AI.spy

Home screen
Detecting AI
Lobby Screen (Mock Up)
Gameplay (Mock Up)

💭Inspiration

AI is pervasive, and its role in art is particularly controversial. As artists ourselves, we ask the important question— “How can we differentiate AI-generated art from traditional art?”

With art theft, fraud, and other issues prevalent in unregulated online communities, it is crucial to be able to preserve the integrity and humanity of art ✊😔🌵

🕵️What it does

🎯 Use AI.spy as a tool or play it as a game!

AI.spy is an online tool that allows companies, artists, and everyday people to use our AI model, the Spy, to identify AI-generated art. Simultaneously, the data collected helps to train and improve the Spy to better identify AI art.

Additionally, AI.spy comes with fun games to encourage better AI art recognition and provide a fun outlet for people which simultaneously improves the Spy through the data received during these games.

Our goal is to make AI identification of images accessible, smarter, and more fun.

🚧Challenges we ran into

Overall, we struggled with: 🔎 Real-time multiplayer coding 🔎 Getting our model properly trained data

A quote from one of our team members: "Funnily enough, one major trouble was actually the first step: the imports. for some reason, keras recently made a change that made it impossible to import from tensorflow.keras, and you just had to write "keras.model" or "keras.layers", etc. something else that gave me a lot of grief was making sure the dataset was good. Making a mistake in the images used meant that the model could be completely unusable. I had to spend a lot of time on getting images, and tweaking the flask endpoints so that when the images were being retrieved from database or uploaded, they were actually correct."

💪Accomplishments that we're proud of

We’re proud of coming up with a novel idea and sticking to it despite all the hardships we encountered along the way. These include the typical turmoil associated with training a classification model from scratch, as well as the pain of handling real-time communication between users. We also are proud of the polished final product we were able to churn out in spite of how much time was spent taking care of backend shenanigans. Overall, we feel good about the potential impact our app can bring to the world and are excited for its future.

📚What we learned

This project taught us the importance of choosing our battles. Even though it was possible, in trying to accomplish all our ambitious goals, we ended up with too little time to set up the little details which were most important. In a vacuum, our components worked well, however, in trying to mesh them together, we were met with many hardships. Moving forward, we will choose our battles and complete the most important tasks first.

It was also the first time for a lot of our members with certain technologies like machine learning, websockets, tailwind, etc. We learned a lot from each other and the mentors, and were able to improve our own coding abilities because of it. In all, we learned a lot.

🔮What's next for AI.spy

🔎 Improving the AI art detection model. 🔎 Expanding more game modes and adding more players.

Despite there still being many flaws with our product, the technology has the capabilities for our cause. For our next steps, we plan on improving on our games to include a larger variety that can appeal to a larger audience, fine-tuning our AI to be more powerful and precise, and also plan on protecting some weaknesses with our model. Currently, malicious actors aiming to worsen our product can intentionally feed it wrong data, which would worsen our model. This is something we plan on combating in the future.