• Login
  • Register

Work for a Member organization and need a Member Portal account? Register here with your official email address.

Project

Unspectacular Feed

Creative Commons

Attribution-NonCommercial-NoDerivatives 4.0 International

Saetbyeol LeeYouk

Project Contact:
Groups

Unspectacular Feed is a platform level chrome extension for the automatic detection and disclosure of AI generated content on social media. When a content uploader uploads a photos, the system automatically analyzes whether artificial intelligence was involved in its creation. If AI use is detected, the system generates a credit indicating the confidence level, the specific regions of the image that were altered, and the assumed prompts applied to those regions. Uploaders may revise this information and add references to the AI models they used.

For viewers, the system offers a filtering function. Activating the "Unspectacularize" feature overlays a mask on AI generated regions within a post and surfaces the associated prompts, allowing viewers to understand the creative intent behind the use of AI in that particular piece of content.

Creative Commons

Attribution-NonCommercial-NoDerivatives 4.0 International

We are currently recruiting participants for a user study on Unspectacular Feed. If you meet any of the criteria below and are interested in experiencing the system, please find further details here. We are looking for:

  • Content creators, including (1) digital artists, (2) journalists, and (3) issue focused content creators engaged in secondary journalism, who maintain an active presence on at least one social media platform such as Instagram, TikTok, Facebook, or YouTube, and who regularly upload original content at least once a week.
  • (4) Active social media users who encounter news, issue coverage, or informational content in their feeds at least three times a week.

Why It Matters

While misinformation and fabricated news have long been concerns in social media discourse, the photorealistic fidelity of contemporary AI generation models substantially lowers the threshold at which audiences accept manipulated content as authentic. The human eye can no longer serve as a reliable verification mechanism. Critically, this epistemic instability operates in both directions: fabricated content is increasingly believed, while authentic content is increasingly dismissed under doubting as "AI generated."

Existing image forgery detection and localization systems are not designed for the contexts of social media feed in which manipulation posts most commonly shows. Detection requires users to seek out external tools or verification websites, transferring the burden of authenticity checking from platforms and content creators to end users. Platform-side labels (e.g., Instagram's AI generated tag) are applied only when the platform's own AI tools are used, leaving externally generated content unlabeled and effectively unaccountable. As a result, responsibility for verifying content authenticity is displaced onto end users rather than enforced at the platform level.

Disclaimer Note

Unspectacular Feed is an early-stage research project currently being developed at the MIT Media Lab. A paper describing this work is being prepared for submission for peer review. Until that review is complete, all findings and demonstrations should be understood as preliminary.

In simple terms, this means the system is still being built and tested. Independent scientists have not yet evaluated it, and its conclusions should not be treated as final.

There are several important limitations to keep in mind at this stage:

  1. The detection model is still in development and does not yet perform reliably. In some tests, images that were clearly AI-generated were not flagged as such. Significant improvement is needed before the system can be trusted to detect deepfakes accurately.
  2. The system has only been tested as a Chrome extension overlay on Instagram. It has not yet been evaluated across other social media platforms or browsing environments, and results may differ in those contexts.
  3. Deepfake detection is a moving target. As AI tools for generating images and videos improve, detection models must be continuously updated to keep pace. The accuracy of the current system will degrade over time without ongoing retraining.
  4. The "Human Made Badge" feature, which certifies content as created without AI, relies on documentation provided voluntarily by uploaders. The standards for what counts as sufficient documentation have not yet been formally defined or independently validated.
  5. The empirical study evaluating the system's usability and social impact is currently in its early stages. No findings from user research are available at this time.
  6. The broader questions this research raises — including how different types of AI use should be disclosed, and how platforms and viewers should respond — remain open. The system currently focuses on image based content. Video detection involves distinct technical challenges and is not yet within the scope of this project. These are active areas of ongoing research.

This work represents a promising early step toward addressing a real and growing problem. We share it to be transparent and invite the research community to engage as it develops.