Thesis: S.M., Massachusetts Institute of Technology, School of Architecture and Planning, Program in Media Arts and Sciences, 2019 / Cataloged from PDF version of thesis. / Includes bibliographical references (pages 41-43). / This thesis lies in the intersection of three avenues: physical non-screen interfaces, storytelling promoting community connectedness and hyper-locality afforded by decentralization. The central question addressed is whether the design and underlying technology of entry points to a network change the way people interact with it and the experience they have. To explore this, I designed and engineered a set of playful physical objects which function as nodes of a hyper-local network. Information bestowed upon this network remains within these nodes, cryptographically secure, and accessible only to local community members who are aware of the network's existence and mode of operation. I successfully deployed this network in four locations across the MIT campus, where members of the MIT community could record and listen to audio messages using the physical objects. Given the choice, people engaged with the decentralized, closed network more than they did with the control (open) one. The messages collected by the former had a more positive sentiment score and their language was more personal than the control. This work is indicative, but not definitive, evidence that the suggested decentralized and closed network fostered a more positive and expressive discourse than its control counterpart. / by Kallirroi Retzepi. / S.M. / S.M. Massachusetts Institute of Technology, School of Architecture and Planning, Program in Media Arts and Sciences

Thesis: S.M., Massachusetts Institute of Technology, School of Architecture and Planning, Program in Media Arts and Sciences, 2019 / Cataloged from PDF version of thesis. / Includes bibliographical references (pages 111-118). / Talk radio exerts significant influence on the political and social dynamics of the United States, but labor-intensive data collection and curation processes have prevented previous works from analyzing its content at scale. Over the past year, the Laboratory for Social Machines and Cortico have created an ingest system to record and automatically transcribe audio from more than 150 public talk radio stations across the country. Using the outputs from this ingest, I introduce "hierarchical compression" for neural unsupervised summarization of spoken opinion in conversational dialogue. By relying on an unsupervised framework that obviates the need for labeled data, the summarization task becomes largely agnostic to human input beyond necessary decisions regarding model architecture, input data, and output length. Trained models are thus able to automatically identify and summarize opinion in a dynamic fashion, which is noted in relevant literature as one of the most significant obstacles to fully unlocking talk radio as a data source for linguistic, ethnographic, and political analysis. To evaluate model performance, I create a novel spoken opinion summarization dataset consisting of compressed versions of "representative," opinion-containing utterances extracted from a hand-curated and crowd-source-annotated dataset of 275 snippets. I use this evaluation dataset to show that my model quantitatively outperforms strong rule- and graph-based unsupervised baselines on ROUGE and METEOR while qualitatively demonstrating fluency and information retention according to human judges. Additional analyses of model outputs show that many improvements are still yet to be made to this model, thus laying the ground for its use in important future work such as characterizing the linguistic structure of spoken opinion "in the wild." / by Shayne O'Brien. / S.M. / S.M. Massachusetts Institute of Technology, School of Architecture and Planning, Program in Media Arts and Sciences

Thesis: S.M., Massachusetts Institute of Technology, School of Architecture and Planning, Program in Media Arts and Sciences, 2019 / Cataloged from PDF of thesis. / Includes bibliographical references (pages 78-80). / Various studies have demonstrated that privately owned cars will become significantly less prevalent in the city in the next 10 to 15 years. Other efficient alternatives for mobility platforms within the city are in demand around the world. One example is the emergence of the PEV (Persuasive Electric Vehicle), an agile autonomous bike-sharing platform (M Lin, 2015). Based on this trend, it is reasonable to anticipate that increasingly more mobility systems of different forms will emerge in urban areas in the future. These new mobility systems might not necessarily be similar to cars; they may instead be a new class of social robot that could blend into the city more seamlessly. Moreover, when there is no longer a driver within each vehicle, designing human-machine interface (HMI) that is simple for users to process will be more important than ever. / For example, if a pedestrian encounters a lightweight autonomous vehicle for which it is apparent that no one is in the vehicle, how can the pedestrian understand the intention of the vehicle? And how can we, as designers, make this more intuitive and seamless? This thesis presents IDK, which is an Interaction Development Kit equipped with essential tools to help facilitate the design and prototyping process. IDK could be physically installed in PEVs, thereby enabling designers and developers to prototype human-machine interactions in a rapid and intuitive manner. This thesis also identifies multiple situations that a lightweight autonomous vehicle may encounter while navigating through streets and proposes a range of interactions that can tackle these problems. All prototypes from this thesis are based on the latest version of the PEV as an interactive platform. / The proposed interactions are evaluated through outdoor testing as well as indoor exhibitions to determine how people respond to these new norms of communication. My hope is that the results of this thesis will provide useful insights for designers and developers who seek to develop interactions that allow humans to seamlessly interact with lightweight autonomous vehicles. / by Jerry Wei-Hua Yao. / S.M. / S.M. Massachusetts Institute of Technology, School of Architecture and Planning, Program in Media Arts and Sciences

Thesis: S.M., Massachusetts Institute of Technology, School of Architecture and Planning, Program in Media Arts and Sciences, February, 2020 / Cataloged from student-submitted PDF version of thesis. / Includes bibliographical references (pages 60-69). / The following thesis lies where mindfulness and technology meet, in particular, in the design, implementation, and testing of HeartBit. HeartBit is a device designed for haptic heart rate biofeedback. A handheld heart beats alongside your own, mirroring the size, weight, and movement of a hidden internal organ, now external and tangible, in real-time. HeartBit offers a medium for users to self-regulate in moments of stress, anxiety, or exertion: Control your heart to control your breath and body--for relaxation, performance enhancement, or augmented self-awareness. In partnership between the MIT Media Lab and the Ellen Langer Mindfulness Lab at Harvard, HeartBit has been tested under two scenarios. First, showing how people can learn to voluntarily control their heart rate using mindfulness, assisted by portable technology to accelerate the learning process. Second, showing how empathy towards unlike-minded individuals can be increased simply by feeling someone else's heartbeat in the palm of one's hand. / by Oscar Rosello. / S.M. / S.M. Massachusetts Institute of Technology, School of Architecture and Planning, Program in Media Arts and Sciences

Thesis: S.M., Massachusetts Institute of Technology, School of Architecture and Planning, Program in Media Arts and Sciences, February, 2020 / Cataloged from student-submitted PDF version of thesis. / Includes bibliographical references (pages 70-71). / Biology creates assemblies with orders of magnitude more parts than any known human designed process. Molecular biology functions on the premise that fundamental building blocks assemble into chains, are zipped into strands and folded into structures. This thesis is a macroscale implementation that aims to do the same, assemble, zip and fold, in an inorganic system. This system, Zipped, utilizes distributed coalescence of parts, aiming for faster assembly while incorporating error correction into the fabrication process. This thesis presents a design for 0-dimensional building blocks that snap together to form 1-dimensional strands. Strands zip together, interlocking to form 2-dimensional beams that can branch and merge to create patterns or flat sheets. Strands can zip to each other out of plane as well, allowing 3-dimensional construction. / All steps of the construction process are reversible; parts can be assembled, dis-assembled and re-assembled without damage to the part or altering structural performance. No energy, formwork or pre-load is required to maintain the parts position once it is assembled. The system can assemble rigid as well as flexural elements, including chains and revolute joints. Increased stiffness or flexibility can be designed into structures by changing strand geometry and zipping. This ability to tune local structural properties allows actuators to be added to the construction system and form mechanisms. Zipped pieces are demonstrated as the structural element for a robot's body, which can locomote on itself or foreign terrain. Initial studies also demonstrate automated construction with this system. The fundamental principles of this system are demonstrated in many materials, via different manufacturing processes and across several scales, showing applicability to a diverse scenario space. / For ease of fabrication and lab use a centimeter scale part was selected and several thousand parts were manufactured. This 0-dimensional part is presented and used to form larger scale assemblies which are mechanically characterized. From here, mission architectures and real-world applications are described. The Zipped system enables human-scale, controlled and reversible assembly, zipping and folding. This allows reusability, reconfigurability and universality - attributes we often credit to nature. / by Sara Falcone. / S.M. / S.M. Massachusetts Institute of Technology, School of Architecture and Planning, Program in Media Arts and Sciences

Thesis: Ph. D., Massachusetts Institute of Technology, School of Architecture and Planning, Program in Media Arts and Sciences, February, 2020 / Cataloged from student-submitted PDF of thesis. / Includes bibliographical references (pages 277-285). / Fifty-two years ago, Sergeant Pepper's Lonely Hearts Club Band redefined what it meant to make a record album. The Beatles revolutionized the recording process using technology to achieve completely unprecedented sounds and arrangements. Until then, popular music recordings were simply faithful reproductions of a live performance. Over the past fifty years, recording and production techniques have advanced so far that another challenge has arisen: it is now very difficult for performing artists to give a live performance that has the same impact, complexity and nuance as a produced studio recording. Live performance production technology is now used almost exclusively to recreate studio albums exactly as they were recorded. Recently, this approach has dominated the entertainment industry. In an attempt to reach superhuman levels of perfection and complexity, many elements that make live performances emotionally meaningful for audiences have been given less priority -- / or lost altogether. The mission of the work described in this dissertation is to reverse this trend by investigating methods of integrating technology and live music performance in such a way that the technology allows for flexible musical expression, sound and connection to the audience, while still enabling exciting, sophisticated and "magical" production values. This dissertation identifies six objectives for the human-centric design and integration of technology in musical performance, and a methodology to support each objective. These have been developed, refined and tested with artists and performers through a series of ten large-scale projects and approximately 300 individual performances. Through this work, I demonstrate that it is possible to combine high-value production with interactive musical performance. / We are now on the cusp of redefining live musical performance production as an art form just as Sergeant Pepper's Lonely Hearts Club Band redefined studio album production as an art form fifty years ago. / by Benjamin Arthur Philips Bloomberg. / Ph. D. / Ph.D. Massachusetts Institute of Technology, School of Architecture and Planning, Program in Media Arts and Sciences

Thesis: Ph. D., Massachusetts Institute of Technology, School of Architecture and Planning, Program in Media Arts and Sciences, February, 2020 / Cataloged from student-submitted PDF of thesis. "February 2020." / Includes bibliographical references (pages 309-342). / Social learning is a crucial component of human intelligence, allowing us to rapidly adapt to new scenarios, learn new tasks, and communicate knowledge that can be built on by others. This dissertation argues that the ability of artificial intelligence to learn, adapt, and generalize to new environments can be enhanced by mechanisms that allow for social learning. I propose several novel deep- and reinforcement-learning methods that improve the social and affective capabilities of artificial intelligence (AI), through social learning both from humans and from other AI agents. First, I show how AI agents can learn from the causal influence of their actions on other agents, leading to enhanced coordination and communication in multi-agent reinforcement learning. Second, I investigate learning socially from humans, using non-verbal and implicit affective signals such as facial expressions and sentiment. / This ability to optimize for human satisfaction through sensing implicit social cues can enhance human-AI interaction, and guide AI systems to take actions aligned with human preferences. Learning from human interaction with reinforcement learning, however, may require dealing with sparse, off-policy data, without the ability to explore online in the environment - a situation that is inherent to safety-critical, real-world systems that must be tested before being deployed. I present several techniques that enable learning effectively in this challenging setting. Experiments deploying these models to interact with humans reveal that learning from implicit, affective signals is more effective than relying on humans to provide manual labels of their preferences, a task that is cumbersome and time-consuming. However, learning from humans' affective cues requires recognizing them first. / In the third part of this thesis, I present several machine learning methods for automatically interpreting human data and recognizing affective and social signals such as stress, happiness, and conversational rapport. I show that personalizing such models using multi-task learning achieves large performance gains in predicting highly individualistic outcomes like human happiness. Together, these techniques create a framework for building socially and emotionally intelligent AI agents that can flexibly learn from each other and from humans. / by Natasha Jaques. / Ph. D. / Ph.D. Massachusetts Institute of Technology, School of Architecture and Planning, Program in Media Arts and Sciences

This electronic version was submitted by the student author. The certified thesis is available in the Institute Archives and Special Collections. / Thesis: Ph. D., Massachusetts Institute of Technology, School of Architecture and Planning, Program in Media Arts and Sciences, 2019 / Cataloged from student-submitted PDF version of thesis. / Includes bibliographical references (pages 107-116). / The ability to predict the success or failure of an early stage company is critical for accelerator programs and investors. Prior studies marked human and social capital as important factors determining the potential of a startup to succeed. However, very little is known about the effect that founders' interpersonal relationships have on the success of their companies, and the effect of their relationships with other startups located in the same innovation space on their performance. To investigate these relationships, we propose a combination of methodology and field experiments that make use of Rhythm, a wearable sensing platform designed for measuring social interaction. We first describe the design of the platform and its evaluation process. Then, we describe a large-scale field study in a university startup accelerator program and its results. Finally, we propose future enhancements to the platform and directions for future research. / Financial support from the Kauffman Foundation / by Oren Lederman. / Ph. D. / Ph.D. Massachusetts Institute of Technology, School of Architecture and Planning, Program in Media Arts and Sciences

Thesis: S.M., Massachusetts Institute of Technology, School of Architecture and Planning, Program in Media Arts and Sciences, 2018 / Cataloged from PDF version of thesis. / Includes bibliographical references (pages 101-102). / Participating in online debate can expose people to diverse viewpoints, and thereby reduce polarization of opinion over controversial issues. However a lot of online debate is hostile and further dividing -we need tools that facilitate meaningful back and forth discussion. For my thesis work I created such a tool in the form of an artificial agent that engages users in debate over controversial issues. By engaging in debates with many users, the agent will start to gain insight into things like: what kinds of arguments do people find persuasive? Or, what can we predict about a person's argumentative behavior from their moral sense? Or, what is the characteristic debate path for someone who becomes persuaded to change his mind completely? The agent will then use what it has learned to help users on either side of an issue better understand each other by exposing them to compelling arguments from both sides. To identify these arguments, the agent develops a model of the user that predicts which arguments the user will like. I measure the agent's performance given different models of the user. I then evaluate the performance of each model against the random agent, which does not attempt to model the user. / by An Yuan. / S.M. / S.M. Massachusetts Institute of Technology, School of Architecture and Planning, Program in Media Arts and Sciences

Thesis: Ph. D., Massachusetts Institute of Technology, School of Architecture and Planning, Program in Media Arts and Sciences, May, 2020 / Cataloged from student-submitted PDF of thesis. / Includes bibliographical references (pages 111-113). / Target tracking is necessary across a wide range of disciplines and scales, such as in monitoring tissues and cells, beam bending, fluid dynamics, human-computer interaction, and traffic. Due to these widespread applications, advances in target tracking drive cascades of new medical, social, and scientific capabilities. In particular, this dissertation advances magnetomicrometry, a technology that tracks visually-obscured magnetic beads implanted within biological tissue to monitor in-vivo tissue length and speed within freely moving animals and humans. There are many methods to track visually-obscured objects, but magnetic-target tracking has the advantages of being low-cost, portable, and safe. However, current magnet tracking technologies are slow, precluding high-speed real-time magnetic-target tracking. This is due to the mathematics of magnet tracking, whereby magnet positions are traditionally determined via numerical optimization, suffering from instability and significant delays. This dissertation develops the mathematics for an improved method to track one or more magnets with high speed and accuracy and validates this method by demonstrating real-time muscle length tracking. We develop a high-speed, real-time, multiple-magnetic-target tracking method using the analytic gradient of the magnetic field prediction error. We extend this method to compensate for magnetic disturbances in real time using a simpler, more portable strategy than currently-published disturbance compensation methods. Validating our method in a physical system against state-of-the-art motion capture, we demonstrate increased maximum bandwidths of 336%, 525%, 635%, and 773% for the simultaneous tracking of 1, 2, 3, and 4 magnets, respectively, with tracking accuracy comparable to state-of-the-art magnet tracking. / by Cameron Roy Taylor. / Ph. D. / Ph.D. Massachusetts Institute of Technology, School of Architecture and Planning, Program in Media Arts and Sciences