Designing in the age of ChatGPT
Disruption, opportunities, and the evolving role of design.
What’s going on
The tweet above refers to ChatGPT-4, which was released on March 14th. Unless you have been living under a rock, you have probably heard about or even used ChatGPT 3 or 4. ChatGPT is the interface to Open AI’s GPT (Generative Pre-trained Transformer) large language model (or LLM), currently in its fourth iteration. ChatGPT is one of a number of language models that are related to the image, video, and sound generation models such as Midjourney and Dall-E that all fall under the umbrella term of generative AI.
This article is my attempt to understand, as a designer, what is going on with Generative AI at the moment, focusing specifically on ChatGPT, and what this might mean for the design industry. The technology, and its applications, are evolving at an incredible speed, so the examples shown may well be out of date next week, but I think the core ideas that underpin them, and shifts in how we work, will remain.
How it works
The GPT model was trained on vast amounts of internet text to generate human-like conversational text in response to user questions, or prompts. The user-interface is a deceptively simple single page chat interface where users can ask complex questions (known as prompts) and, crucially, follow up with additional queries that retain the context of the conversation.
It was developed by Open AI, an American AI research lab that also created the DALL.E AI image generator. Microsoft recently invested $10 billion into the company and is already starting to integrate the technology into Bing and Office products. Google released a similar model called BARD and Meta also recently announced their version called LLaMA.
The ChatGPT model itself is not that new, with GPT-1 announced in 2018. It is essentially an incredibly capable auto-complete that works by predicting the next word based on the words it has seen so far. It takes in a sequence of words as input and tries to predict the next word. It does this by analyzing patterns and relationships between words. The more words it sees, the better it gets at understanding context and generating relevant text.
“ChatGPT is merely pulling out some ‘coherent thread of text’ from the ‘statistics of conventional wisdom’ that it’s accumulated”. — Stephen Wolfram
Generative hype or genuine innovation?
The pace of adoption and innovation has been incredible by many measures. Bill Gates thinks this “is as revolutionary as mobile phones and the internet” and many more agree. The rate of adoption tells a story in itself; it took ChatGPT only 2 months to reach 100 million global monthly active users. For comparison, it took TikTok, (next in line) 9 months to do the same and Instagram (third) 30 months.
There is no doubt a massive amount of hype with this. There has been similar hype before, most recently with Web3.0 and the Metaverse but I do believe it’s different. ChatGPT (and related models) kicked down the door with tangible, valuable and easy to understand examples that people could intuitively understand and, importantly, easily try out and see value.
Within DataRobot, we are already starting to use ChatGPT both as a tool when collaborating internally and also as a feature in our product. We recently demoed a GPT-powered (specifically ChatGPT in Azure OpenAI service) Assist feature in our code-first Notebook. This augments workflows for data scientists. It suggests, corrects and finishes code with a few natural-language commands. It’s already showing incredible promise at increasing the speed of workflows for data scientists. The idea of a more pervasive GPT-style assistant that works across the platform is something we are currently exploring more.
In order to understand how this may play out in the longer term, it’s helpful to look at examples elsewhere. There is a lot of overlap with our Notebook Assist functionality and GitHub’s CoPilot that was launched 2 years ago. Andrej Karpathy, the former director of AI at Tesla and ‘one of the most elite AI developers’, published this tweet last year, it made a lot of people sit up and take notice.
“Copilot has dramatically accelerated my coding, it’s hard to imagine going back to ‘manual coding’. Still learning to use it but it already writes ~80% of my code, ~80% accuracy. I don’t even really code, I prompt. & edit.” — Andrej Karpathy
Satya Nadella later responded to this by saying:
“This is one of the most elite AI developers who’s saying that they’re being that much more productive. This is literally changing the productivity curve for software engineers.” (WSJ)
We think that we can similarly change the productivity curve for data scientists in the same way that copilot changed that curve for software developers, but it doesn’t stop there with ChatGPT.
In addition to the Notebook Assist feature, we are also using ChatGPT to explain what is happening with AI models in our platform. The conventional approach is that a data scientist builds a model using our UI and then generates a number of visualizations, or insights, to understand the performance of those models and subsequent predictions. We are using ChatGPT to generate natural language descriptions of what is happening in behind the scenes and explain why the model predicted what it did.
Additionally, by allowing users to ask follow-up questions and delve deeper, they receive recommendations on what to do next. This approach has the potential to be much more disruptive compared to the Notebook integration.
ChatGPT will not only boost the productivity and engagement of many products, but it will also profoundly alter the essence of some. Benedict Evans describes this disruptive opportunity well in the context of what incumbents in other industries:
“Microsoft & Google adding generative AI into office apps is a classic pattern of incumbents making the new thing a feature. But the new thing generally also enables completely new ways to solve the problem. ‘Easier spreadsheets’ is less important than ‘why is that a spreadsheet?’” — Benedict Evans
By going back to first principles and asking ‘why is that a spreadsheet’ (or any other product or feature) in the first place, ChatGPT enables us to re-think the very fundamental proposition of many products and services. In addition to ChatGPT itself, we are already starting to see the potential power of the combinatorial effect when integrated with other services as plugins. It’s still early days but I think we have only scratched the surface of potential applications and will see much more growth here.
Responsible innovation
As a product designer, I find this potential incredibly exciting but also a little intimidating and concerning. At DataRobot we are, like many other companies, still grappling with the fact that ChatGPT is fundamentally a probabilistic tool that is sometimes wrong.
As a design team, I think a lot of our focus will be in managing potential output errors, or hallucinations as they are known. This may mean ensuring that any interfaces have adequate feedback loops in place to enable people to report inaccuracies, which can in turn fine-tune the AI models an improve performance. We also need to ensure that we combine AI solutions with sufficient human oversight. This is particularly important for critical applications such as healthcare where we need the highest levels of accuracy.
I have been thinking about this quote by Kevin Kelly (founder of Wired magazine) a lot lately and worry that the technology is racing ahead faster than we can understand all of the implications.
“We are morphing so fast that our ability to invent new things outpaces the rate we can civilize them” — Kevin Kelly
OpenAI and other companies are already making some good progress in terms of establishing some checks on the technology. Google has published their AI Principles that outline the criteria for assessing AI applications and Microsoft published the HAX Toolkit to help AI builders to create effective and responsible human-AI experiences.
“We are optimistic about the incredible potential for AI and other advanced technologies to benefit current and future generations. We also recognize that advanced technologies raise important challenges that we need to address clearly, thoughtfully, and affirmatively.” — Google AI Principles
How will this impact design?
The examples above show that tools like ChatGPT will not just change what we do, but they will also fundamentally change how we do it. It seems increasingly likely that such technologies will change many of our roles in design in significant ways.
There are parallels in history that we can learn from. For example, CAD and the internet changed how we design but it didn’t result in mass redundancies. The image below shows draftsmen (and it seems they were all men) working at the Ford motor company in 1942. This room full of people has probably been replaced by a handful of designers and engineers. The reality is that job and role definitions change over time. Yes there aren’t as many professional drafts-people around today but there are a whole host of new roles that didn’t exist 20 years ago. As people continue to debate whether they should be called UX or (digital) Product designers, don’t forget that these roles are very new. In fact there are now more designers than ever before — the total number has doubled in Ireland in the last 5 years alone.
To understand how our roles in design might change, it’s worth looking for parallels in other disciplines. The role of technical [software] architects emerged out of the need to make sense of the complexity of delivering large software projects. Perhaps this is suggestive of how product design roles like mine might evolve. The role is focused on framing problems, establishing constraints and defining the criteria for success (e.g. prompt engineering). What happens in between could be largely procedural and AI driven by tools like ChatGPT. There are already several examples emerging that point to this.
Diagram is a design tools company that is re-imaging UI design in the era of generative AI. Their vision is to create radically powerful tools to improve the lives of UI designers and allow new people to design their digital world. Their products include Automator, which lets designers create their own automation tools without writing any code. Genius (not yet released) combines the latest AI models and engineering to deliver an auto-completing smart assistant experience right within tools like Figma.
Tools like Google Workspace will change how we work and collaborate. This example harnesses the power of generative AI to enable people to create, connect, and collaborate like never before. Once again, the role of the designer here is to establish the framework and steer the tools towards the desired end result.
This GPT-4 demo shows how someone went from a crude sketch to a functional website using some new GPT-4 capabilities. You could imagine that something like this, if paired with your company’s design system, or generative image tools like Mid-journey or Dall-E, it could be a very powerful way to accelerate design workflows. It may not provide the polished end result but it might automate much of the grunt work needed to get to a (quite advanced) point where you can add polish.
The baseline has shifted
The examples above demonstrate that there are clear opportunities to improve how we work, as well as the potential to disrupt what it means to be a designer. While there will be cause for concern among many, I prefer to take an optimistic view that this will ultimately change how we work for the better. The reason is that these tools will enable us to create new things that weren’t previously conceivable. It’s hard to see where we might go, much like it was hard to see how the internet might unfold when Marc Andressen released Mosaic in 1993. It simply wasn’t possible to consider how that technology, when mixed with things like GPS and computers in your pocket could result in transformative products like Uber, AirBnb and Instagram.
I think Prakhar Mehrota’s description of the potential impact of ChatGPT on writing is a very helpful way to think about what this could mean for design. He outlines how tools like GPT will essentially elevate the baseline for quality, enabling more people to get to a ‘good enough’ level faster while allowing skilled practitioners to go further than they could before. It also outlines the challenge however that new entrants will face when trying to learn new skills. When you can get to ‘good enough’ with a GPT, why bother spending 10,000 hours learning your craft?
I think that the answer here is that the goalposts have simply moved. What was previously good enough, is now the baseline for entry. Much like digital tools like Quark and InDesign had a profound effect on desktop publishing, I believe that generative AI tools will elevate the baseline for design. They will open up a new opportunity space to create new things, go further and deeper than was previously possible.
The next generation of designers that rise to the top will be those who embrace generative AI tools so that 1+1=3. Humans are remarkably adaptable so people will come to expect this baseline as the norm very quickly. Keep in mind however that generative AI tools are fundamentally derivative so the ability to combine AI outputs with personality, humanity and authentic innovation (by this I mean things that didn’t exist previously) will elevate work from the ‘good-enough’ baseline.
Another reason why we need to push ahead of the baseline is that these models will need to be fed new information in order for the outputs to stay fresh and relevant. There is a very real risk if we don’t do this that we could become creatively blunt and generative-AI tools could loose their impact and become stagnant. David Truog, analyst at Forrester describes this very well.
“The content that the neural networks underlying generative AI produce is the result of remixing material created by humans in the first place. So unless humans continue to create original material that is fed into neural networks to refresh them, generative AI’s pools of inspiration will become stagnant, growing only incestuously through the addition of material that itself comes from generative AI.” — David Truog, Forrester
Human-machine differentiation
There are various aspects of design that GPTs have yet to excel in, which can provide insight into our evolving role as designers. Skills such as framing problems, considering broader technical, business, and user contexts and constraints are vital for ensuring successful adoption.
Furthermore, understanding people, interpreting emotions to discern their true intentions, and probing deeper when clarity is lacking are essential design skills that are difficult to automate.
Additionally, the capacity to present and promote a vision is crucial — many great ideas begin as delicate, partially formed concepts that require nurturing to flourish. GPTs may overlook much of this subtlety. While they are highly adept at synthesizing information and drawing upon their knowledge base, they still lack the ability to genuinely reason.
Where to next
Whether or not you subscribe to all of the hype, it feels clear that we are at an inflection point and tools like GPT are here to stay. As a designer, there are some very obvious threats to what we do but at the same time, there are incredible opportunities to create new things and work in ways that just weren’t possible only a few months ago. We now all have an all-knowing, incredibly insightful and indefatigable creative partner on hand at all times which is game-changing. Individuals can now do the same work as studios, novices could conceivably compete with experts in some areas.
Here are a some key take-aways that that I think will be important to make the most of this new era of augmented creativity.
- Disrupt yourself
Embrace these new tools: Get acquainted with generative AI tools like ChatGPT and prompt engineering. Explore how they can augment and disrupt your workflow before someone else does it for you. - Find the AI baseline
Use generative AI to establish a solid foundation, find out quickly what the established knowledge base is and then identify areas where you can add value. - Enforce limits
This is especially important if starting off in your career. While the temptation to use generative AI tools will be strong, it will be important to set boundaries and force yourself to focus on developing core skills without the aid of tools like ChatGPT. While sometimes painful and inefficient, dead-ends and bad ideas are important parts of the creative process that are crucial to enabling us to create great work. - Push beyond
Invest time and effort into pushing the boundaries of AI-generated designs to create novel products and experiences. Use it as a creative partner to explore new territory and challenge yourself with to find the truly innovative ideas. - Human-centred AI
Since AI services look to become a central part of everyday life, it’s critical that as designers, we keep people’s needs front and centre. Consider human-centred design best practices to ensure that your AI solutions are designed to be accessible, inclusive, and effective for users.
In conclusion, the rise of generative AI tools like GPT present both challenges and opportunities for the design profession. By adapting our roles, embracing new technologies, and continuously striving to elevate AI-generated designs, we can redefine our profession and unlock the full potential of generative AI.
This article was written with the assistance of GPT-4.
—
Additional notes:
As noted above, I have been using GPT-4 to help me write this article and that in itself was an enlightening experience and helped me understand it at a deeper level. You might think that all I needed to do was enter a few prompts and bang, there’s your article. I did try to do this (for research purposes, obviously), but the reality is that badly formed ideas well written are still bad ideas. ChatGPT helps you move along but it doesn’t organise your thinking for you. I think this quote by Leslie Lamport still rings true, with our without GPTs.
“Writing is nature’s way of telling us how lousy our thinking is.”
I for one welcome our new GPT overlords and look forward to collaborating with them on more articles, designs and whatever else it is that I will be doing in the future.
John Moriarty is a Dublin-based designer leading the design team at DataRobot, a US-based start-up. DataRobot is an enterprise AI platform that helps data scientists to build, deploy and manage machine-learning models. It is used by 40% of the Fortune 50 companies, such as BMW, Autodesk and the Boston’s children’s hospital to help make critical business decisions.
Before this, he co-led the Fjord design team in Accenture The Dock and prior to that he worked with HMH and Design Partners.
