Agent–Computer–Human Interaction
Agent–Computer–Human Interaction = Agent-computer interaction + Human-computer interaction
Fragments of thought about how the existing HCI paradigm might change, assuming that infrastructure for Agent-computer interaction, such as MCP, becomes commonplace and foundational technologies such as LLMs are widely available.
The conversational user interface, which has remained popular since ChatGPT, is not a particularly good UI.
Assumptions
Agent-computer interaction, including MCP, becomes commonplace. What does commonplace mean?
- Remote MCP servers become sufficiently diverse, or tools such as Playwright MCP and OmniParser (now that I write this down, both came from Microsoft) enable AI agents to use existing systems better.
- There is a well-maintained directory of trustworthy service providers (an MCP registry? MCP discovery service?).
- The practical requirements for serious use of MCP and similar technologies—security, authentication, billing, and so on—are in place.
Foundational technologies such as LLMs become widely available. What does widely available mean?
- The cost of using a sufficiently capable LLM approaches zero.
(What else? ToDo)
Revisiting the existing paradigm
Revisit the research at Xerox PARC. In particular, examine which assumptions they made still hold, which no longer hold, and which could be reconsidered, and develop ideas from there.
Possible topics to reconsider:
- Humans: Humans themselves have not changed biologically at all. But 1) cognitive psychology was in its infancy then, leaving considerable room to reconsider the psychological principles used at the time in light of developments in embodied cognition, personality psychology, emotional psychology, and other fields; and 2) even if human biology has not changed, we need to consider how greatly our social context has changed (the internet, social media, etc.).
- Computers: We need to consider the enormous gains in computer performance (computation speed, network transmission speed, storage capacity, and so on), the spread of XR devices, advances in neural-network-based generative AI such as LLMs, and the spread of the internet of things.
(ToDo)
Reference:
Fragments of a new paradigm
(ToDo)