---
id: "claim-ai-lacks-novelty"
type: "claim"
source_timestamps: ["§ Behavioral Change"]
tags: ["creativity", "ai-limitations"]
related: ["concept-human-value-add", "contrarian-ai-novelty-myth"]
confidence: "high"
testable: true
speakers: ["Tom Davenport", "John J. Sviokla"]
sources: ["spine"]
sourceVaultSlug: "hbr-seg-spine"
originDay: 1
articleStem: "hbr-cl-95-6-disciplines-genai"
sourceUrl: "https://hbr.org/2024/07/the-6-disciplines-companies-need-to-get-the-most-out-of-gen-ai"
sourceTitle: "The 6 Disciplines Companies Need to Get the Most Out of Gen AI"
---
# Gen AI outputs are rarely truly novel

**Claim:** Because generative AI models are trained exclusively on existing online content, it is highly unlikely their raw outputs will contain truly novel ideas — necessitating human intervention for high-value tasks. This underpins [[concept-human-value-add]] and its contrarian framing [[contrarian-ai-novelty-myth]].

**Confidence: high · Testable: yes** (but see the nuance below).

Enrichment validation (partial; needs nuance): LLMs approximate the *distribution of observed data* rather than inventing new conceptual structures, so much output is derivative in form and content ([[prereq-llm-mechanics-d1]]). **However**, "trained *exclusively* on existing online content" is **over-stated** — major models train on mixed sources (web pages, books, code repositories, licensed corpora, sometimes synthetic data). The general point (training is based on existing data) holds; the "online only" restriction does not.

Novelty literature: LLMs can generate **combinational novelty** (new combinations, unusual analogies) — Boden's "combinational/exploratory" creativity — but are weak at **transformational** creativity (radically new conceptual spaces). **Counter-perspectives:** practitioners note outputs can be *practically novel* ("new to the firm/team") and economically valuable even if globally derivative; IP lawyers warn derivative training raises plagiarism/originality risk, reinforcing the need for human value-add.