AI Platforms for Interactive Storytellers: New and Improved Tools
At the beginning of 2018, I curated a list of five accessible tools aimed at anyone with basic Python scripting abilities for Docubase. I was a beginner in the field of machine learning and overwhelmed by its complexity. Machine learning can be a fantastic tool for creators, but integrating AI into your workflow is a challenge for those who don’t have much coding experience.
Two years later, I recognize it was naive of me to think that it is possible to just “discover the complex ideas and building blocks of deep learning and machine learning technology” with only basic familiarity with Python. The responsibility to uncover and challenge the stories behind the machine learning algorithms is not on the individuals themselves, but the creative community as a whole.
Luckily, wonderful people are thinking creatively about how we can bring everyone into the process of working with complex technologies. It is our responsibility towards one another. I hope this piece will encourage more people to join the conversation and diversify the voices that manufacture our reality. In the last two years, more accessible platforms have been developed; therefore, we find it necessary to update this list. If I missed something, or you have suggestions, please contact me at shirinanlen@gmail.com.
- RunwayML by Cristobal Valenzuela What started as Cristobal Valenzuela’s thesis project grew into a brilliant product with a simple goal: to put machine learning in the hands of creators. No code is required to play and explore. Anyone with a computer can start creating, thanks to an excellent user-friendly interface. For the more adventurous ones, this software has a full developer kit that can provide options for integrations and trainings. RunwayML, connected to a cloud service, is constantly being updated with new models to explore and engage with. https://runwayml.com/
Captured
from RunwayML website2. ML-agents by Unity
The Unity Machine Learning Agents Toolkit (ML-Agents) is an open-source project for training intelligent agents in a game environment. It is fairly easy to install within the game engine software and a variety of training scenarios are possible, depending on how agents and rewards are set up. Unity provides several pre-trained environments and agents to start and play with, and there are already a bunch of easy and fun tutorials to follow online.
3. ml5.js by ITP under the guidance of Daniel Shiffman One of the tools on my list last time was Tensorflow.js. Although it provides a kind of straightforward way for users to train computers to perform complex tasks and a great “playground area” for interactive learning about neural networks, the emphasis here is on the “kind of.” You still need to have basic knowledge of Python or C++ programming to find your way around. That’s why I was excited when the ml5.js project was announced. It’s a JavaScript library, built on top of Tensorflow.js, that brings a friendlier interface to machine learning on the web. There is a bit of code needed to browse and play with different machine learning models, but there are good documents to follow and some of the models are available for good, old-school web interaction. The project is developed by ITP and heavily inspired by the open-source platforms Processing and p5.js. https://ml5js.org/
ml5.js
website cover image4. OpenAI GYM OpenAI GYM is a python library that provides a large number of test-game environments to work with a Reinforcement Learning (RL) agent’s algorithms and shared interfaces for writing general algorithms and testing them. You need to have some understanding of Python, but the documentation is pretty accessible. The problem that OpenAI is trying to tackle is to remove the barrier of setting up training environments for RL agents. Working with RL can be extremely difficult, and this might increase engagement in the field of AI and provide tools with which everyone can learn the basics. https://gym.openai.com/
from a paper
on OpenAI GymAdditionally, OpenAI announced that they are releasing an API for accessing a new AI language model — GPT-3. It is being said that GPT-3 is better than any prior program at producing lines of text that sound like a human could have written them. Anyone can now register on their waiting list to request access.
5. Pandas, Python data framework. Sometimes we just need a useful, efficient tool to help untangle the mess of big data so we can see clearly the underlying stories. One of the links on my previous list is Scikit-Learn. After playing with Pandas, I would like to replace it. I’m sure there are pros and cons between the two, but for accessibility purposes, Pandas is the absolutely winner! Pandas is an open-source software library written for the Python programming language for data manipulation and analysis. It relies on Numpy and Matplotlib, two of the main libraries in the Python ecosystem. In a nutshell, it is Excel for Python with rows and columns. It is easy to use and a super flexible, robust data analysis, and data visualization tool.
6. ML4Artist by Gene Kogan I already mentioned Gene Kogan last time — but ML4Artist is so great that I have to mention it again. This is not a tool or piece of software, but a free book about machine learning for artists, with a collection of effective resources. ML4A is an initiative by Gene K,.mogan, an artist and a programmer who collaborates with numerous open-source software projects. https://ml4a.github.io/
Captured
from Gene’s websiteAdditional educational interactive web tools:
- GauGAN network by Nvidia: You can explore how categorized images are generated by using the paintbrush.
- A web tool of Pix2Pix network: You can explore how image-to-image translation is being translated with a Conditional Adversarial Network.
- Explore the latent space of the various image layers in this Deep Conventional Generative Adversarial Network.
- This Person Does Not Exist might seem like a random portraits site, but every single photo on the site has been created by using a special kind of artificial intelligence algorithm called generative adversarial networks (GANs).
Immerse is an initiative of the MIT Open DocLab and The Fledgling Fund, and it receives funding from Just Films | Ford Foundation and the MacArthur Foundation. IFP is our fiscal sponsor. Learn more here. We are committed to exploring and showcasing media projects that push the boundaries of media and tackle issues of social justice — and rely on friends like you to sustain ourselves and grow. Join us by making a gift today.