Human-centered AI asks how intelligent systems can support human agency, judgment, learning, and responsibility.
More in this topic
Writing connected to this thread.
guide
Semiotic Processing: Build the Rest of the Computer
Five principles for using language models as processors of meaning while keeping reliable state, rules, and verification in the surrounding system.
note
When does an interpretation become system state?
A model can propose what someone meant. The system has to decide whether that proposal becomes a lasting change.
note
When does the system need a language transform?
A language transform earns its place where a designed system encounters meaning that its ordinary rules cannot use directly.
note
Test the transform at the boundary
Evaluate whether a language transform preserves the distinctions the system needs, then test the system's rules separately.
note
Stateless doesn't mean contextless
A language transform can receive the context it needs for one operation while the system remains responsible for memory and state.
note
Start with the system, then find the transform
Design the behavior people need first. Then ask whether a stateless language transform makes one part of that system better.
essay
What if LLMs Are Something Else Entirely?
When we really look at the technology without preconceptions of AI, LLMs stop looking like "synthetic intelligence" and more like a very weird, very new kind of compute.
essay
Systems from the Gaps
Why and how I do what I do in all my systems work, at SkillSaige and beyond.