Wikipedia Advising Report: Generative AI

As a college student at the University of Washington who has just finished the Wiki Ed coursework, I am intimately acquainted with Wikipedia culture and generative AI systems. The Wikipedia community will find that generative AI integrated into the platform can be an effective tool to support contributions by providing a tremendous amount of automated assistance on the more repetitive tasks such as initial moderation and copyediting, ultimately improving quality. Wikipedia relies on volunteer involvement to create and consolidate quality information, so while AI could raise the potential output of content, it also jeopardizes community participation and the values that define Wikipedia: accuracy and neutrality. By leveraging the efficiency of AI in performing repetitive work — correcting grammar problems and flagging spelling mistakes, human editors can spend more time on verification and writing/content aspects as well.[1]

At the same time, there is a flip side to being mindful that AI has room for development and is still maturing. AI should aid the editing process, not be a substitute for human editors. This is why a well-thought-out integration of AI into the Wikipedia database should define AI as augmenting, not replacing human contributions, enabling the Wikipedia community to harness the benefits of AI in contributions while retaining the intrinsically human-centric spirit. This report recommends how the Wikipedia community can elicit a response, using theories from community motivation and norms in the context of policies that uphold Wikipedia's neutral point-of-view and include multiple viewpoints while preserving its culture.

Recommendation #1: Define AI's Scope as an Assistive Tool

If Wikipedia is to integrate AI into its platform responsibly, then it needs to establish clear guidelines that restrain the use of AI to support, low-stake editorial tasks: style consistency, grammar correction, and initial moderation. In defining the role of AI in such a manner, quality improvement can be allowed without diminishing the creativity and authority of human editors. WMF should define AI's scope with the following roles:

Copyediting Assistance: AI can improve Wikipedia's readability by identifying and correcting punctuation, grammar, and minor stylistic issues. AI automating routine language editing can help ensure a polished, encyclopedic tone while enabling human editors the time to dedicate more focus to complex content tasks.

Preliminary Moderation: AI can help Wikipedia detect potentially biased language, spam, and low-level vandalism. Then, it can alert human moderators to review flagged edits and thereby help speed up the moderation process of Wikipedia without sacrificing Wikipedia's standards for neutrality and accuracy, and it keeps human moderators involved.

Content Creation Limits: WMF should prohibit or at least restrict AI from generating original content, summarizing complex topics, or sourcing information. Allowing direct content generation by AI with no restriction imposes a risk to Wikipedia's standards of neutrality and sourcing, as AI models do not consistently interpret sources or evaluate reliability.

Supporting Theory: This recommendation aligns with the concept of motivation crowding out and motivation theory. Restricting AI to supportive tasks helps preserve the intrinsic motivation derived from directly contributing to the Wikipedia community.[2] When automated tools take over meaningful tasks performed by real individuals, editors may feel that their contributions are less valuable or less valued, reducing their engagement. By Wikipedia framing AI as a supportive tool instead of a replacement tool, WMF can protect editors' intrinsic satisfaction and commitment to Wikipedia's mission.

Recommendation #2: Human Oversight Policies Shall Be Developed

Another step toward increased quality would be establishing human oversight mechanisms in everything AI generates. Experienced human editors are required by AI assistants to review the edits before publication. Bring full accountability and accuracy in upholding Wikipedia standards with experienced human editors positioned as the last line of verification on Wikipedia content.

Human Review for Quality Control: A protocol of flagging all AI-generated edits for human review before publication provides an essential check on potential inaccuracies or biases that AI might produce. Although AI is knowledgeable and advanced in technology, it can misinterpret specific nuances. Hence, it is crucial that human editors confirm that AI suggestions enhance rather than compromise content quality.

Guidelines for Prompt Engineering: WMF can also provide training for human editors on prompt engineering to ensure editors are knowledgeable about using AI effectively. Crafting prompts emphasizing specificity and neutrality allows editors to help guide suggestions that align with Wikipedia guidelines and standards.

Supporting Theory: This recommendation is rooted in norm enforcement and quality control within collaborative online communities. Wikipedia's descriptive and injunctive norms require active human involvement to uphold.[3] Wikipedia can reinforce these norms by establishing a protocol for human editors to review AI contributions, ensuring that AI is a tool instead of a replacement for quality improvement. Commitment theory plays a significant role here as transparency and oversight and AI use can foster identity-based commitment, reinforcing the role of custodians of Wikipedia standards.[4]

Conclusion: Reinforcing Wikipedia's Mission through Thoughtful AI Integration

Consequently, AI as a supportive tool, not a replacement tool, for copyediting and moderation enables the Wikipedia community to improve the quality of its platform in harmony with respect for its community values and guidelines. By defining AI's role as a tool or assistant, establishing human oversight, and implementing training on prompt engineering, the Wikipedia community can harness the strengths of AI while preserving the communities' intrinsic motivation and engagement. This approach allows Wikipedia to benefit from AI's efficiency in handling routine tasks while maintaining its commitment to high-quality, human-curated educational content.[5] Through thoughtful, limited AI integration, Wikipedia can ensure that AI supports its mission to provide reliable, high-quality information to users while sustaining the involvement and engagement of its dedicated editor community.