Pseudo-Random Number Guessing AI

Inspiration

I've been wanting to do something like this for a while and this was a good opportunity to make that a reality

What it does

Guesses Pseudo Random numbers about 1.5 standard deviations above guessing

How we built it

Using Google Colab and a library called gspread(allows for google Colab and sheets to interact) we created 7000 random numbers between 1 and 100. We then used TensorFlow, Numpy and other AI/ML libraries to created the model.

Challenges we ran into

Many Ai/Ml problems use inputs that contain quite a lot of information I.E. images our AI got only the previously generated random number. The result of the limited data was that some approaches that can prove effective with large inputs(like CNN's) where off the table and MLP had to be used. We also experienced a large amount of issues with data types and gspread and Pandas did not want to get along.

Accomplishments that we're proud of

As previously mentioned the algorithm achieved 1.5 standard deviations above guessing witch if it were achieved on a truly random number generator would break most of modern cryptography.

What we learned

How to get google sheets(or exel) to interact with google Colab in real time as well as too much about what loss algorithms work for a give AI problem

What's next for Pseudo-Random Number Guessing AI

Time permitting I would like to try to use a similar algorithm on a true random number generator with more data and time. Or use the current algorithm as a discriminator as part of a GAN.