this fall, i'll be spending six weeks at the Recurse Center in New York City.
i've been fascinated by the spirit of the Recurse Center ever since i first heard about it one year ago—this community of kind, brilliant, caring humans who contribute amazing things to the world through programming.
during this time, i'm thankful to be supported by a fellowship which encourages the undertaking of ambitious projects by a diverse group of actors.
first, some motivation
my experience in machine learning to-date has been across supervised + unsupervised learning: recommender systems + learning to rank, classification problems, social network analysis, embeddings, natural language processing on tweets. i've also done some silly things like finetuning GPT-2 on tweets to study whether it can produce comedy.
recently, i've been trying to understand where i'd like to specialize—machine learning is a big place. i'm fascinated by two separate areas in particular: deep reinforcement learning, and how we can use machine learning for combating misinformation in our shared information ecosystem.
i'm interested in deep reinforcement learning because of its generalizability. it's curious to me that the same key algorithms can be taught to play Go, optimize traffic systems, and so much more. i'm curious how this research will advance our progress toward artificial general intelligence. at this point, this is a pursuit driven out of personal curiosity. i think that building artificial general intelligence is a compelling problem and the path to getting there will result in many unexpected advances over our lifetime.
i'm also interested in studying the integrity of our information ecosystem. along with Pippin Lee, i've recently been working on automating fact-checking of news claims with natural language processing methods via the 2019 Leaders' Prize competition. this has been extremely fun, but we're finding that the Kaggle-like competition nature of this imposes arbitrary constraints which leave us spending too much time on unnecessary engineering optimization (the submission pipeline must run with arbitrary memory/time constraints and stacktraces are not returned) as opposed to experimentation and research with new methods, which is ultimately our end goal.
more broadly, i'm interested in the impact of artificial intelligence on the way we interact with truth in what is quickly becoming a post-truth world, and the threats to our democracy. advances in machine learning are allowing malicious actors to produce and spread synthetic media of all sorts faster than ever.
along these lines, questions which interest me include:
how do we encounter disinformation in our everyday internet interactions? what shape does it take? how do we identify when this is happening?
what is the impact of state-backed misinformation on our collective democracy? i hope to study this via data released by Twitter.
who are the most vulnerable populations for disinformation? how can we build tools to help them protect themselves?
how can we use machine learning to drive countermeasures? how can machine learning and natural language processing help with automated fact-checking, bot detection, predicting & measuring the spread of misinformation, and social network analysis on sites like Twitter?
i'm convinced that misinformation/disinformation and our relationship to truth (and to each other) is going to be one of the most important problems of our lifetime, and there are not enough people working on technical countermeasures and policy reform for the scope of this problem.
what i hope to learn over these six weeks
throughout my time at RC, i'm hoping to understand a few things about these topics, myself, and my relationship to the technology ecosystem:
where can i have the most positive impact with my current skillset?
how should i continue to grow my skillset to effectively make progress on these problems?
who can i collaborate with on these problems to maximize the impact we can have together?
how should i share my progress to help others working in this field or those looking to understand more about these problems?
what are the most promising avenues for further research in these domains?
how do we, as technologists, contribute to the intentional politicization of artificial intelligence in ways that benefit everyone?
here's a tentative plan for the six weeks. i'll be writing about my experience throughout my time, and i'm really excited to receive feedback and be surrounded by the talented people at Recurse Center as well as the wider technology community. send me thoughts via Twitter!