The Jagged Frontier of AI is a concept from Co-Intelligence used to describe the unpredictable and uneven nature of AI capabilities.

It is explained using the analogy of a fortress wall with towers and battlements that extend unevenly¹, representing the capabilities of AI. Everything inside this metaphorical wall can be accomplished by AI, while everything outside poses a challenge. However, the challenge lies in the invisibility of this wall, making it difficult to predict which tasks AI can or cannot perform.

Think of it like this: you might assume that writing a sonnet and a 50-word poem are equally difficult tasks for AI. However, due to AI’s reliance on processing “tokens” rather than understanding words like humans, it might excel at the sonnet but struggle to limit a poem to precisely 50 words. This unpredictability arises because AI excels in tasks where it can leverage patterns from its training data but struggles when faced with novel situations or tasks that require a deeper understanding of human context. For instance, AI might surprise you with its ability to generate creative ideas but disappoint you with its grasp of basic mathematics.

The Jagged Frontier highlights the importance of continuous experimentation with AI to understand its capabilities and limitations better. By experimenting, individuals can gain expertise in utilizing AI effectively for specific tasks within their domains. This exploration allows individuals to map out the contours of the Jagged Frontier through trial and error, identifying which tasks AI can augment and which ones still require human intervention.

Iteration with AI yields the best results.

source::Co-Intelligence (book)

Footnotes

1. The imagery of a Star Fort popped up as soon as I read this analogy. ↩