Revolutionising Human Centred Design: The AI impact

00;00;09;14 - 00;00;26;21
Jake Fernandez

Hello and welcome to this episode of the CGI Experience Podcast, where we will be exploring how A.I. will revolutionise human centred design. I am Jake Fernandez, a UI designer in our UK HCD team, and I'm pleased to be joined by my colleague Alex Lowe. If you'd like to introduce yourself.

00;00;27;13 - 00;00;36;19
Alex Lowe

Yeah, hi. So my name is Alex Lowe and I'm a user experience strategist working as part of the human centred design team here at CGI. So, looking forward to our discussion today.

00;00;36;26 - 00;00;54;29
Jake Fernandez

Well, there's a few topics within tech at the moment with the level of intrigue that A.I. has around it. It has been a rapidly growing field over the past few years that has gained serious traction off the back of emerging technologies such as Chat GPT and Midjourney. But to kick things off, Alex, what are these technologies all about?

00;00;55;16 - 00;01;26;22
Alex Lowe

So, the term A.I. or Artificial Intelligence, with which you're probably familiar, refers to the ability of machines to perform tasks that typically require human intelligence. And you've probably heard about it being talked about a lot over recent months with A.I. tools such as OpenAI's, Chat GPT and Midjourney appearing regularly in the news. And these are examples of large language models or LLMs, and they're created by training neural networks and extremely large amounts of text data.

00;01;27;20 - 00;01;57;07
Alex Lowe

They're trained in different ways. But GPT, for example, uses an interesting method known as RLHF, and that stands for reinforcement learning through human feedback, which is essentially a human saying if response it gives is correct or incorrect, then has a reward system in place, understands to keep the correct response and dispose of the incorrect response. And it's important that these A.I.s understand the intent behind what's being asked of them.

00;01;57;14 - 00;02;26;10
Alex Lowe

And RLHF can really help with this. There are many applications and they’re incorporating layers of AI into their products, such as Microsoft's Bing using Chat GPT is now Google's Bard, which is another conversational A.I. Until something like notion uses A.I. to generate and organise content. So, I guess why is this such a massive increase in interest in these large language models all of a sudden?

00;02;27;01 - 00;02;47;01
Jake Fernandez

I think one of the main reasons GPT has become so popular is it's an evolution of search engines. So, in a search engine, a user would have to manually review the content, return to figure out what it is that they want; with something like Chat GPT the intent from the user is better understood and therefore a more relevant answer is made available.

00;02;47;21 - 00;03;12;26
Jake Fernandez

So, let's create a scenario comparing GPT to a search engine. I want to find ideas for a budget summer holiday to somewhere hot and create a seven-day itinerary. Google search may find loads of lengthy blogs trying to troll through opinions and long-winded answers. When I just really want to brainstorm. I may end up taking quite a while just to get to this point, likely exhausting myself in the process.

00;03;13;15 - 00;03;34;28
Jake Fernandez

GPT can create a response in a digestible way, however I'd like. I could have it draft a carefully written blog which hits all my criteria, or I could have it boil down to the things into simple bullet points. I could even ask it to write in a tone of voice that I engage with, like my favourite comedian in the style of a local expert.

00;03;35;15 - 00;03;44;22
Jake Fernandez

I think that the aspect of allowing it to mimic a specific tone of voice will be really, really helpful in the future of online chat bots. But what do you think about Alex?

00;03;45;10 - 00;04;21;20
Alex Lowe

Traditional chat bots are trained on product information, customer service logs, online forums in order to gain an expert level of knowledge that can provide help to a customer. The emerging tools like Chat GPT follow the same methodology, but using the vast datasets as we described earlier, which lets it do the kind of things that you were just talking about. And looking at activity in particular. as a user, I'm able to ask it to reflect it in inverted commas on a response it's given me, and it's able to kind of self-analyse, look what it's written and adjust its response accordingly.

00;04;22;04 - 00;04;24;27
Jake Fernandez

And so, what does this mean on a day to day basis?

00;04;25;20 - 00;04;45;06
Alex Lowe

Well, for everyday tasks, you could ask these elements to do things like organise your diary. Perhaps you want a personal fitness program to work around your week. You could input the details and ask it to create a workout program and a meal plan, say. Or perhaps you just want a recipe suggestion based on what's in your fridge at the moment.

00;04;46;02 - 00;05;09;27
Alex Lowe

These LLMs are really useful for summarising long form text or explaining complex ideas as well, and they're starting to get good at creating text for emails, letters, presentations, essays, etc. I've even actually used it to create some specific tests to help me learn a new language and we may or may not have used it to create some of the content for this podcast; It'll be interesting to see if you can tell.

00;05;09;27 - 00;05;25;01
Alex Lowe

And you could, if you want to get it to summarise something like the Magna Carta in, say, two hundred words, but in the style of someone like Billy Connolly, which we did try and is interesting to see what those results were. But, you know, this is just a start really, I guess, what else might be possible in the future?

00;05;25;16 - 00;05;44;23
Jake Fernandez

I think it could also be very beneficial for users who may have special requirements and helping them plan ahead. So, knowing facilities provided by locations such as disabled access or minimizing travel time between that might help a lot of people. And there's a handful of people that are already using the multitasking capabilities of these tools to earn money.

00;05;45;04 - 00;06;10;28
Jake Fernandez

So, in particular, Jackson Fall, who you can follow on Twitter; he's been acting as a human liaison for Chat GPT, asking for it to act as an entrepreneur with the intention of making as much money as possible, starting with $100. And this AI-trepreneur has managed to create a live website with content and images and sold a 2% stake in the business for $500, which effectively values that company at $25,000.

00;06;11;01 - 00;06;23;06
Jake Fernandez

I think it's a great early adoption of the technology to see how it's impacting our world and bridging into HCD through web design and content creation. But Alex, how are these technologies impacting our industry?

00;06;24;15 - 00;06;48;16
Alex Lowe

Well, as human centred designers, I think we really need to consider the implications of incorporating A.I. language models into our industry. While the technology is still in its early stages, we need to identify its strengths, its weaknesses and how it can be leveraged in our field. I think it's a real opportunity to explore new ways to optimise our workflows and enhance our creative processes, really deliver better outcomes for our clients.

00;06;49;00 - 00;07;16;24
Alex Lowe

And one of the primary applications of A.I. language models, in design I think, is probably around content generation at the moment. While the content produced might not always be perfect initially, I think it can serve as a solid foundation for iteration and refinement. This could be particularly beneficial maybe for those who might struggle with articulating ideas in a language that isn't necessarily their native tongue, for example, as it provides a starting point for further improvement.

00;07;17;12 - 00;07;43;15
Alex Lowe

This also allows designers to focus on the more strategic aspects of content creation, like tone messaging and storytelling. A.I. language models can also contribute to enhancing collaboration within design teams, I think. You know, you can really provide quick outputs from brainstorming sessions and these models can stimulate discussions, maybe inspire new ideas and enable teams to explore multiple creative directions more quickly.

00;07;44;04 - 00;08;12;08
Alex Lowe

As designers, we can use air as a collaborative partner to help us refine our idea is and discover more innovative solutions. A.I. language models hold great potential for revolutionising the design industry by automating certain tasks and streamlining processes. However, I think it's really important to remember that human creativity, the critical thinking, and the intuition that we bring as HDD designers will continue to be essential elements in the design process.

00;08;12;21 - 00;08;21;25
Alex Lowe

As designers, we must learn how to harness the power of A.I. while maintaining our unique human touch and embracing the opportunities and challenges that A.I. brings to our profession.

00;08;22;18 - 00;08;29;28
Jake Fernandez

And so, we've talked a lot about GPT, but what else is there out there for us to use, as human centred designers?

00;08;29;28 - 00;08;59;00
Alex Lowe

There are a few I can list here which aren't necessarily LLMs but still use A.I. in how they work. So, for example, maybe something like research A.I., which automates user research, streamlining the data collection process. There's something like Smart One, and this offers sentiment and intent analysis, enabling better understanding of user needs. Maybe there's something like UI-zard, which is a versatile tool that can generate mock ups from simple text.

00;08;59;10 - 00;09;42;04
Alex Lowe

It can convert screenshots into editable mock ups. It can create text for designs, generate UI themes. It can even take like a hand-drawn sketch of a wireframe and produce a mock-up based on that and even produce predicted attention heat maps to analyse user flows based on what you're designing. Perhaps something like VAS or Vas predicts where users will look on a screen providing insights into design effectiveness and then coming down the road fairly soon, it's going to be something like Adobe Firefly, which uses A.I. not only to create new images like something like Midjourney might, but also you could manipulate existing ones in novel new ways.

00;09;42;06 - 00;10;14;02
Alex Lowe

On the coding side, there's something like MS Co-pilot so that aids in coding tasks while something more like Notion, which uses A.I. to generate content like blogs and business plans and sort of learning as you type. And lastly, perhaps something like Fig GPT, this focuses on content generation for prototyping within Figma, which is really handy for us as designers. And these A.I. tools offer designers a wealth of resource for streamlining their work and enhancing their creative output.

00;10;14;05 - 00;10;44;05
Jake Fernandez

Yeah, we've been using a few of those already like Fig GPT and for me, UI’s is definitely next up on my list and speaking about HCD specific use cases for these technologies, the major one being the improvement of user experience. So being able to analyse user behaviour and provide personalised experiences tailored towards user needs, this could take away a lot of those laborious tasks for professionals and allow them to concentrate on what really matters, which is creating the best solution for users.

00;10;44;07 - 00;11;10;22
Jake Fernandez

It could also provide starting points for UX designers to move past issues like project paralysis, and those automated tasks could include gathering and processing for UX designers and researchers. It could take a long time to synthesise user research outputs, such as survey responses and the methods in which we do may not be optimal every time. So, think about creating word clouds to get an idea for user feedback.

00;11;10;24 - 00;11;35;16
Jake Fernandez

They may be a great tool, but how about getting a full descriptive analysis with statistics to back it up? You can make much clearer research backed decisions with a tool to conduct all of that hard work for you, and there's also A.I. opportunities in content writing. So right now, there is still a big need for content designers in this space to develop tone of voice, correctly manage the tool and review content and tweak as needed.

00;11;35;19 - 00;11;45;17
Jake Fernandez

But a lot of the writing could be generated through A.I. tools, and then content designers can focus their time on curating the best output and have a range of options to choose from.

00;11;45;28 - 00;11;49;28
Alex Lowe

So, do you think there are any new methodologies that might be created through A.I.?

00;11;50;12 - 00;12;21;08
Jake Fernandez

Yeah, definitely. I think a few that come to mind are a full analysis of social media sentiment to create user personas. We could automate usability testing as a first pass of a product before going out to real users. We can use tools to analyse, use behaviour such as the ones that you mentioned before, like attention heatmaps or a mode of recognition, the latter of which is extremely helpful for professionals with Alexa theme year, which is a difficulty in understanding or reading emotions.

00;12;21;10 - 00;12;35;26
Alex Lowe

Yeah. I mean, we could even look further to the future and how I could assist with something like VR testing, providing real time feedback to improve user experiences. It might be something CGI could look to implement with our metaverse testing for that.

00;12;36;10 - 00;12;58;20
Jake Fernandez

Yeah, exactly. And it's also very helpful for checking and evaluating work. For example, a few the developers on projects I've been working on recently have been using Ghetto Labs Co-pilot, and that checks over the code and provides fixes. It's been really, really helpful for them to make sure that things are working without needing to sit down with the more senior developer and analyse it themselves.

00;12;58;23 - 00;13;23;28
Jake Fernandez
And there are some ethical implications that we need to consider when using these tools. We may face encoded bias, which is where the dataset of the A.I. has a subconscious bias, and it's extremely difficult to avoid as us humans providing these have our own subconscious biases or what the world may largely accept as right and wrong. So, thinking human rights, racism, LGBTQ+ views.

00;13;24;00 - 00;13;46;12
Jake Fernandez

There are also quite a few legal concerns that have cropped up, but I'll mention those a little bit later on. So, we're also interested in what happens when you combine the capabilities of multiple A.I. systems like Chat GPT and Midjourney. So, Alex, how could they be used to create customised prompts for generating images and assisting in web design projects?

00;13;46;15 - 00;14;14;21
Alex Lowe

Yeah, absolutely. So, for example, we started experimenting with Chat GPT and by training it to create prompts for Midjourney and so Midjourney is a text to image A.I. system. We fed Chat GPT information about Midjourney’s capabilities, so information like details about different lens selections, what Midjourney’s capabilities were and other details into Chat GPT, which in turn generated detailed prompts.

00;14;14;24 - 00;14;34;19
Alex Lowe

As a result, we could enter commands like something like create me a prompt for an image of a young family playing in the park, say, and Chat GPT would generate a detailed description, including suggestions on focal lengths, aspect ratios and quality. We then fed that into Midjourney which generated the corresponding images.

00;14;34;21 - 00;14;36;07
Jake Fernandez

But what were those results like?

00;14;36;27 - 00;14;48;23
Alex Lowe

Well, not too bad. However, to get really good results, we still need to refine the prompts and experiment quite a lot, but with a bit of playing around you can definitely get to some really good results.

00;14;48;25 - 00;14;55;14
Jake Fernandez

Yeah. We also looked at extending these sorts of approaches to other applications like web design. So how did that turn out?

00;14;56;03 - 00;15;35;14
Alex Lowe

Well, we experimented with using Chat GPT to generate prompts for web design, sort of similar to what we talked about for images. So, we put in something like creating a basic design for something like a cosmetics company, and after generating a prompt, we fed that into Midjourney to visualise what that design would be. We also used Chat GPT to create basic steps for user research and UX design strategy, so we would feed information from popular sites like Nielsen Norman Group and then ask it to create a UX approach for a hypothetical project using the example of something like the cosmetics company that we just talked about.

00;15;35;16 - 00;16;00;12
Alex Lowe

So perhaps I need a new website. What would Chat GPT suggest as an approach? And although it's generated steps that were not in depth, we were able to produce a fair first stab at a process which might be useful for basic projects, or perhaps if you were new to human centred design and needed sort of some guidance. So, for example, it's suggested we do user interviews.

00;16;00;14 - 00;16;33;02
Alex Lowe

So, we then got it to generate an interview protocol for use research based on some specific criteria. We then used that protocol to actually interview some users, get back the user research responses, fed that back into it, and then suggested next steps. So, we, including building user journeys, personas, that kind of thing. One thing I would say though, in general about a lot of responses is that can be a lot of extraneous sort-of padding and fluff in these responses, especially on the first pass of a prompt.

00;16;33;24 - 00;16;45;09
Jake Fernandez

And it seems like A.I. systems like Chat GPT have the potential to handle some heavy lifting tasks. Have you tried incorporating any specific guidelines into your A.I. generated prompts?

00;16;45;11 - 00;17;28;10
Alex Lowe

Yes. So, we had a little play with it looking at GDS guidelines. So obviously at CGI, we do a lot of government work, so we thought would be interesting to sort of train GPT on the GDS guidelines and see what it came out with. So those A.I. generated prompts, we then put into Midjourney. And although I think the results that came out, what really suitable for GDS projects is definitely fun to see what the interpretation was, and you can definitely see potential there. And I think as these A.I. technologies continue to develop and improve, there's no doubt that the possibilities for combining A.I. systems like Chat GPT and Midjourney would definitely become more exciting.

00;17;28;12 - 00;17;34;06
Jake Fernandez

Yeah, it sounds really exciting and very promising start. So, what's next then?

00;17;34;09 - 00;18;08;28
Alex Lowe

Well, as we said before, we're in the very early stages of this technology and its possibilities and the impacts are yet to be determined, I think. With regards to HCD specifically, I think we've seen a bit of pushback against it. I think there's a fear that it might make a lot of our work obsolete. Perhaps. Will it take our jobs? - No, I don't think so. Well, at least not for a while. It's I think it's a tool for experts to get the busy work done quicker. You know, we still need that human element and an expert’s knowledge to guide the process ultimately.

00;18;09;01 - 00;18;24;05
Jake Fernandez

Yeah. And when it comes to product design, a lot of the nice-to-have features don't end up being built due to time and budget constraints. So having these A.I. systems will ramp up the speed to allow for larger scopes of work to be completed. In that same amount of time.

00;18;24;08 - 00;18;47;21
Alex Lowe

Yeah. And so, looking a bit further in a bit outside of HCD you know, on top of that, there are lots more forms of media that can be produced, for example, stuff like music or voice sensitisation and video work as well. We've already seen stuff like the Pope in a puffer jacket image that came out and went viral a couple of months ago.

00;18;47;27 - 00;19;17;04
Alex Lowe

And I think we're going to see similar things like that happening with video as these eyes become more accurate and they're able to produce more realistic outcomes. I think, you know, we're definitely seeing an increase in the degree of accuracy in these LLMs You know, you can see GPT 3 is much less accurate than the current version GPT 4 and those future iterations is already going to get better and closer and closer to being, you know, 100% accurate.

00;19;17;07 - 00;19;46;19
Jake Fernandez
Yeah, definitely looking at the ethical concerns as well. So, as I alluded to before, legal issues are already being spoken about. So, the government and regulator discussions are upcoming to try and get ahead of the game and ensure that future developments of A.I are looked after. There's also been a huge lawsuit recently with Getty Images attempting to sue Stable Diffusion for $1.8 trillion for the use of over 12 million copyrighted photos for A.I. training.

00;19;46;21 - 00;20;08;20
Alex Lowe

Yeah, it's interesting to think about, you know, who has the rights to images created based on someone else's work. You know, A.I. sort of being trained on different photographers’ imagery. Where do these legalities and copyright law sort of fall, that’s all being worked out at the moment. But you can see some companies are looking at this already. So, we mentioned before Adobe Firefly.

00;20;08;22 - 00;20;26;09
Alex Lowe

And so, Adobe have their own image stock image site as well. And they're introducing a thing where content contributors to that stock site can mark their work to not be used by Adobe's A.I. and traded. So that's interesting as well.

00;20;26;12 - 00;20;50;10
Jake Fernandez

I think ultimately every single industry will be affected to varying degrees from agriculture to law. I don't think there's a single industry that will not be impacted through A.I., but as it has done several times before, when new technologies adopted, the world will change and adapt. Right now it's critical that we don't resist this change. We really need to learn, understand, and utilise it.

00;20;50;13 - 00;20;55;28
Jake Fernandez

And at the end of the day, these tools and processes have the potential to create better work and home lives for us all.

00;20;56;13 - 00;21;13;16
Alex Lowe

Absolutely. So, it's been great to chat about this today, and I'm really looking forward to seeing what happens in the next six months and what it brings to the world of A.I.. And, you know, how as human centred design designers we are going to adapt? So, thanks very much for joining me today.

00;21;13;19 - 00;21;14;05
Jake Fernandez

Thank you.

00;21;14;25 - 00;21;15;29

Alex Lowe
And thank you all for listening.