The launch of ChatGPT in November 2022 got everyone talking about AI. At first there were rumblings about how it would make copywriting and web design much easier. Then came talk of how AI content generators would render writers and designers obsolete. In this post, I want to examine how accurate these claims and concerns are.
We’ll start by digging into what this new generation of chatbots are all about and what they are and aren’t capable of. Then I want to have a frank discussion about the risks of using this technology in web design.
What is ChatGPT?
ChatGPT is an AI chatbot built by OpenAI. Here is what the application looks like:
The basic functionality is similar to the chatbots we build into digital products. Users input their queries into a chat field. The AI chatbot then responds.
However, that’s where the similarities end.
For instance, this chatbot conversation on the Aveda website is commonly what we encounter when we engage with chatbots today:
Their prompts may sound conversational in tone. However, there’s very little leeway for what users can ask or say, which is why they’re often given pre-written answers and choices. In addition, users reach a point where they need to be passed onto a human support representative.
That is not the experience users have with ChatGPT.
This chatbot is the third generation of generative pre-trained transformer (GPT) technologies. This large language model (LLM) AI has undergone some serious training.
Here’s a diagram from OpenAI that explains the process of supervised and reinforcement training used to fine-tune the AI:
A lot of work has been done to perfect the chatbot’s ability to sound human, comprehend complex inquiries, and piece together appropriate responses. And it’s still learning, thanks in part to all the people flocking to play around with the prototype while it remains free.
While ChatGPT is a text generator, it has an AI image generator counterpart called DALL-E. The concepts are similar — users ask the AI to perform a task and it responds. However, DALL-E’s specialty is in the generation of images and artwork based on a natural language description.
What AI Generators Cannot Do
I’ve seen people declaring “R.I.P to web designers” now that ChatGPT is here. However, if you look at what AI content and image generators can actually do, you’ll see that there’s a lot of misleading information out there. I think a lot of the hype around it has people misunderstanding its capabilities as well.
For instance, I recently saw this post on Twitter from @mertsusur. My initial thought was, “Wow! I can’t believe ChatGPT can do that.”
ChatGPT for web design pic.twitter.com/9PZsJvgtvX
— Mert Susur 🦇🔊 (@MertSusur) February 27, 2023
I was right to be shocked because ChatGPT can’t do that. In the tweet’s GIF, the first prompt entered is this:
“An onboarding screen for a dog walking app”
When I put that exact prompt into ChatGPT, I got a wall of text with five written steps:
ChatGPT is a content generator, so it can’t create designs for you. What it does instead is provide a text-based response that describes what you should design, write, or code.
In terms of the guidance provided on what to write in the example above, ChatGPT hasn’t done a great job. For an onboarding screen, this is far too much copy. The information is accurate and the copy is free of errors, but it’s not something that could reasonably be used — not just in terms of space, but in terms of users’ memory capacity and attention span.
So I went over to DALL-E to see if it could do better. This is what I got using the same prompt:
Perhaps if I had given DALL-E more specific input it could do a better job of creating mockups for an onboarding screen. However, the basic description seems to be insufficient for the AI. Either that or DALL-E doesn’t know much about modern design.
Upon closer inspection, the tweet that had been labeled as “ChatGPT for web design” was a promotion for another AI called Galileo, which has yet to launch. I suspect this tool will be to design what Grammarly is to writing — an AI-powered assistant built for a specific use case as opposed to a general chatbot assistant like ChatGPT.
Even then, I don’t know if AIs built for specific tasks like design will actually be able to produce anything close to the onboarding screen or other designs from the GIF above.
I’m wary of anything that promises to make perfect, modern, and responsive designs with an 8-word prompt. I’d be just as wary if a designer said they could run with it if I gave them that succinct of a request. There are just too many complexities and extenuating circumstances in design and marketing to be able to create something so easily that is pixel-perfect and usable.
What I do know is this: As of writing this, neither ChatGPT or DALL-E can do what you see in the GIF above. So what we need to focus on next is what ChatGPT is good for… If anything.
What Designers *Can* Use ChatGPT For
I think there are some ways in which you can put tools like ChatGPT to use. However, I don’t think there are quite as many as some people have been suggesting.
I’ve seen posts with lists of 20+ things that ChatGPT can do for web designers. Things like:
- Write copy for a website.
- Code an entire website.
- Create wireframes and user flows.
- Teach you about web design, principles, and trends.
- Provide you with visual inspiration.
- Choose the right UX research and testing methods.
- Find the best tools or resources to use.
For starters, you 100% cannot use ChatGPT to write copy for a website. I’ll explain why in the next section.
Secondly, it seems strange to me that anyone would open up a chatbot to ask it something that will take at least a few seconds for the AI to type out. Why not just input the same query into your browser bar and search engine and get numerous expert responses in a second or two? Especially if those responses come with descriptive visuals and useful anecdotes?
If you want to know how ChatGPT can be of use in web design, here are some ideas:
1. Create Placeholder Content
We all know the struggle of waiting around for clients (and sometimes copywriters) to send over the copy they promised for a website or app we’re working on.
You could just put the job on hold until you get what you need. However, that could mess up the timeline for the job as well as others you’re working on.
Rather than add Lorem Ipsum to the pages as a placeholder, you could ask ChatGPT to whip something up for you. For instance:
“Can you write copy for an About page for a law firm in Miami”
If you do this, it’s strictly so you can build the design and features around the amount and type of copy you expect to receive. It all needs to be replaced with custom-written content before you publish it to the web though.
2. Check Out the Competitive Landscape
Market and competitive research are important first steps in the web design process. Now, ChatGPT can’t provide expert analysis on UX design or why one digital product performs better than another. What it does well, however, is deal in documented facts.
If you want a quick view of the competitive landscape for a product you’re building, ChatGPT can help.
For example, I asked the AI:
“What are the best dating apps and why?”
It declined to answer my question directly because, according to ChatGPT, it’s an AI and unable to give an opinion. However, it was able to briefly list out the most popular dating apps:
What’s nice about this list is that the key differentiating features of each app are spelled out. So I can see:
- Similarities between the most popular existing solutions
- Different ways you can handle critical functionality in these types of apps
- Gaps in features and functionality that none of these solutions have been able to solve
If you’re feeling stuck or overwhelmed by all the information you’re working with, ChatGPT might be able to whittle it down to something more manageable.
3. Create the Basis for a User Persona
Nothing will be able to replace the research you do into your client’s business and the target customers they serve. Nor will the time you spend interacting with those users or evaluating the data you gather from their interactions with the product you built.
However, ChatGPT can give you a place to start from when it comes to generating user personas. For instance, here’s what I got back when I asked for the following:
“Create a target user persona for a doggy daycare”
The response includes a lot of the information you’d need to include in your user persona workup. For instance:
- Age
- Gender
- Occupation
- Location
- Hobbies
- Goals
- Motivations
- Challenges
- Online Behavior
There’s some good information here. However, this user persona description doesn’t feel or look very human. You and your teammates will want something you can look at and instantly get a sense for the person you’re designing an experience and product for.
To do that, you’d want to make these responses shorter, convert some of the text into quotes that sound like the person you’re designing for, and add visualizations. You'll also need to get to know the users themselves to ensure that this description is accurate. But again, it’s a good start.
4. Improve Internal Communications
Do you ever worry that the email you’ve sent to a prospect, client, teammate, or collaborator won’t come off sounding the way you intended? If so, you could use ChatGPT to help you change the tone or sentiment of your message to ensure that there are no misunderstandings.
Another way to have ChatGPT fix your internal communications is by having it provide you with a message framework.
For example, let’s say that you’ve shared a new logo design with your client. In return, they said they hated it. You’ll likely feel disappointed, confused, and insulted. You might even feel angry.
Rather than rush to shoot off a defensive and emotional response, you could have ChatGPT help you take a more professional approach:
In this example, the AI tells me to:
- Thank the client for the feedback.
- Ask for clarification.
- Take a step back.
- Offer solutions.
- Stay professional.
This is one of the advantages of using a chatbot instead of, say, going to a coworker who might have the same type of heated response as you do. The AI will instead lay things out logically, objectively, and unemotionally.
5. Write and Debug Simple Code
ChatGPT can’t custom-code an entire website or application on your behalf. However, if you need help generating code and don’t want to have to sift through the top Google searches and resulting web pages to find the snippet, the AI could offer a quicker solution.
For example, let’s say you can’t remember how to add an anchor link. ChatGPT won’t just explain how to do it, it’ll give you sample code to use for the link and anchor:
You could also ask it to perform more complex coding tasks as well as to help with debugging. There’s no guarantee that it’ll know exactly how to execute the command you’ve given it, but it may be worth trying if you don’t have a go-to reliable resource to help you out.
Other Tasks
The list above is by no means exhaustive. What it should do is give you an idea of what the AI’s strengths are.
- For instance, while it can’t and shouldn’t write anything from-scratch for you, ChatGPT can provide you with starter frameworks for what you do need written.
- It can also help you refine what you’ve written. That said, if all you need is some proofreading assistance, there are other AI tools — like Grammarly — that are built for that purpose and will do a better job of it.
- ChatGPT can also help shrink down vast amounts of information and make it easier to analyze. This would be especially useful if you’re working on pitching a client or drawing up a proposal and you don’t want to spend a ton of time on something that might not turn into paid work.
Why Designers Need to Be Careful About Using ChatGPT
Before we wrap up here, I want to address the limitations and risks of using ChatGPT in web design and, really, any kind of online content creation. Here’s what you need to know:
1. ChatGPT Isn’t Omniscient
The AI will occasionally explain that because it’s an LLM that it can’t provide opinions or recommendations. So while it can deal in hard facts that it’s been fed by its trainers or that it’s extracted from other resources, subjective matters are not its specialty.
For instance, Matthias Ott asked ChatGPT for a book recommendation for someone wanting to learn more about typography. This is how it responded:
“I'm sorry, but I am not able to recommend books as I am a large language model trained by OpenAI and do not have the ability to browse the web or access external information. I am only able to provide information based on the knowledge I have been trained on, which has a fixed cutoff point.”
It also lacks expertise. One of the reasons why creative professionals who niche down get paid more is because they are specialists. They write, design, and otherwise create from genuine expertise and deep knowledge. AIs cannot do this. What they will give you is generic and basic.
2. ChatGPT is Derivative
If you value originality and creativity, and clients are paying you well for those capabilities, then ChatGPT is not your friend. The same goes for DALL-E.
These LLM-based AI are not creators. They take information they’ve been given and then provide responses based on their training and the parameters that their trainers have set for them.
Assume that anything you receive from these AIs has been pieced together from something someone else has said, written, or done somewhere else before.
3. LLMs Could Lead to Another Material Design Phase
Remember when Material Design first launched and every app suddenly looked like it was modeled after Google’s design system? And then it started to bleed into website design?
If designers become too dependent on the ideas generated by ChatGPT and the designs created in a platform like Galileo, we could end up with another design era where everything becomes homogeneous.
This, in turn, could lead to apathetic, disengaged, or, worse, negative user experiences. As a result, designers will be stuck with a ton of restoration work as everything that came from and was inspired by AI needs to be stripped out of their products.
4. ChatGPT Isn’t Human
I know this seems obvious. But that’s going to be a huge problem for web designers. Empathy is a critical component of UX design and it’s something that no machine can fake.
So, yes, you can use ChatGPT as a basic reference for some things and to carry out technical tasks like creating placeholder text or code. However, when it comes to performing user research, running and analyzing tests and experiments, and creating human-first designs and experiences, ChatGPT will be useless.
Designers will always need to be at the helm of website and app development because of this.
5. ChatGPT Can Be Inaccurate, Inconsistent, and Biased
According to ChatGPT’s own maker, it is an unreliable source. In the November announcement of the LLM chatbot, OpenAI says:
“ChatGPT sometimes writes plausible-sounding but incorrect or nonsensical answers. Fixing this issue is challenging, as: (1) during RL training, there’s currently no source of truth; (2) training the model to be more cautious causes it to decline questions that it can answer correctly; and (3) supervised training misleads the model because the ideal answer depends on what the model knows, rather than what the human demonstrator knows.”
What’s more, you can get different responses to the same question from ChatGPT:
“ChatGPT is sensitive to tweaks to the input phrasing or attempting the same prompt multiple times. For example, given one phrasing of a question, the model can claim to not know the answer, but given a slight rephrase, can answer correctly.”
Unlike something like the search engines which often provide the same results even if you tweak the arrangement of words or the wording itself, ChatGPT doesn’t seem to have that agility built into it. There have also been issues reported from some users that bias has tainted ChatGPT responses.
6. ChatGPT Could Put Everything You’ve Done at Risk
You want to be very careful about where and how you use what you receive from ChatGPT. I suspect we’re going to see a rise in plagiarism and copyright infringement allegations as a result of this.
That’s not the only thing to be worried about either. There’s talk that ChatGPT could be bad for SEO. For example, I received this email from a client recently:
“We’ve decided against using AI writing tools. The main argument is that ChatGPT was developed by Google's direct competitor.
At this very moment, we're sure that Google is working on algorithm changes that will either give priority in search results rankings for human-written content or that will outright disregard ChatGPT content from ranking.
There are already online tools that can detect if a text was written by an AI, such as GPT Zero - which already has 1 million users. We're almost certain that Google will implement something similar in its algorithms.”
This isn't surprising in the SEO world. There's been talk for the last year that Google was going to start penalizing bot-written content. Draft, for instance, wrote about whether Google can detect AI content and Waya discussed how human content was going to be prioritized in search results.
It’s too early to tell now, but I suspect we'll start to see massive waves of ChatGPT backlash start around late 2023. And SEO and copyright infringement lawsuits will likely be the driver behind it.
Wrapping Up
ChatGPT is still a really new technology. While there’s a lot of hype around it taking designers’ (as well as writers’) jobs, I just don’t see that happening. There are tons of limitations in terms of what it can do and there’s also a lot of risk in taking what it’s given you and publishing it.
So does that mean you should avoid it altogether in web design?
If you look at ChatGPT as nothing more than a really smart and fast personal assistant, I think you can make good use of it. But to think of it as something that can replace what you or your peers do would be a mistake.
Also, if your concern is with how efficiently you’re working or the quality of your output, I would advise you to look at other AI tools on the market. As I mentioned above, there are different tools developed and in development for specific use cases like copywriting and web design. They’ll likely be more useful to you than a general purpose chatbot like ChatGPT.
No comments:
Post a Comment