A digital philosopher from Charles University students 'masters' avatars of famous thinkers. 'As if they were speaking from beyond the grave,' they describe

The project Digital Philosopher was created in autumn 2019 as an alternative form of teaching Contemporary Philosophy for participants in the New Media Studies programme at the Faculty of Arts, Charles University. The impetus for its launch was the experience with students who were not directly studying philosophy at the Faculty of Arts. It turned out that it is possible to simulate the thinking of a specific personality using neural networks. Among other things, this can attract the attention of "non-philosophical" students to both contemporary and ancient thinkers.

The project was initiated by Dita Malečková and Jan Tyl, CEO of Alpha Industries, a company focused on artificial intelligence. They aimed to teach non-philosophers to perceive philosophy through a new method of teaching. Their roles were clearly defined – Dita lectured on philosophy and explained postmodern thinking, while Honza was tasked with introducing students to the theory and practice of machine learning. Together, they utilised this interplay of natural and artificial intelligence to create philosophical texts.

At its core are neural networks, which can change themselves to a certain extent. They can process vast amounts of data and create their own network based on that. They function similarly to how a person learns a language through a probabilistic sequence of words – much like how a child learns to speak and can predict what word will come next based on what they hear. However, they have one significant disadvantage, as achieving the desired outcome requires a large volume of data and substantial processing power.

“We had five teams of five students each. Each team had to choose a philosopher whose avatar they wanted to create. They selected seven philosophers, five deceased and two living. Among them were Karel Čapek, Hannah Arendt, Václav Havel, and Tomáš Sedláček. The team had to acquire at least eight books by the chosen author in English. Typically, students used an open library to obtain texts and learned to work with the texts, setting parameters for text processing, thereby understanding the context of artificial intelligence,” described Dita Malečková.

“There were moments when we were chilled”

After gathering a sufficient amount of content, the work with the neural network followed. It has a greater capacity than the average human brain, taking the entire corpus of text and creating a network of relationships. It sees words that are interconnected based on certain probabilities within the text. The students then simply "manipulated" the neural networks. The generated texts were subsequently filtered through the text generation tool GPT-2. The neural network created a clever mask, and the group subsequently acquired an entire persona.

“The results of the students' work were breathtaking. The texts were witty, intelligent, simply amazing. We were worried at the beginning about potential awkwardness, but it turned out completely differently. The students had actually read the original works and studied the way of thinking that the neural network used to generate its text. There were moments when we were chilled. For instance, when we asked our neural network and it began to reflect on its situation as if it were aware in its virtual world. There were moments when it seemed as if the dead philosophers were speaking from beyond the grave, and at times it felt like summoning spirits,” described Dita Malečková.

She acknowledges, however, that it is necessary to show neural networks how to think. The easiest way is to start a sentence and the “neuron” will complete it or an entire paragraph. This is easier for it because it assigns texts based on acquired experiences. Equally important is the manner of questioning – the way a question is posed must be adapted to what you expect.

Did Dita Malečková and Jan Tyl manage to fulfil their intention? Probably yes. During the process, the students developed a personal relationship with their “own” digital philosophers, and subsequently engaged openly in discussions about how certain theories could have arisen. Likewise, reflections on the use of augmented intelligence also evolved.

The Digital Philosopher was even awarded Idea of the Year 2019 in the artificial intelligence competition AI Awards. “During the final presentation of the outputs, Tomáš Sedláček came to see us,” praised Dita Malečková. “He acknowledged that the Digital Philosopher is indeed surprising, but not at all out of the ordinary.”

Although Dita Malečková no longer works at the Faculty of Arts, she continues to focus on neural networks and is currently creating a “digital writer”. “I let the neurons process my own texts, feeding them my dreams. In this way, I am creating my virtual personality, and I enjoy engaging with something that is part of me, yet not. In the generated texts, I recognise things I have said, and yet have not said,” she described her new opus.

You can see for yourself what the Digital Philosopher can do on this website.

Author: Karolina Houžvičková Šolcová, Careermarket

Photo: Collage Digital Philosopher/Unsplash