Explorations in Artificial Intelligence

Although I didn't pursue an academic path in Artificial Intelligence, I purposefully navigated my career to better understand how humans interact with computers on a practical level in the hopes that one day I would get to work with it. This journey was significantly informed by my work as a UX Designer for large enterprises, where I leverage my interest in behavioral psychology as much as possible into my day-to-day tasks and am in constant pursuit of balancing System 1 and 2 thinking for users attempting to accomplish goals.

New Skills

Prompt Design Processes

  • With a recent successful completion of MIT xPro’s course in Generative Computational Design I learned to utilize LLMs to transform information into other more machine-readable forms of information via transformations into graph data structures, applications of context-free grammars, and other techniques enabling the creation of high-quality synthetic data, iterative design algorithms, and design spaces.

  • Through explorations utilizing Generative AI tools (MidJourney and ChatGPT) to create and speed up the development of an art and design brand I developed a deep understanding of how iteratively expanding and contracting a design space can be applied through different means. I am currently experimenting with utilizing an emoji translation system to explore novel HCI interactions.

AI UX Design Insights

  • I’ve conducted several tests with Operators (my preferred term for AI users) to evaluate receptiveness to different types of interactions as compared to base models with my prompts applied over them. By generating very comprehensive prompts and iterating over time, creating invisible conversational experiences that nudge users towards ambiguous end goals seems to be a successful interaction paradigm to pursue.

Seeking Answers Beyond My Day Job

My curiosity extends beyond my professional responsibilities, leading me to explore the specific value that interacting directly with a large language model (LLM) can offer humans. I am particularly interested in how we can incorporate these technologies into our everyday lives in a way that is not only beneficial but also safe—both in the present and several layers of abstracted interaction away from the original interaction.

Initial Experiments with GPT-4

In late 2023, I began probing GPT-4 to understand its logical consistency and relevance on topics like human behavior and learning. Initially, I found it lacking, so I embedded a set of scientific papers within its knowledge base to fill the gaps. After a month of iterative questioning and logic refinement, the model's responses aligned with my expectations. I then introduced the model to users to assess their interactions. The findings mirrored my earlier prototypes from 2017, revealing a high level of confusion and disappointment with the chat interface, hindering meaningful work for non-expert users.

Trust and Collaboration

In 2017, I worked on a Wizard-of-Oz prototype for Gap using a publicly available voice chat prototyping tool. To make the prototype feel like a good experience, it became quickly apparent that crossing an uncanny valley via smart design would be necessary. While interacting with the prototype, I realized that building trust between a system and its users is crucial, especially when it resembles human interaction. I hypothesized that early design work for GenAI systems would focus on obfuscating the chat interface to build trust or to make up for a lack of flow design capability, which I believe might also obscure the technology's untapped potential. My facilitator's bias leads me to believe that thoughtful communication and collaboration are key to extracting maximum value from these systems—an idea rooted in the notion that collaboration allows human civilization to thrive in the first place.

Your [apprentice]

After extensive testing and iteration, I launched my first publicly available GPT on the GPT store: Your [apprentice]. This tool is designed to be collaborative, helping the Operator understand everything needed to make informed decisions. Your [apprentice] guides users through goal setting and strategizing, addressing their pain points and contextual challenges.

Success Stories

  • A Grandmother: Created songs and bedtime stories for her grandchild with the proactive assistance of her [apprentice], overcoming the “generic advice” from standard GPT-4.

  • A Medical Device Salesperson: Used his [apprentice] to manage supply shortages and logistical challenges with “more empathy and situational awareness” than standard GPT-4 responses.

  • A Documentary Journalist: Found relevant sources and story leads for a piece on opioid addiction in women's prisons, thanks to her [apprentice]'s quick understanding of her angle.

  • A Childcare Provider: Received contextually relevant business process improvement suggestions, better aligning with her vision than standard GPT-4.

Early Wireframes of Future [apprentice] UX

Above: Wireframes for the visual version of Your [apprentice].

My plans for Your [apprentice] are to have every input generate a UI with contextual actions available to help Operators navigate through in order to enable even novice users to accelerate their learning, strategic planning, and task management.

In the example above, an Operator arrives in a cold-start state with a pre-populated list of popular prompt subjects that span a variety of topics. This is intended to allow for the least amount of onboarding possible, putting the onus on Your [apprentice] to learn more about you by being helpful. In this instance, the Operator has a specific task in mind so they input “how to fix car” as [Interaction 1].

In [Interaction 2], a real Your [apprentice] Pro output is shown, formatted in the wireframe to showcase the structure of the prompt output when it is applied to a visual vs textual hierarchy.

In [Interaction 3], the Operator has read through some of the previous information and has chosen to open the search tool via the [Magnifying Glass] icon. Populated within the search tool are typical search-related queries. In concept, each search-related query would trigger a flow based on currently known information about the Operator. If we are in a cold-state such as this one, the suggestions are related specifically to both the original input of the Operator and the context of the [apprentice] response in [Interaction 2].

This flow was arrived at via interviews with Operators attempting to use Your [apprentice], noting moments of confusion or indecisiveness on behalf of the Operator. Operators have so far found the chat interaction intuitive, with a natural progression and flow through tasks being present throughout.

Holograph

Building on the success of Your [apprentice], I conceptualized Holograph—a broader idea acknowledging the collective value of interactions with intelligent systems. This project aims to create new experiences powered by LLMs to augment the capabilities of the humans who work with them.

My next project, Your [apprentice] Pro, is complete and awaiting release. It enhances the collaborative experience by generating various options and paths based on user input, enabling effective and creative work completion.

Lightweight Future

In March of this year, I took a month away from LLM investigation to let some things sink in. During that time I spent some time building an Instagram account utilizing artwork designed using a unique workflow incorporating generative AI, photography, historical research, and a host of other fun easter eggs. I was able to sell a few one-off prints that I made and am looking forward to jumping back in this summer to begin fully automating the operation. The current account was built utilizing generative AI for every aspect of the creative process as well as to enable myself to scale from 0 to 1400 followers within the first month. Prints are available at Lightweight Future.

My journey with AI has been driven by a deep-seated curiosity and a commitment to understanding and improving human-computer interactions. From probing GPT-4 to developing Your [apprentice] and Holograph, to making and sellmy work continues to explore the practical and safe integration of AI into our daily lives.