AI Job Generation: GPT UX/UI Designer and Researcher
Predicting a new job category: GPT UX/UI Designer, Researcher, and Consultant.
The upcoming onslaught of new GPTs, which will be created by World-Class AI Experts as well as the AI Weekend Warriors, is going to spawn a new class of job: A GPT UX/UI Designer, Researcher, and Consultant.
There is much talk about how AI is going to kill jobs. But I've noticed an emerging need for a new type of job altogether even as I've been building some GPTs and being asked to review others'.
First, Distilling the GPT Framework
Before getting into this new role, it's important to distill using a GPT into three (and sometimes four) components so we have a baseline understanding of where this role could best fit:
#1. The "Data"
This is going to be what the model is trained on. Things like Bard and ChatGPT are trained on their own unique (and massive) datasets. X's/Twitter's new AI, Grok, is uniquely leveraging the wealth of content that has been on Twitter over the years. The uniqueness of each dataset is going to be the foundation upon which a GPT can be built, and it will affect the abilities and output.
You can also introduce your own data to the model (transcripts from interviews, blog posts you've written, quantitative datasets, etc.) to further augment this base.
This can also be augmented through plugins, which I'll touch on a bit down below.
The greater relevance and quality of input, the greater the foundation for the GPTs usefulness.
#2. The "Machine"
This layer is the actual GPT programming, which will be the behind-the-scenes instructions that guide it when sifting through and compiling information from the "Data" layer.
It determines the types of information used, the ways that data is evaluated, and how the GPT will output information/content back to the user.
The greater the precision, efficiency, and relevancy of the GPT instructions, the greater the impact of the output.
#3. The "Interface"
This is the layer where the user interacts with the GPT through prompts/prompting, inputting their need and any additional information/context in order to achieve an end-goal.
When using a base generative AI tool (like ChatGPT) directly, this step requires a lot of forethought and careful wording in order to extract the most useful output. However, one major benefit of GPTs are their ability to walk (talk?) the user through the process.
Inputs can range from text, to voice, to data, to imagery (and beyond).
The user's ability to thoughtfully and articulately provide input to the GPT is still critical, but now there is a shared onus of responsibility for elevating that 'conversation', as the GPT's designer has to architect those interactions by "pre-prompting" the Machine on how to interact with and provide output to the user.
Optional #4. The "Plugins"
For the purpose of this post, we'll leave these out, but essentially they are 3rd party tools that interface with the GPT that can perform independent actions through APIs. These actions can be either retrieval (seeking information from other places to augment the Data layer) or activations (performing a task), and would essentially be nodes coming off of the Data layer.
A few examples:
A retrieval action might be to review a certain customer's or respondent's behavioral data, check the weather forecast, review recent news articles, etc.
Activations/task actions can be having the GPT send emails based on certain conditions, book flights, create a file, tag a customer, etc.
Factors for choosing a GPT
If there are hundreds (or thousands, or more) GPTs that are being created for a specific use case, how is someone to decide which to use??
There will be 3 key factors driving GPT selection:
1. Price for usage (obviously)
2. Effectiveness: This can be self-judged, perhaps a rating system, or some sort of empirical tests for different GPTs (yielding another future role that will emerge)
3. Usability Experience/User Interface (UX/UI): How easy, efficient, and enjoyable the GPT is to use.
Price will be marketplace driven, and are dependent on Effectiveness and UX/UI (and of course branding/reputation as well)
Effectiveness will be driven and constantly iterated on by the GPTs creator and, at its core, is the primary function of the GPT, largely driven by science.
But the UX/UI is the component that requires a deeply human understanding, an empathetic orientation, and a skillful interaction engineering ability.
The New Role
The role of the GPT UX/UI designer will be to partner with the GPT developer to collaboratively design the Interface layer (what the user sees/interacts with) and its connection point with the Machine layer (the way the GPT responds and continues the process until the end-goal is met).
A GPT UX/UI designer is going to need to:
1. Be fluent with AI models to understand their limits, as well as know how to maximize engagement with the model. (hard skills: technical understanding)
2. Have a systems mindset, to understand the infinite ways a given interaction with a GPT could go in order to anticipate implications for designing what rules should govern that interaction/conversation and how to do so in a way to achieve the optimal end-goal. (soft skills: strong logical thinking, problem solving abilities, flexibility, human understanding, empathetic approach)
3. Be deeply empathetic and human-centric when designing and evaluating/testing the GPT UX/UI, both from their own experience in testing the GPT but also in capturing user feedback through research and testing. (hard & soft skills: research, insights, strategy)
The testing can be qualitative, quantitative, and observational.
My instinct is that a GPT will never be "finished" or "finalized" - it will be constantly evolving across all layers of the GPT. This means the UX/UI evolution will also be continual as a GPT is iterated to increase its effectiveness and usability.
The role will apply to commercial/external GPTs (made by pros, for pros) as well as internal (to drive a company's IP secret sauce/defensibility).
There will also be three ways the role is going to be needed: At the giants like Salesforce, McKinsey, and Adobe (requiring multiple roles spending a lot of time each focused on a few commercialized GPTs), in the long tail of smaller companies like agencies, startups, and hobbyists (requiring a singular role spending a moderate amount of time spread across a handful of internal GPTs), and at AI-focused consultancies (depending on the size of the consultancy, requiring multiple roles spending a lot of time focusing on a few client GPTs each).
--
While it's [very] early on with GPTs, still interesting to consider the 2nd and 3rd order organizational effects they will have as they mature and become ubiquitous.
Would love any additive thoughts, questions, or other considerations you may have as we continue to learn in public together.
Onwards…
Matt Walker