Collaborating With AI: Thinking, Learning, Career Planning With AI as Co-Creator

Purpose & objective

As part of their daily work, educators and students work closely with AI systems - such as combinations of recommendation systems, digital assistants, and behavior monitoring systems - often without realizing they are working with AI or how AI is influencing their behavior and thinking. This workshop will introduce teachers, professional development and curriculum specialists to the basic structures of generative AI - a type of AI that uses existing content like text, audio files, or images to create new plausible content - and strategies for collaborating and creating with an assortment of generative AI tools. Participants will first experience the learning activities as a student to learn how these AI systems work, and then reflect upon the activities and pedagogy that they can use to increase student awareness and knowledge of these systems and their ethical and societal implications. Participants will also learn pedagogical practices tested in our prior work to encourage students to think AI as a collaborative partner on a project and in the workforce (e.g., asking students to collaborate with a generative AI tool to create a story, showing examples of human working with AI in work). As a result of the workshop, participants will be able to implement a series of scaffolded learning experiences designed to teach students basic structures of generative adversarial networks (GANs), strategies for co-creating media with GANs, and lead discussions to further student understanding of the social and ethical implications and possibilities of human collaboration with AI.

The curriculum being presented was developed as part of the DAILy Curriculum, designed by MIT educators and experienced facilitators. It features hands-on and computer-based activities on AI concepts, ethical issues in AI, creative expression using AI, and how AI relates to students’ future. A scaffolded series of engaging activities guide students through processes of investigating bias in machine learning applications, use GANs to create novel works of art, and learn to recognize the AI you interact with daily and in the world around you.

Importantly, explicit instruction, authentic examples, and personable role models have all been shown to motivate student participation and lower barriers to accessing curriculum, particularly for students with diverse backgrounds and a variety of educational needs . The DAILy curriculum was designed with these pedagogical moves at its heart. Activities include accessible directions, scaffolds, and structured modeling to lower cognitive barriers for accessing content. Learning experiences include authentic, real-world examples not only of the tools but also of the people who use them. The curriculum includes a curated collection of videos exhibited by professionals in AI related fields from diverse backgrounds, using personable language and stories to briefly explain how they entered their STEM field. Each of these methods plays a key role in addressing one of the main objectives of the DAILy curriculum, which is to make it available and accessible to historically marginalized communities.

Outline

In the proposed 1.5-hour deep-dive workshop, participants will first be introduced to generative adversarial networks (GANs), a generative AI technique, through a succinct presentation including short video interviews of AI practitioners describing their work and the technical and social applications for this technology (e.g., https://www.youtube.com/watch?v=YUVgGxS4u7M). Participants will then be given a brief period of time to view video interviews with a number of professionals across a variety of fields (e.g., transportation, social media, art) from historically marginalized communities who describe their work with AI and the challenges and benefits of collaborating with AI. As a knowledge-consolidation activity, participants will engage in heterogeneous knowledge groups (whose members watched different videos) to discuss lessons learned from each video and decide on generalizable lessons learned from across the videos to answer the driving question, “What do our stakeholders (i.e., students, teachers, administrators, the people we coach or work with) need to know about the current landscape of professionals collaborating with AI?”

Small group work will be followed by whole group discussion to achieve learning objectives 1a and 1b to summarize how professionals are working and collaborating with AI. Discussion will be guided towards achieving the understanding that professionals work with AI to be more flexible, complete tasks faster, increase the scale and/or scope of their work, and inform their decision making processes. This understanding is key to messaging the utility and purpose of developing and working with AI. As the discussion closes, participants will share how they intend to message this understanding to their colleagues, teachers, and/or students. If time allows, these messages will be shared, considered by the group, and tuned for precision and accuracy.

Next, participants will engage in a series of interactive condensed versions of activities from the DAILy curriculum, designed by MIT educators and experienced facilitators to introduce the components, processes, and limitations of GANs to middle school students. This will begin with a brief introduction to portfolio of tools and learning activities for teaching high level concepts regarding the what and how generative adversarial networks (GANs) function, including Style transfer (https://mitmedialab.github.io/GAN-play/), Portrait AI (https://portraitai.app/), This person does not exist (https://thispersondoesnotexist.com/), and Text generation from books, (https://mitmedialab.github.io/generative-text/).

Then participants will engage in a small set of scaffolded learning experiences designed to introduce GANs, its abilities, limitations, and social implications. First, participants simulate the adversarial exchange within a GAN by role-playing a simulated series of exchanges between a discriminator and a generator. Then, participants investigate a set of simple web-based GANs such as GPT-2 text generation (https://transformer.huggingface.co/doc/distil-gpt2), Dall-E decoder (https://huggingface.co/spaces/flax-community/dalle-mini), Music generation (AI-duet) (https://experiments.withgoogle.com/ai/ai-duet/view/), Beat blender (https://experiments.withgoogle.com/ai/beat-blender/view/), and Drawing generator (https://magic-sketchpad.glitch.me/). These experiential and explorational learning activities will be followed by a whole group discussion to achieve learning objectives 2 and 3.

Finally, participants engage in a collaborative AI media generation project involving collaborative creation of a story with a GAN, “Storytelling with AI” from the DAILy curriculum. In this condensed presentation of a mini-project participants 1) choose a model text (a renowned novel from a digital library) as training data, 2) write a sentence or phrase as seed text, and then pass it to the trained GAN. The GAN in turn, generates a paragraph of novel text informed by the model and the seed. 3) Participants edit this paragraph so as to tune it to better meet their aesthetic goals, then 4) give the edited text back to the GAN as seed text. The cycle repeats at least three times. The project is concluded by 5) participants map each version of their paragraphs (numbered to show the sequence of iterations describing the evolution of the text) on a Jamboard for reflection and sharing.

As a quick culminating experience, participants engage in a brief gallery walk, in which they simply scroll through jam board pages and leave colored tags (like post-it notes) to compliment creators for a) creativity (thinking outside the box), b) craftsmanship (careful fine tuning), c) humor (playfulness, awareness of the absurd), d) tone (preservation of emotion, mood, or a clear theme).
As a final whole group discussion to achieve learning objectives 4 and 5, the group will discuss collaboration with GANs as a process. Participants will be prompted to address the driving questions. How would you describe this process? How is writing with a GAN different from writing with a pen or keyboard? How are the cognitive activities different (i.e., planning, editing)? Once participants are comfortably examining human cognitive work involved in collaborating with GANs, shift the conversation to discuss the big picture (learning objective 5). GANs are highly responsive to our behavior and can thus easily interact with us in real-time, but the initial work feels clunky and awkward as neither the human nor the AI has developed a familiarity with each other's abilities and limitations. Practice and experience can improve the flow of this relationship. What advice would you give someone learning to collaborate with an AI? How can this advice be woven into your message?

Throughout, we will have participants synthesize their lessons learned and finalize a message for the educators and students they work with about the potential and implications of collaborating with generative AI. Participants will leave the workshop with this coherent message as well as a portfolio of open, web-based learning activities that model generative AI.

Timing & Process
10 min - Show short video on careers with generative AI and discuss impressions
10 min - Participants explore AI Careers video library
10 min - Small group discussion
10 min - Whole group discussion
5 min - Break
5 min - Introduction to portfolio of tools and learning activities
10 min - Small group activity and game
5 min - Whole group discussion
15 min - Individual activity / condensed mini-project called “Storytelling with AI”
5 min - Gallery walk
5 min - Whole group discussion

Supporting research

Utility of correct perceptions of AI - Pew Research; Automation in Everyday Life: https://drive.google.com/file/d/1-8eG5-eWcJm3sQd0RVd3FOWmOFjtt7Oa/view?usp=sharing
Smith, A., & Anderson, M. (2017). Automation in everyday life. Pew Research Center.

Collaborative Intelligence: Humans and AI Are Joining Forces
https://hbr.org/2018/07/collaborative-intelligence-humans-and-ai-are-joining-forces
Wilson, H. J., & Daugherty, P. R. (2018). Collaborative intelligence: humans and AI are joining forces. Harvard Business Review, 96(4), 114-123.

Implementation of AI Education for Secondary Schoolers: https://drive.google.com/file/d/1B8wOHdCSRxulM0pPCX3blCnwDwZMaq6W/view?usp=sharing
Forsyth, S., Dalton, B., Foster, E. H., Walsh, B., Smilack, J., & Yeh, T. (2021, January). Imagine a More Ethical AI: Using Stories to Develop Teens’ Awareness and Understanding of Artificial Intelligence and its Societal Impacts. In RESPECT 2021 virtual conference; Annual Conference on Research in Equity and Sustained Participation in Engineering, Computing, and Technology.

An Overview of Creative AI: https://www.media.mit.edu/projects/creative-ai-a-curriculum-around-creativity-generative-ai-and-ethics/overview/