iiinova — where your digital self keeps evolving

Arvind Satyanarayan

@arvind.bsky.social

5.2K followers 404 following 13 posts

Asst Professing at MIT CSAIL. Data visualization, ML interpretability, accessibility, cognitively convivial information.

Posts Replies Media

Arvind Satyanarayan @arvind.bsky.social · 18/10/2024

A full house in Boston for @kanarinka.bsky.social #ieeevis closing capstone. Thank you to our hosts @Northeastern @KhouryVIS for the gorgeous space!

0111

Arvind Satyanarayan @arvind.bsky.social · 31/07/2024

Perhaps most interesting is the line of control that separates human and AI agency. Its squiggles reflect that responsibility wobbles back-and-forth for different tasks. Rather than a fixed wobble, designers can now demarcate a space for users and AI to (re-)negotiate control.

Once a designer has specified the configuration of agency for the range of conceivable tasks in an application, the line of control emerges and is relatively fixed. Illustration by Sarah Gephart.

Because GenAI is more flexible in terms of possible inputs and outputs, designers can leave some spaces open to future negotiations between users and AI co-agents. Illustration by Sarah Gephart.

110

Arvind Satyanarayan @arvind.bsky.social · 31/07/2024

There are many things I find exciting about this model of fused agency. At base, it suggests many different configurations of agency. For instance, different cultures may emphasize different 50-50 splits. Or, might we track the progression of agency, flagging undesirable modes.

The possibility of GenAI as a co-agent raises questions about whether constituent capacities of composition and interpretation should be balanced (a) symmetrically or (b, c) asymmetrically. Illustration by Sarah Gephart.

How might we characterize a progression of fused agency that moves from relative balance towards GenAI asserting first greater interpretive agency, and then greater compositional agency? Illustration by Sarah Gephart.

100

Arvind Satyanarayan @arvind.bsky.social · 31/07/2024

So, what would it mean to treat GenAI as our co-agent? Well, people often move from acting independently to more collaborative relationships—with deep collaborations yielding a kind of "mind meld" or fusing of agencies. By analogy, we could consider fusing our agency with AI.

Independent agents become co-agents by establishing common ground, composing actions for each other to interpret. Illustration by Sarah Gephart.

When their agencies are so fully blended as to fuse together, co-agents continuous re-negotiate the line of control. Illustration by Sarah Gephart.

By analogy, we can imagine a model of fused agency between a human and AI co-agent. Illustration by Sarah Gephart.

100

Arvind Satyanarayan @arvind.bsky.social · 31/07/2024

In essence, how might we move beyond measuring only the _competence_ of GenAI and evaluate whether it is meaningfully empowering or disempowering people? First, we need a definition of agency: composing and interpreting actions, and being held accountable for those actions.

Agency is the capacity to compose, interpret, and be held accountable for goal-direction action. Illustration by Sarah Gephart.

110