For half a century, MIT students have embraced the Independent Activities Period (IAP), a distinctive mini-semester spanning just four weeks, strategically positioned between fall and spring terms. This year, despite the pandemic necessitating a virtual format, the essence of innovative exploration and creative discovery remained unshaken.
Embodying this innovative spirit, the Department of Electrical Engineering and Computer Science (EECS) presented diverse IAP offerings ranging from entertaining to intellectually profound.
Many MIT learners leverage the condensed IAP timeframe to explore subjects beyond their academic comfort zones. The "Deep Learning For Art, Aesthetics, and Creativity" course was specifically designed to facilitate this type of intellectual adventure. Course instructor Ali Jahanian draws inspiration from his artistic background: "My foundation in design, painting, and visual arts has continuously fueled my creative endeavors. In my research, I've focused on understanding and quantifying aesthetics and design, but following my doctoral studies, I became increasingly captivated by the fascinating concept of learning through creation," he explains. "While many colleagues concentrate on understanding intelligence, I'm particularly fascinated by the creativity and innovation aspects of human cognition. For example, how can we extrapolate from existing data to generate novel content beyond the scope of training datasets?"
Jahanian's methodology appeals not merely to computer scientists but to anyone intrigued by AI's artistic potential. "The course objective is to help students comprehend how we can harness AI for creative purposes, and how creativity can enhance AI development," Jahanian states. "Students first need to grasp the current landscape of AI and creativity in 2021, then recognize the limitations of these concepts and explore ways to expand those boundaries." He views artificial intelligence application as intrinsically creative: "AI captivates me because it mirrors our essence. We possess an innate drive to replicate and recreate ourselves, which explains why artists derive fulfillment from creation."
The potential applications of creative AI—or systems capable of predicting human aesthetic preferences—are extensive. "We can certainly learn and quantify our preferences. Such quantitative algorithms have broad implications for understanding emotions and feelings," Jahanian notes. "Consider a home assistant robot: you'd want it to comprehend your desires, preferences, and personal taste." More immediately, Jahanian believes his students benefit from art's tangible nature as an educational tool. "I conceptualize this process as learning through problem definition; when the problem is engaging and concrete—something relatable—we're more likely to immerse ourselves," he explains, referencing the scholarly term "visceral aesthetics" to describe the satisfaction of problem-solving. "Hopefully, students will feel motivated to learn AI concepts through this compelling medium. It's often challenging for students to grasp loss functions, but visualizing them while attempting to match images might facilitate intuitive understanding of the underlying mathematics. This approach might even inspire further learning."
For those seeking excitement with their education, the Pokerbots competition, celebrating its tenth anniversary, offered participants the opportunity to win substantial prizes by designing and deploying competitive card-playing algorithms in a virtual tournament. "I joined during my freshman year because I was fascinated by the intersection of mathematics, computer science, and game theory; Pokerbots represented an ideal fusion of these disciplines," shares Shreyas Srinivasan, now a junior studying mathematics with computer science and serving his second year as MIT Pokerbots president. "Developing a bot compels you to apply theoretical concepts, delve deeply into implementation details, and enhance your understanding. It delivers a profound sense of accomplishment and the unique experience of creating an AI capable of outperforming humans in poker."
Srinivasan's assessment of the bots' capabilities isn't exaggerated. "There's a meaningful parallel to chess and other games solved through computational approaches," observes Stephen Otremba, a senior in Computer Science and Engineering and this year's Pokerbots head instructor. "With advancements in machine learning and reinforcement learning, algorithms have progressed beyond human performance." Each Pokerbots participant had the opportunity to test this assertion; traditionally, at the conclusion of each IAP, organizers arrange sessions for players to challenge their own creations. "In recent years, across the game variants we've established, the bots consistently defeat their developers," Srinivasan reports. These variants change annually—this year, organizers drew inspiration from the popular Blotto minigame by providing each player with an individual chip allocation. Participants then created bots capable of strategically distributing these resources across three simultaneous poker games. Appropriately, quantitative trading firms primarily sponsored this year's competition. Srinivasan explains the connection: "Both trading and poker center on risk management, and this year's game variant specifically focuses on managing your initial chip resources across three boards. This mirrors portfolio management in trading, where you allocate resources among various stocks and securities to maximize returns."
Both Srinivasan and Otremba acknowledge the pandemic's impact on their cherished competition, but these changes brought unexpected benefits. "We're considering maintaining lecture livestreaming in future years because it enables greater student accessibility," Srinivasan notes, adding that booking guest speakers became more convenient through Zoom. This year, these speakers included renowned computational poker researcher Noam Brown, now an AI researcher at Facebook, who developed Libratus, among the first AIs capable of defeating professional poker players. Whether Pokerbots participants eventually dominate competitive circuits or join trading firms, their experience will undoubtedly prove valuable. Otremba reflects: "The Pokerbots competition provided not only a real-world project environment but also taught me about implementing ideas in code, effective team collaboration, and navigating all stages of software development—from debugging to theoretical challenges. It offered invaluable software development experience and served as a springboard to even more complex projects."
"Code For Good" presented another opportunity for students to apply their technical expertise for positive global impact. This long-running workshop collaborated with nonprofit organizations to address technical challenges, allowing students to contribute their time and skills to causes aligned with their personal interests. "I strongly believe in Code For Good's mission—assisting nonprofits in overcoming technological obstacles," states Lucy Liao, a senior majoring in Computer Science and Engineering who joined Code For Good as a junior organizer, a project management role connecting student teams with nonprofits. "It proved enlightening because beyond considering projects and code, it involved substantial communication—coordinating with nonprofits, ensuring deadline awareness, maintaining project momentum," Liao explains.
These projects vary widely in scope, from focused and manageable to extensive and ambitious. Senior Victoria Juan outlines the application process: "Throughout the academic year, nonprofits apply for assistance via a Google Form requesting details about themselves, their project requirements, and existing resources or budgets. At IAP's commencement, we present these projects to students, who select based on project parameters; sometimes, participants are also passionate about the nonprofit's mission." These nonprofits represent diverse causes, ranging from a peace-building organization in Afghanistan to the Cambridge, Massachusetts, YWCA.
While the Code For Good team previously exercised editorial control over project scope, they recently experimented with allowing students to undertake as much as they believe achievable—with surprising outcomes. "Several projects seemed so extensive that I doubted their feasibility within one month, yet I've been amazed by students' progress throughout January," Liao admits. "Witnessing their accomplishments in such limited time is truly remarkable."
Beyond the resume-enhancing technical achievements, Code For Good participants develop leadership capabilities. Juan reflects: "Helping organize IAP provided invaluable experience as I assumed classroom leadership responsibilities: establishing deadlines, conducting weekly presentations, embracing accountability, and ensuring teams and projects remained on schedule—or troubleshooting when they didn't." Liao emphasizes the unexpectedly crucial role of communication in this technology-focused initiative: "A significant aspect involves learning to communicate with non-technical stakeholders. We frequently interact with technically proficient individuals at MIT, which comes naturally to us. However, engaging with people unfamiliar with computers and uncomfortable with technical terminology presents distinct challenges. This communication experience proves tremendously valuable."
Quite remarkable for a compact learning session!
