Summary: Improv is vital for a Human-Centered AI future. The reason is that teams work better together and produce better results with spontaneous collaboration. Likewise, to meet human expectations of adaptability to random, unscripted, and contextually appropriate exchanges, AI needs to be better at improvisation.
Chess lesson: AI beat master, humans + computers beat AI
The year is 1996, and IBM’s custom-built supercomputer beats chess champion Gary Kasparov. The defeat inspires Kasparov to leverage Moravec’s paradox, which says machines and humans often have opposite strengths and weaknesses. In 1998, he returned with an ‘advanced chess’ tournament idea featuring humans + computers working together against a supercomputer, which at the time could calculate 200 million positions per second. The game was a draw 4:4. Moreover, a few years later, more “freestyle chess” matches proved successful with human + computer teams. The setup followed Kasparov’s hunch: allow basic PCs to process tactics and let humans focus purely on strategy in collaboration as a team. The results were remarkable: the human and PC teams beat the most powerful supercomputer. Again in 2014, the experiment was reproduced, with players beating a supercomputer- this time made up of amateur chess players using basic PCs.
It turns out humans have strategy and improvisation superpowers. Yea, we suck at tactical computation, unlike computers.
Improvisation is also why it took IBM’s Deep Blue another 20 years to beat players at the game board “Go”, which has a more significant number of move choices than chess.
Improv makes it better
Our improvisation abilities are possibly hard-wired into our humanness. Remarkably, improv may even facilitate the neural foundation for creativity. Indeed, studies on the use of improv by design teams yield impressive results:
- Improv enhances brainstorming effectiveness by design teams (Gerber 2009).
- Enhances team collaboration performance in design (Lacerda & Chung 2011).
- Boosts divergent thinking capabilities (understanding the problem space deeply), leading to richer design solutions later (Lewis, C., & Lovatt, P. J. 2013).
Service Designers regularly use improv, for example, for prototyping intangible services. Interestingly, employees in the service ‘Back Stage’ also benefit: Employees are more capable of service recovery and service delivery when using improv skills.
Why Improv Matters for Human Centered AI
To succeed, AI needs to be steered toward ethics in UX and toward human-appropriate uses of AI. See Ben Schneiderman’s excellent book called Human-Centered AI for more. Inclusive Design strategies that reduce bias and harm, such as adversarial testing or red-teaming, are also vital.
For Meta’s latest rollout of AI chatbots, they spent 6,000 hours red-teaming the model to find problematic use cases…(Alex Heath, The Verge, Sept 27, 2023).
The future of AI is not machine takeover; it’s machine-human augmentation. Improvised interactions between humans and AI systems will characterize this augmentation. This is why improv is important to leverage as we approach an AI-augmented future.
Theater-based improvisation, or improv, is gaining status as a useful practice for interaction design teams. Demand for improv practice grows as practitioners and instructors realize the effectiveness of using improv to foster productivity. For example, Facebook and Google teams use improv practices when creating digital content for their users and customers, Twitter includes improv classes as a part of their employee’s benefits, and Northwestern’s and Stanford’s Design Institutes teach oversubscribed improvisation classes every semester. Source: Funology 2 ‘Improv for Designers: From Usability to Enjoyment’, July 2018
Generative AI is showing us how being able to pivot in real-time can amplify outputs. As we directionally guide AI, we must be ready for anything, including AI-related nonsense and ‘hallucinations’. That’s why we need to Learn improv to get more from AI.
Improv is also a killer feature direction for AI systems
As crucial as improv is for humans to learn, improv is also a killer feature direction for AI systems.
For example, Amazon’s new generative AI-enabled Alexa “can also adjust its tone and response to express things like affirmation, excitement, laughter, and surprise, Amazon says — adjusting to a person’s natural pauses and hesitation to deliver an ostensibly more free-flowing conversation” (Kyle Wiggers, Techcrunch, Sept 20, 2023).
AI pivots to context and intent. For humans, these are colored by social and emotional scenarios. This is likely why AI has been studying emotion as a requirement for decades before the UX community started focusing on Emotion Design.
Facebook is also rolling out generative AI with more social cue detection.
The Meta team spent time “refining additional data sets for conversations so that we can create a tone that is conversational and friendly in the way that the assistant responds. A lot of existing AIs can be like robotic or bland.” (Alex Heath, The Verge, Sept 27, 2023).
This points to the need to understand sociability in AI and UX design.
Since AI is improving at improvisation, we need to improve our improv skills. Not only does improv offer the ability to listen and respond in real time, but it also expands your pattern adaptation. Humans get stuck on patterns or familiar responses. Instead, we must be ready to dance on our toes to keep our AI outputs sharp and helpful and to make them do what we want: expand our imaginations and thoughts.
Go Deeper: Join Frank Spillers’ UX Inner Circle event on Improv for UX