Think about a boombox that tracks your each transfer and suggests music to match your private dance model. That’s the thought behind “Be the Beat,” one in every of a number of initiatives from MIT course 4.043/4.044 (Interplay Intelligence), taught by Marcelo Coelho within the Division of Structure, that have been introduced on the thirty eighth annual NeurIPS (Neural Data Processing Methods) convention in December 2024. With over 16,000 attendees converging in Vancouver, NeurIPS is a aggressive and prestigious convention devoted to analysis and science within the area of synthetic intelligence and machine studying, and a premier venue for showcasing cutting-edge developments.
The course investigates the rising area of massive language objects, and the way synthetic intelligence will be prolonged into the bodily world. Whereas “Be the Beat” transforms the artistic prospects of dance, different scholar submissions span disciplines akin to music, storytelling, vital considering, and reminiscence, creating generative experiences and new types of human-computer interplay. Taken collectively, these initiatives illustrate a broader imaginative and prescient for synthetic intelligence: one which goes past automation to catalyze creativity, reshape training, and reimagine social interactions.
Be the Beat
“Be the Beat,” by Ethan Chang, an MIT mechanical engineering and design scholar, and Zhixing Chen, an MIT mechanical engineering and music scholar, is an AI-powered boombox that means music from a dancer’s motion. Dance has historically been guided by music all through historical past and throughout cultures, but the idea of dancing to create music isn’t explored.
“Be the Beat” creates an area for human-AI collaboration on freestyle dance, empowering dancers to rethink the normal dynamic between dance and music. It makes use of PoseNet to explain actions for a big language mannequin, enabling it to investigate dance model and question APIs to seek out music with related model, power, and tempo. Dancers interacting with the boombox reported having extra management over creative expression and described the boombox as a novel strategy to discovering dance genres and choreographing creatively.
A Thriller for You
“A Thriller for You,” by Mrinalini Singha SM ’24, a current graduate within the Artwork, Tradition, and Expertise program, and Haoheng Tang, a current graduate of the Harvard College Graduate Faculty of Design, is an academic sport designed to domesticate vital considering and fact-checking abilities in younger learners. The sport leverages a big language mannequin (LLM) and a tangible interface to create an immersive investigative expertise. Gamers act as citizen fact-checkers, responding to AI-generated “information alerts” printed by the sport interface. By inserting cartridge mixtures to immediate follow-up “information updates,” they navigate ambiguous situations, analyze proof, and weigh conflicting data to make knowledgeable choices.
This human-computer interplay expertise challenges our news-consumption habits by eliminating touchscreen interfaces, changing perpetual scrolling and skim-reading with a haptically wealthy analog system. By combining the affordances of sluggish media with new generative media, the sport promotes considerate, embodied interactions whereas equipping gamers to raised perceive and problem at this time’s polarized media panorama, the place misinformation and manipulative narratives thrive.
Memorscope
“Memorscope,” by MIT Media Lab analysis collaborator Keunwook Kim, is a tool that creates collective recollections by merging the deeply human expertise of face-to-face interplay with superior AI applied sciences. Impressed by how we use microscopes and telescopes to look at and uncover hidden and invisible particulars, Memorscope permits two customers to “look into” one another’s faces, utilizing this intimate interplay as a gateway to the creation and exploration of their shared recollections.
The system leverages AI fashions akin to OpenAI and Midjourney, introducing totally different aesthetic and emotional interpretations, which ends up in a dynamic and collective reminiscence area. This area transcends the restrictions of conventional shared albums, providing a fluid, interactive surroundings the place recollections will not be simply static snapshots however dwelling, evolving narratives, formed by the continued relationship between customers.
Narratron
“Narratron,” by Harvard Graduate Faculty of Design college students Xiying (Aria) Bao and Yubo Zhao, is an interactive projector that co-creates and co-performs youngsters’s tales by way of shadow puppetry utilizing massive language fashions. Customers can press the shutter to “seize” protagonists they need to be within the story, and it takes hand shadows (akin to animal shapes) as enter for the primary characters. The system then develops the story plot as new shadow characters are launched. The story seems by way of a projector as a backdrop for shadow puppetry whereas being narrated by way of a speaker as customers flip a crank to “play” in actual time. By combining visible, auditory, and bodily interactions in a single system, the undertaking goals to spark creativity in shadow play storytelling and allow multi-modal human-AI collaboration.
Good Syntax
“Good Syntax,” by Karyn Nakamura ’24, is a video artwork piece inspecting the syntactic logic behind movement and video. Utilizing AI to govern video fragments, the undertaking explores how the fluidity of movement and time will be simulated and reconstructed by machines. Drawing inspiration from each philosophical inquiry and creative follow, Nakamura’s work interrogates the connection between notion, know-how, and the motion that shapes our expertise of the world. By reimagining video by way of computational processes, Nakamura investigates the complexities of how machines perceive and characterize the passage of time and movement.