minimize risk to residents and response teams + Design to support wearable &amp; home hardware <img width="800" height="600" width_o="800" height_o="600" data-src="https://freight.cargo.site/t/original/i/13a6c9583f3df8390ad5eba83b2691eb3c8c34473ef020c5098272d24f1ac2b3/Foremost_Slide6_Workflow.jpg" data-mid="18768734" border="0" src="https://freight.cargo.site/w/800/i/13a6c9583f3df8390ad5eba83b2691eb3c8c34473ef020c5098272d24f1ac2b3/Foremost_Slide6_Workflow.jpg" /> Foremost was also presented as a research poster at Northeastern University's RISE (Research, Innovation and Scholarship Expo) in March 2015. <img width="3456" height="2593" width_o="3456" height_o="2593" data-src="https://freight.cargo.site/t/original/i/1fd96c2d534a86eb1ebef2fb4b71ab79c71d87c3a1ca5a96cc676bc845e82b89/2015-RISE-Poster-Horizontal_666.jpg" data-mid="18768735" border="0" src="https://freight.cargo.site/w/1000/i/1fd96c2d534a86eb1ebef2fb4b71ab79c71d87c3a1ca5a96cc676bc845e82b89/2015-RISE-Poster-Horizontal_666.jpg" /> https://mahimapushkarna.com/Monitor Wed, 19 Oct 2016 15:45:31 +0000 Mahima Pushkarna https://mahimapushkarna.com/Monitor Seems like you've reached an in-progress project. Thank you for your patience - this content will be made available shortly! Here is a sneak peek of what to expect: <img width="1280" height="800" width_o="1280" height_o="800" data-src="https://freight.cargo.site/t/original/i/1a8b188e2b09ac9c032cbc381a41aaade348f01fdd54060adb94bd4a7aba37b5/iPhone6-White-Car-2.jpg" data-mid="18768723" border="0" src="https://freight.cargo.site/w/1000/i/1a8b188e2b09ac9c032cbc381a41aaade348f01fdd54060adb94bd4a7aba37b5/iPhone6-White-Car-2.jpg" /> <img width="1280" height="800" width_o="1280" height_o="800" data-src="https://freight.cargo.site/t/original/i/9f85a9974a19c4cb79f31dd88c1f8efec1aaed9f1e87046e4df9a483727dba85/iPhone6-White-Comp-Standalone-2.jpg" data-mid="18768724" border="0" src="https://freight.cargo.site/w/1000/i/9f85a9974a19c4cb79f31dd88c1f8efec1aaed9f1e87046e4df9a483727dba85/iPhone6-White-Comp-Standalone-2.jpg" /> <img width="1280" height="800" width_o="1280" height_o="800" data-src="https://freight.cargo.site/t/original/i/0f7395879280c754798a0db24ca572ecf63ae7d2608c04eca193adebc455c43b/iPhone6-White-Comp-Standalone2-2.jpg" data-mid="18768725" border="0" src="https://freight.cargo.site/w/1000/i/0f7395879280c754798a0db24ca572ecf63ae7d2608c04eca193adebc455c43b/iPhone6-White-Comp-Standalone2-2.jpg" /> https://mahimapushkarna.com/Pre-experiences Wed, 19 Oct 2016 14:13:58 +0000 Mahima Pushkarna https://mahimapushkarna.com/Pre-experiences Pre&nbsp;experiences&nbsp;How can information and interaction design make our data more meaningful in our daily lives? A pre-experience is an ephemeral and situated information experience that offers actionable knowledge so users make better decisions in a larger, transactional experience. A pre-experience is preparatory. They “reframe” personal experiences to be more responsive to ever-changing human contexts. <img width="2304" height="2880" width_o="2304" height_o="2880" data-src="https://freight.cargo.site/t/original/i/cbbaddf8252d70c61cec5d9cc48c9770141f067002b8b80d333a9ec490f9f9db/Pre-experiences-4x.jpg" data-mid="18768726" border="0" data-scale="100" src="https://freight.cargo.site/w/1000/i/cbbaddf8252d70c61cec5d9cc48c9770141f067002b8b80d333a9ec490f9f9db/Pre-experiences-4x.jpg" />What is a pre-experience?&nbsp; Each day, the world produces 2.5 quintillion bytes of data, most of which has been supposedly created in past two years. Each of us has a quantifiable digital self, which grows each time we interact with a smart device, platform or interface. Without a meaningful context to use it in, our data remains invisible and unexplored. How can information design transform it into something more useful to us as individuals? Pre-experiences are a theoretical model for presenting actionable knowledge in a timely and contextual manner derived from a user’s data. Using this model, product designers can identify meaningful touchpoints in mundane scenarios in which users must process complex information from the real world to make decisions towards self-defined goals, for instance, staying within budget at the grocery store. A dedicated pre-experience can assist users in navigating complex information experiences that typically overlay daily experiences&nbsp;without forfeiting agency and control in decision-making. What makes a pre-experience different from a notification or alert? Like a notification or alert, pre-experiences are messages sent to users. The difference is that rather than an update, a pre-experience gives users the optimal amount of information needed to make a series of decisions. These decisions are made in larger, complex experiences, rather than a product surface or device.&nbsp; The content of a pre-experience is tailored and personalized to the “IRL” context of the user to encourage them towards a pre-determined goal or aspiration they have set - such as saving money, managing a chronic condition, etc.The “IRL” context or event in which a user&nbsp; benefits from a pre-experience typically meets the four criteria: <img width="1600" height="1614" width_o="1600" height_o="1614" data-src="https://freight.cargo.site/t/original/i/1287fcdabce575e2030e7cfff4e6c1c7f69627f9dfeb4402eb43a217fa1c5eb7/The-Understanding-Continuum-4x.jpg" data-mid="18768727" border="0" src="https://freight.cargo.site/w/1000/i/1287fcdabce575e2030e7cfff4e6c1c7f69627f9dfeb4402eb43a217fa1c5eb7/The-Understanding-Continuum-4x.jpg" /> Above: In theory, pre-experiences expand the DIKW continuum by including action and feedback, paving the way for utilization of data. 1 Routine &amp; Variation A fairly routine event, such as grocery shopping, which is influenced by natural variations in the user’s daily life. Each occurence of the event usually has different constraints and outcomes. These events are susceptible to impact from other events in the day, and conversely, can impact other events and contexts. Why Each pre-experience is tailored to individual users and resources available to them at that point in time.&nbsp; Content is updated with detected change in the user’s personal, social and environmental contexts.The message of a pre-experience is minimal, clear, and digestible-at-a-glance. Responsive and optimized content speaks neutrally to the user’s value system, so they feel the ownership to their decisions and the benefits they experience. In Context: Grocery Shopping Sources of variation in grocery shopping can come from the products that users need to purchase per visit, or a visit to an alternate store with different prices. Other factors that influence such a scenario could range from available cash to a hosting a dinner on thanksgiving. Despite these variations, users must routinely go grocery shopping. A financial pre-experience would change each time based on the shopping list or the location at which users are shopping. 2 Decisive Action &amp; Continuous ImprovementAn event in which users must repeatedly make a set of decisions and act upon them. These decision-action pairs cannot be automated – they require making value judgements. However, each decision offers an opportunity to continuously improve or optimize towards a goal, without the need for any external regulation. Why Pre-experiences are timed to precede experiences so users can utilize actionable knowledge from their own data in decision making. PEs enable affordances of an experience that work with the user’s motivations. Users focus on engaging with the experience, independently making decisions while being aware of critical information that affect the output of the experience. The assimilation of actionable knowledge in an experience provides grounds for hands-on learning. In Context: Grocery ShoppingWhile the end goal (purchase all items on the shopping list) is fixed, a user could be optimizing for healthier food choices or increased savings, each time they decide between competing products.If the user were optimizing for savings, then a pre-experience would constitute a budget customized to that specific grocery run; this number could then assist users in approximating the total cost when deciding optimal products to purchase. 3Output &amp; Influence The outcome(s) from an event affect the circumstances of users. Cycles of feedback between users, events and pre-experiences are created each time users act on decisions. Actions in the real world affect the overall “system state” of the user and the platform, which tune pre-experiences for the next such event.Why Pre-experiences encourage users to discover and try new behaviors in repeating events, that respond to the immediate context without upsetting the larger picture. This builds resilient behaviors in a variable experience. As an affinity for a particular iteration of behavior grows, self-incentivized habits biased towards positive deviance are formed.In Context: Grocery ShoppingA example of trade-offs a user makes when shopping for groceries is money spent vs quality vs quantity. Each of these have a direct impact on the user’s life, and can influence the decisions they make the next time they go grocery shopping.4Data Streams &amp; Information Environments The environment of the event must be connected and produce rich, useful data. Events should generate multi-dimensional data which can then be mapped to useful signals used to tune and deliver pre-experiences.Why To derive good and healthy actionable knowledge from personal data, the algorithm that creates a pre-experience must rely on multiple signals. Actionable knowledge or knowledge that can be ‘acted upon’, is created from the convergence of multiple streams of data, enabled by the internet of things and a personal family of devices that have permissions and features that support tasks related to the premise for the pre-experience. In Context: Grocery Shopping Grocery shopping creates a variety of digital artefacts, some that are incredibly private and must be treated accordingly. For a pre-experience that focuses on savings or maintaining budgets, tasks and shopping lists and prior reciepts from grocery stores are digital artefacts of importance. Personal and protected information streams in this case would include budgeting apps, location tracking services, and spending data. Explore Monitor, a concept for an intelligent finance management and budgeting application which uses pre-experiences.This thesis was submitted to Northeastern University, Boston, MA in May 2016 towards the Master in Fine Arts in Information Design &amp; Visualization program. Read the complete thesis as a PDF or a poster. https://mahimapushkarna.com/About Fri, 30 Nov 2018 16:43:40 +0000 Mahima Pushkarna https://mahimapushkarna.com/About Mahima Pushkarna is a design lead at the People + AI Research Initiative at Google. She brings design thinking and human-centered design into Human-AI Research, and has a rich history of leading product design &amp; strategy projects in emerging technologies of large scope. Mahima has designed tools and frameworks that make complex information useful and understandable and even fun – such as visualizing interpretability results for Machine Learning models and creating controls for Generative AI.&nbsp;These have been widely used to advance better development practices by companies like Huggingface, Google, Disney, Yahoo! and GoPro, as well as in academia by MIT and Harvard University. Mahima believes that design can be a powerful tool for understanding and addressing the needs of people who are impacted by technology. Her work draws from a mix of human-centered, participatory, and speculative design practices to bridge the gap between upstream developer practices and their impact on end user experiences and society. Mahima is also interested in exploring the intersection of design, technology, and society, and is always looking for new ways to use design to make the world a better place. Prior to Google, Mahima worked as a product designer at Innovation by Design, a global think-tank, consulted at MIT's Design Lab, and designed visualization tools at Ion Interactive. Sometimes she writes on her&nbsp;Personal Blog and Google DesignOccasionally she speaks or runs workshops. Reach out here for speaking. Articles &amp; Videos The Data Cards Playbook: A Toolkit for Transparency in Dataset Documentation Article with Andrew Zaldivar, for the Google AI Blog. 2022 Mahima Pushkarna is making data easier to understand Interview, The Keyword Blog, Google, 2022 Future Of Canada: Growth &amp; Recovery Panel on AI, Globe and Mail, 2022 Participatory ML: Using PAIR Tools: What-If and Tensorflow.js Talk,&nbsp;People + AI Research Symposium, Google, London, 2019.&nbsp; How UX changes the world, One AI at a Time Panel, UXPA Boston Annual Conference, 2019. Through the Looking Glass&nbsp;World IA Day Boston 2019, Massachusetts College of Art &amp; Design. 2019. The What-If Tool Talk, Google Developers Summit, Cambridge MA , 2019 Six AI Terms UXers Should Know Article written with Reena Jana, for Google Design. 2018 Learning Machine Learning: Implications for Design Talk, UXPA Boston 17th Annual User Experience Conference, 2018.&nbsp;Forbes Coverage Machine Learning, Implications for DesignInvited Talk, Northeastern University, College of Arts, Media and Design. 2018. Research &amp; Publications &nbsp; &nbsp; Pushkarna, M., Zaldivar, A. και Kjartansson, O. (2022) ‘Data Cards: Purposeful and Transparent Dataset Documentation for Responsible AI’, στο 2022 ACM Conference on Fairness, Accountability, and Transparency. New York, NY, USA: Association for Computing Machinery (FAccT ’22), σσ. 1776–1826. doi: 10.1145/3531146.3533231. &nbsp; &nbsp; Pushkarna, M. and Zaldivar, A., 2021. Data Cards: Purposeful and Transparent Documentation for Responsible AI. In 35th Conference on Neural Information Processing Systems (pp. 1776-1826). &nbsp; &nbsp; Tenney, I., Wexler, J., Bastings, J., Bolukbasi, T., Coenen, A., Gehrmann, S., Jiang, E., Pushkarna, M., Radebaugh, C., Reif, E. and Yuan, A., 2020. The language interpretability tool: Extensible, interactive visualizations and analysis for NLP models. arXiv preprint arXiv:2008.05122.&nbsp; &nbsp; Ghassemi, M., Pushkarna, M., Wexler, J., Johnson, J. and Varghese, P., 2018. Clinicalvis: Supporting clinical task-focused design evaluation. arXiv preprint arXiv:1810.05798. &nbsp; &nbsp; IEEE VIS 2019 -&nbsp; Wexler, J., Pushkarna, M., Bolukbasi, T., Wattenberg, M., Viegas, F., &amp; Wilson, J. (2019). The What-If Tool: Interactive Probing of Machine Learning Models. arXiv preprint arXiv:1907.04135.&nbsp; Recording of presentation by James Wexler. &nbsp; The What-If Tool: Code-free probing of machine learning models for fairness and interpretability, Workshop, ComputeFest 2019, Harvard University, January 2019