iiinova — where your digital self keeps evolving

Special thanks to @katygb.bsky.social and @sethlazar.org for organizing a fantastic @knightcolumbia essay series on AI and Democratic freedoms, and other authors of the essay series for valuable discussion + feedback. Check out the other essays here! knightcolumbia.org/research/art....

Artificial Intelligence and Democratic Freedoms

000

Finally, autonomy levels can help with agentic safety evaluations and setting thresholds for autonomy risks in safety frameworks. How do we know what level an agent is operating at, and what appropriate risk mitigations should be put in place? Read our paper for more!

110

We also introduce **agent autonomy certificates** to govern agent behavior in single- and multi-agent settings. Certificates constrain the space of permitted actions and help ensure effective and safe collaborative behaviors in agentic systems.

120

Why have this framework at all? We argue that agent autonomy can be a deliberate design decision, independent of the agent's capability or environment. This framework offers a practical guide for designing human-agent & agent-agent collaboration for real-world deployments.

100

Level 5: user as an observer. The agent autonomously operates over long time horizons and does not seek user involvement at all. The user passively monitors the agent via activity logs and has access to an emergency off switch. Example: Sakana AI's AI Scientist.

100

Level 4: user as an approver. The agent only requests user involvement when it needs approval for a high-risk action (e.g., writing to a database) or when it fails and needs user assistance. Example: most coding agents (Cursor, Devin, GH Copilot Agent, etc.).

100

Level 3: user as a consultant. The agent takes the lead in task planning and execution, but actively consults the user to elicit rich preferences and feedback. Unlike L1 & L2, the user can no longer directly control the agent's workflow. Example: deep research systems.

100

Level 2: user as a collaborator. The user and the agent collaboratively plan and execute tasks, handing off information to each other and leveraging shared environments and representations to create common ground. Example: Cocoa (arxiv.org/abs/2412.10999).

100

Level 1: user as an operator. The user is in charge of high-level planning to steer the agent. The agent acts when directed, providing on-demand assistance. Example: your average "copilot" that drafts your emails when you ask it to.

100

Web: knightcolumbia.org/content/leve.... arXiv: arxiv.org/abs/2506.12469. Co-authored w/ David McDonald and @axz.bsky.social . Summary thread below.

Levels of Autonomy for AI Agents

100

Here are links to the paper: Web: knightcolumbia.org/content/leve.... arXiv: arxiv.org/abs/2506.12469. Co-authored w/ David McDonald and @axz.bsky.social .

Levels of Autonomy for AI Agents

000

Special thanks to @katygb.bsky.social and @sethlazar.org for organizing a fantastic @knightcolumbia.org essay series on AI and Democratic freedoms, and other authors of the essay series for valuable discussion + feedback. Check out the other essays here! knightcolumbia.org/research/art....

Artificial Intelligence and Democratic Freedoms

100

100

120

100

100

100

100

100

100

📢 New paper, published by @knightcolumbia.org We often talk about AI agents augmenting vs. automating work, but how exactly can different configurations of human-agent interaction look like? We introduce a 5-level framework for AI agent autonomy to unpack this. 🧵👇

Screenshot of a paper on a webpage, with a figure showing 5 levels of autonomy for AI agents and the tradeoff between user involvement and agent autonomy as the levels increase.

1204

Knight First Amendment Institute @knightcolumbia.org · 28/07/2025

Hi Dave, this is super cool! I'm currently working on a research prototype that brings domain experts together to draft policies for AI behavior and would love to chat more, esp about how I can integrate something like this

020

Reposted by Kevin Feng

In an essay for our AI & Democratic Freedoms series, @kjfeng.me, @axz.bsky.social (both of @socialfutureslab.bsky.social), & David W. McDonald outline a framework for levels of #AI agent autonomy that can be used to support the responsible deployment of AI agents. knightcolumbia.org/content/leve...

Levels of Autonomy for AI Agents

033

Kevin Feng @kjfeng.me · 27/04/2025

Excited to kick off the first workshop on Sociotechnical AI Governance at #chi2025 (STAIG@CHI'25) with a morning poster session in a full house! Looking forward to more posters, discussions, and our keynote in the afternoon. Follow our schedule at chi-staig.github.io!

041

Reposted by Kevin Feng

Tzu-Sheng Kuo 郭子生 @tskuo.bsky.social · 27/04/2025

@kjfeng.me and @rockpang.bsky.social are kicking off the panel on sociotechnical AI governance! #CHI2025

074

I'll be at CHI in-person in Yokohama 🇯🇵. Let's connect to chat about anything from designing LLM-powered UX to AI alignment + governance!

020

Thanks to my fantastic collaborators @qveraliao.bsky.social @ziangxiao.bsky.social @jennwv.bsky.social @axz.bsky.social and David McDonald. This work was done during a summer internship with the FATE team at Microsoft Research.

110

Canvil allows you to craft AI-powered user experiences by shaping the behavior of large language models (LLMs) right in your canvas. Works in both Figma and FigJam! Demo (3 mins): https://youtu.be/E0...

We published Canvil on Figma community and it's already used by 2k+ designers worldwide. Try it yourself on Figma today! www.figma.com/community/wi.... Or, check out a demo here: youtu.be/E0l-IH_Lo7Q.

figma.com

Canvil | Figma

100

Finally, Canvil's seamless integration into Figma allowed designers to use their existing collaborative workflows to work with design and non-design teammates on model adaptatation. We believe lowering the barrier to adaptation can catalyze more thoughtfully designed LLM UX.

100

Designers also used constraints from their existing design requirements and UIs to shape model behaviors. For example, many devised ways to embed cultural preferences and customs into models based on the needs and backgrounds of their target users.

100

We used Canvil as a probe to better understand how designers would make use of designerly adaptation in practice. We found that using Canvil, designers identified clear opportunities to tweak their designs based on observed model behaviors to enhance user interaction with LLMs.

100

We propose designerly adaptation, a process that enables design requirements and UX designs to shape and be shaped by LLM behavior. We operationalized designerly adaptation with Canvil, a Figma widget that allows for quick experimentation + adaptation of LLMs in design canvases.

100

We interviewed 12 designers at Microsoft to learn how they design LLM-powered UX. We found that designers often thought deeply about how to adapt model behavior to improve UX, but lacked the means to actually experiment with model adaptation within their workflows.

100