Becoming Narrow Minded

AI: A Case for Low-Hanging Fruit

Goal

To show that combining many single-purpose narrow AIs often is a better
approach to solving real world problems than to focus on general AI
moonshots

Agenda

1. What is Rizom?
2. Low End Theory
3. In Practice

What is Rizom?

Rizom

Knowledge infrastructure that works like a virtual coworker—devoted to
information management across your tools and teams

Virtual Coworker

Moderates your team's communication on messaging apps

Organizes the valuable knowledge that these messages contain

Augments this knowledge base with information from other sources

Low End Theory

General AI

"The hypothetical ability of an intelligent agent to understand or learn any
intellectual task that a human being can."

Narrow AI

"A term used to describe artificial intelligence systems that are specified
to handle a singular or limited task."

Moonshot

"An extremely ambitious project or mission undertaken to achieve a monumental
goal."

The Problem

The combined promise of general AI and moonshots never panned out. It hasn't
led to the promised breakthroughs, while other solvable real-world problems
are being ignored.

Our Approach

Radical Automation: Automate every laborious and boring chore that can be
automated. Any missed opportunity to do so inevitably results in talent waste.

Human In The Loop: On collaborative messaging apps, it is childishly simple
to put an AI in a continuous feedback loop with its users.

Low-Hanging Fruit: While hardly ever successful, most AI initiatives focus
on moonshots. We take the opposite approach—deliberately tackling the 90% first.

AI Stack

In Practice

Feature Development Flow

1. Isolate a specific problem
2. Explore existing message patterns
3. Project desired outcomes
4. Determine data requirements and methods

Synchronize

Problem: Information overload after absence from messaging apps

Data: Trello boards, meeting notes, message history

Method: Rule-based filtering, NLP summarization

Onboarding

Problem: Tacit knowledge is hard for new team members to grasp

Data: Channel history, pinned messages, shared documents

Method: Context extraction, FAQ generation

Research

Problem: Shared links rarely get read or processed, creating knowledge gaps

Data: Shared URLs, link metadata, discussion context

Method: Content extraction, automatic categorization, knowledge graph

Conclusion

Combining many single-purpose narrow AIs is often a better approach to solving
real world problems than focusing on general AI moonshots.

The future isn't one superintelligent system—it's many specialized tools working
together with humans in the loop.