Dr. Piotr Mirowski: AI, Improv, and the Future of Human-Machine Collaboration
[HPP] Raia HadsellApril 10, 202541 min
28 connectionsΒ·40 entities in this videoβBridging AI and Performance
- π§ Piotr Mirowski is an AI research scientist at DeepMind and a passionate theater improviser, having spent 10 years in New York enjoying and performing theater.
- π‘ He realized around 2016 that his two worlds of AI research and performance could connect, leading to innovative projects.
- π His work involves training AI systems to be useful for humans, including robots and algorithms for tasks like weather forecasting or epileptic seizure prediction, even before AI became a buzzword.
Evolution of AI in Creative Arts
- π Early language models in 2013-2015 became fast enough for conversational text generation, sparking a revolution in AI's creative applications.
- π¬ Initial AI-generated creative works like the film "Sunspring" and play "Beyond the Fence" in 2016 were often nonsensical but inspired new forms of human-AI collaboration.
- π€ Mirowski developed an improv show where a toy robot interacted with humans using speech recognition and a language model, creating a live, challenging performance that he built from scratch.
The Dual Nature of AI in Improv
- β AI in improv challenges performers to adapt and become better, with audiences often supportive and cheering for the human effort to make it work.
- β οΈ However, integrating technology introduces significant technical challenges and stress, as seen in instances like robot crashes and logistical issues during remote performances.
- π The experience led to a new saying, "never work with robots," echoing traditional theater advice about children and animals, due to the inherent unpredictability.
Audience Engagement and Ethical Considerations
- π¬ Mirowski's shows feature a robot engaging directly with the audience, initially via a "Wizard of Oz" setup, then transitioning to AI-generated responses to trigger memories and stereotypes.
- π Performances also address contemporary AI issues like deepfakes and copyright related to models trained on vast datasets, highlighting the rapid progress of technology.
- βοΈ The discussion emphasizes the importance of consent when AI generates content in an artist's style, contrasting it with unauthorized use of their work.
Future of Human-AI Collaboration
- π§ AI excels in rigid, rule-based systems like chess or Go, outperforming humans due to finite possibilities within defined universes.
- π However, the human world is infinitely complex, involving emotions and motivations, making human tasks much harder for AI to fully master, as seen with self-driving cars.
- π± AI can serve as a writing aid, helping playwrights and screenwriters with ideation, creating "vomit drafts," or offering different perspectives, rather than replacing human creativity.
- π€ The goal is to use AI as a tool to challenge and inspire humans, making them better performers and creators, similar to how a marathon challenges a runner to improve personally.
Knowledge graph40 entities Β· 28 connections
How they connect
An interactive map of every person, idea, and reference from this conversation. Hover to trace connections, click to explore.
Hover Β· drag to explore
40 entities
Chapters14 moments
Key Moments
Transcript151 segments
Full Transcript
Topics15 themes
Whatβs Discussed
Artificial IntelligenceDeepMindRoboticsImprov TheaterLanguage ModelsSpeech RecognitionHuman-Machine CollaborationChatGPTDeepfakesCopyright IssuesScreenwritingPlaywritingAI as a Writing AidChess AIGo AI
Smart Objects40 Β· 28 links
ConceptsΒ· 9
PeopleΒ· 8
MediasΒ· 11
ProductsΒ· 4
CompaniesΒ· 4
EventsΒ· 3
LocationΒ· 1