How to Strategize Human Content Creation in the Era of GenAI?

TL;DR

The paper analyzes a dynamic competition between a human content creator and GenAI across $k$ topics, capturing time-sensitive value via per-topic discounts $\gamma_i$ and GenAI learning through discounted counts $N_i(t)$. It proves hardness for time-sensitive domains under the rETH and delivers a near-optimal $(1-\epsilon)/2$-approximation via Myopically-Optimize-then-Pause, while in time-insensitive domains ($\gamma_i=1$) it yields a polynomial-time optimal strategy using a reduced DAG and a longest-path computation with complexity $O(Tk^3)$. The work provides extensive simulations showing the proposed methods outperform baselines and offers insights into when to rely on pausing versus continuous content generation. Together, these results guide platform operators on scheduling human content creation and leveraging GenAI data for training in both decay-prone and timeless content domains.

Abstract

Generative AI (GenAI) will have significant impact on content creation platforms. In this paper, we study the dynamic competition between a GenAI and a human contributor. Unlike the human, the GenAI's content only improves when more contents are created by the human over time; however, GenAI has the advantage of generating content at a lower cost. We study the algorithmic problem in this dynamic competition model about how the human contributor can maximize her utility when competing against the GenAI for content generation over a set of topics. In time-sensitive content domains (e.g., news or pop music creation) where contents' value diminishes over time, we show that there is no polynomial time algorithm for finding the human's optimal (dynamic) strategy, unless the randomized exponential time hypothesis is false. Fortunately, we are able to design a polynomial time algorithm that naturally cycles between myopically optimizing over a short time window and pausing and provably guarantees an approximation ratio of $\frac{1}{2}$. We then turn to time-insensitive content domains where contents do not lose their value (e.g., contents on history facts). Interestingly, we show that this setting permits a polynomial time algorithm that maximizes the human's utility in the long run. Finally, we conduct simulations that demonstrate the advantage of our algorithms in comparison to a collection of baselines.

Figures (19)

• Figure 1: Example interaction between the human, GenAI, and Internet user at a given round.
• Figure 2: Accumulated utility vs round for each algorithm in: (top) time-sensitive domain. (bottom) time-insensitive domain.
• Figure 3: An example of the exponential graph for $k=3$ and $T=3$. For simplicity we assume that $\gamma_i=1 , \forall i \in [k]$ so that each node is recording the total number of pulls. The value of the attribute $d$ and no action $\emptyset$ nodes are not drawn to keep the graph simpler.
• Figure 4: Example illustrating the value for the discounted number of contents for an arm $i$ with $D_i=3$ if its delay value is violated.
• Figure 5: Example illustrating the upper bound on the discounted number of contents for an arm $i$ with $D_i=3$ if its delay value is not violated.

Theorems & Definitions (83)

• Proposition 2.0
• Proposition 2.0
• Lemma 2.0
• Theorem 3.1
• Lemma 3.1
• Theorem 3.2
• Theorem 3.3
• Definition 4.1: Exit Mean and Exit Pull
• Definition 4.3: Long Horizon Setting
• Theorem 4.4