DAIID

• 

  Home page

Inspiration

DAIID: Distributed AI Image Detection

People hype up generative AI as a set of tools that will make our lives easier through automation and content generation. However, time has shown that these tools actually seem to make our lives more difficult. Now, much of the internet is filled with AI-generated content that is full of incorrect information. In fact, Amazon reports that over 57% of all content on the internet is AI-generated. This makes it harder for people to interact with authentic human content as they struggle to filter through AI bloat. We are here to help solve this challenge and make technology easier to use. Technology should make our lives easier, not harder.

What it does

Our dApp allows users to upload images to determine how likely they are to be generated by AI. This helps users filter out AI-spam content and be confident that what they are seeing is real—without the cognitive load of doing it themselves. We use the Ethereum blockchain to ensure no single entity can manipulate this service for its own benefit. This means no central authority can use it to push an agenda or censor certain viewpoints. Instead, AI detection is democratized. Anyone can become a node on the Ethereum network and vote on whether an image is AI-generated. Nodes are incentivized to vote correctly through proof of stake and reputation.

How we built it

We used Solidity smart contracts to create a decentralized system where nodes vote on the likelihood of an image being AI-generated. Images are stored using the Interplanetary File System (IPFS), allowing anyone to review them and see the consensus. We built a simple frontend with React, enabling users to upload images to the Ethereum network and watch nodes reach consensus. Additionally, we developed a Chrome extension to demonstrate how this service can filter AI content in real-time.

Challenges we ran into

Our biggest challenge was learning the Ethereum ecosystem. We were all beginners, so we had to quickly learn how to use IPFS, Ethereum smart contracts, and other blockchain tools. There were several instances where nodes failed to communicate over the network properly, making debugging difficult. To resolve these issues, we spent a lot of time researching and using code assistant tools.

Accomplishments that we're proud of

We are proud to have successfully deployed a working smart contract on our local Ethereum network. We are also excited to showcase how this system integrates with the broader internet ecosystem through our web extension, which connects to our blockchain.

What we learned

We learned about Ethereum and distributed systems. We gained experience with smart contracts, deploying decentralized nodes, and implementing a stake-and-reputation consensus mechanism.

What's next for DAIID

DAIID (Distributed AI Image Detection) is just the beginning. This technology can be applied across all types of media, allowing humans to regain control over their internet experience.

Models used for Image Classification

https://huggingface.co/jacoballessio/ai-image-detect

 -trained on the CIFake dataset.  

https://huggingface.co/Organika/sdxl-detector

-created by fine-tuning the umm-maybe AI art detector on a dataset of Wikimedia-SDXL image pairs 

      -https://huggingface.co/umm-maybe/AI-image-detector  

-accuracy: 0.98  

https://huggingface.co/jacoballessio/ai-image-detect-distilled

-Three separate models distilled to make this model   

    -Midjourney vs. Real Images, Stable Diffusion vs. Real Images, Stable Diffusion Fine-tunings vs. Real   Images

-Validation Set: 74% accuracy, Custom Real-World Set: 72% accuracy  

Built With

• ethereum
• ganache
• hugging-face
• ipfs
• javascript
• python
• react
• solidity
• tkinter
• webpack