New Interaction Paradigms & Behavioral Engineering
Talk Recap: Karina Nguyen (OpenAI, Anthropic, Dropbox, NYT)
About Karina Nguyen
Karina was awesome. We invited her to come speak at our very first Future Interfaces event in SF. Despite not knowing any of us, she agreed to come out and drop a ton of knowledge for the audience of designers, engineers, and founders building at the intersection of AI.
Karina leads a research team at OpenAI, creating new interaction paradigms for reasoning interfaces and capabilities. She is responsible for the creation of ChatGPT tasks, ChatGPT canvas, streaming chain-of-thought for o1 models, Claude in Slack, and more via novel synthetic model training. Previously, as a product engineer/designer, Karina worked on R&D prototypes, engineering tools, and product features with teams at Primer.ai, Dropbox, Square, and the New York Times.
Key Insights
Real interface innovation is going to be about training and teaching the model for the interaction patterns you want to see in the end product. Many people focus too much on chat interfaces and surface-level optimizations without considering how to fundamentally rethink / retrain the models. If the model itself can't self-correct accurately, there will not be real interface innovation nor novel affordances. To go beyond chat interfaces, invent new interaction paradigms for human/AI feedback systems so that you can train the models to directly optimize for those goals.
Product designers and model trainers need to collaborate more with each other. There is still a bifurcation between these disciplines when it’s increasingly imperative they must work in tandem to create the best human computer interaction outcomes for AI. This also means product designers need to get more technically adept at AI technology.
Karl Lagerfeld stated "embroidery is not mere lux augmentation, but capability intrinsic to the garment." Determining the best and safest form factor for AI capabilities requires a thorough consideration. How do we enable a form factor that is most intrinsic to the system? What type of interface affordances — file uploads, a multi-branch conversation, or something else—would have been most intrinsic for such a capability? This is an unending effort to thoughtfully manage an evolving relationship between humans and machines.
A good chunk of manual prototyping is being replaced by rapid prompt experimentation. Mocking up APIs and seeing how the model performs on an outcome can give a heuristic as to whether the feature is shippable within the next few days, or if the model needs to be taught to be really good at it (a more resource-intensive approach). The complexity arises when you have to navigate this trade-off between research investment to make the system excellent at a certain thing or ship what works well enough for users with the current state of the system.
Detail-oriented design craft won't just mean pixel perfection or the creation of novel metaphors, but also cultivating AI which has a good taste in and of itself. Ellen Ullman put it nicely, "The world as humans understand it and the world as it must be explained to computers come together in the programmer in a strange state of disjunction."
Check out the full video for the entire discourse. Happy interfacing!
