Digital philosopher

We composed corpuses from texts of several philosophers and used them to train a neural network specialized in understanding language and generating new texts. This way we tried to create virtual characters of "digital philosophers," whose generated texts we analyzed. Using this website we give you access to the trained nets so you can try generating yourself.

Website was created in course Contemporary Philosophy taught as part of New media studies at Faculty of Arts at Charles University during autumn 2019 by Mgr. Dita Malečková, Ph.D. (ÚISK FF UK) and Jan Tyl (Alpha Industries s.r.o.).

Choose corpus

We are using language model GPT-2 by OpenAI introduced in February 2019, specifically the version with 345 milion parameters. GPT-2 model is so capable that OpenAI decided to release versions gradually, because of the risk of it being misused for generating disinformation. That the risk is real is confirmed for example by research made by Sarah Kreps and Miles McCain from Cornell University published in Foreign Affairs, which showed that people find texts generated by GPT-2 almost as convincing as real articles from New York Times. Find out how GPT-2 works under the hood.

Using the interpretation of Czech copyright law by Jan Zibner and Matěj Myška for works created by AI, published at portal Iurium (in Czech only), authors of the generated texts are the creators of the used corpus. In case you write in an original text, which should the generating follow, you are becoming a co-author. As creators of the corpuses we allow use of the generated texts under license Creative Commons Attribution-NonCommercial 4.0.Creative Commons License

Assemblage of this website was linked by Jan Vlček, code is available on GitHub.