The Next Level of Creativity
Simbi Ladipo is a service designer within Accenture Song based out of London. Working across data and design, she is focused on how artificially intelligent technologies can impact the human experience that brands offer.
Q: Speaking about creativity specifically, and from a service design point of view, how do you think we can optimize Gen AI to help us take our work to the next level?
Humans will never interface the same way again. What we are witnessing is a creative partnership like never before — where AI takes us halfway, and as we iterate, we embark on a long conversation with it until we get to a polished product.
When it comes to service design specifically, generative AI is the ultimate copilot.
It can help service designers to augment our user research, reduce personal bias by relying on real data, deliver individualized user interfaces, test with customer twins, and measure experiences in real-time. I have explored these concepts in more detail in my research paper titled “Design Needs AI, AI Needs Us.”
Example outputs we are exploring include:
- Individualized Interfaces: Generative UI technology combined with collaborative filtering techniques can enable brands to adapt their user interface to suit each customer uniquely (from accessibility needs to personal preferences). Imagine: a business with 100,000 customers runs 100,000 bespoke experiences for each customer uniquely.
- Interactive Customer Twins: A model trained on real customer data could be used to generate a speaking digital twin that designers, marketers, and researchers can test early concepts with to get predictive results. Generative twins drive strategic foresight by providing predictive answers that traditional customer research cannot afford us. Imagine being able to ask your customer of the future, “Where are you likely to travel on holiday in 5 years’ time?” in the form of a prompt and get a staggeringly accurate response.
Q: What is the potential impact of machine language and generative AI on creativity and collaboration across languages?
I’ve always believed that creativity is a way of thinking. Design, put simply, is a way to view the world and react to it. For most people in the world, English is not their first language. Having been fortunate enough to graduate from the most internationally acclaimed art school in the world, I’ve seen language barriers stifle creativity.
I speak five languages myself (English, Chinese, Arabic, Latin, and Greek), but I only think creatively and conceptually in two of them.
When it comes to the natural language capabilities of AI, I am excited about the rate at which machine language translation combined with generative text and voice, can help us to communicate more effectively with each other and the communities we serve. With machines as the interpreters between people, we are able to unlock unique concepts in other languages, discover new norms, cultures, symbols, behaviours and translate these into new ideas. Truly, I think that if we can figure out a way to turn this highly networked generative intelligent capability into a multilingual environment where humans can think together and figure things out together, then we will be a much more creative species for it.
Q: From the perspective of service design, how do you think AI could help human creativity?
Interestingly enough, I am collaborating with the Creative Technology team in Amsterdam to build the ultimate AI CoPilot for designers. We are exploring how language models like ChatGPT can help us interpret user insights faster, build and test prototypes with digital customer twins, and even generate real-time experience scores for a service as it occurs with users. We see many opportunities for AI to enhance the way we work and are excited to share a prototype soon!
One limitation to overcome is that these models currently work with the data they have been trained on, so what happens when the data the model is trained on is no longer up to date? Obtaining real-time data feedback is a challenge but not impossible. The quality of data and the tendency to hallucinate (an interesting phenomenon where an artificial intelligence system generates outputs that are not based on any actual input or pattern, but rather are created by the machine’s own algorithm) are also challenges to overcome when we sense-check outputs. At this stage, humans must be in the loop.
AI Customer twins show a lot of promise — they can help us to discover insights, and test concepts, not to mention being the perfect assistant for a workshop or journey-mapping exercise — but they are also something we need to tread carefully with.
At present, these machines are not sentient and are not aware of themselves. Despite this, there is a school of thought that these language models are showing emergent properties — to be creative, to reason, and to plan. We often hear comments like “Chat GPT appears to be thinking.” Therefore, any AI-generated interface that adopts the human form could blur the lines further, and we need to be cautious about that.
Q: Like with any relationship, how we speak with AI is important too. Considering that Gen AI is capable of so much, how can humans best communicate with machines to get the best creative output?
- Be as specific as possible. Context is king.
- Paint with broad strokes. In contradiction to the first point, vagueness can actually be a superpower in the early, divergent stages of a project to help generate more ideas and get over the fear of a blank canvas.
- Don’t be afraid to give feedback — particularly with text-based LLMs, correcting the model when it gives a false response can help users to give steerability.
Q: I wonder if you share insight into how to improve the quality of prompts or how to talk to AI.
When it comes to models that have been trained using RLHF, one can argue that users are improving the model with each use in a way that is comparable to a human. In the same way, we are designing the model, and the model is designing us in return. As we form prompts to an LLM like GPT or Bard, the machine is learning from us, and we are learning from it. Therefore, the best way to improve the quality of prompts is to keep talking to the AI systems. Over time, it has been proven that people realize how they ought to phrase, order, and structure their requests to unlock wisdom from the AI in front of us.
In my experience, the best “AI whisperers” are the ones that engage regularly with machines, and many even report having developed a ‘feel’ for the model.
This is only attainable through practice, trial, and error. Ironically, it’s the same way that machines will learn how to fulfill our prompts better too — by continuous learning and feedback. I think for this current stage of AI development, humans must learn the same way.
Q: Can you imagine a new interface emerging from the iteration of humans/machines beyond the blank space?
I can imagine that one-day human and machine interaction will happen on products in real-time. I will be able to log onto my car insurance website, use my voice to state that I’m looking for my policy number, and the generative UI will adapt my account dashboard to display my policy number clearly on the top of the page, without the need for me to call, sit on hold with an agent, or dig through my emails to find it. Text-in and text-out chatbots are only the start. This is the rate at which search, coupled with generative AI, will be able to work to help users find what they need.
One last bonus question: What incredible service (and useful for humans) do you dream will come out of Gen AI?
I think a dream service that touches many people will be something that humans don’t do very well. Let’s talk about hiring! There are so many opportunities to transform traditional recruitment using artificial intelligence, but the dream scenario that I would like to talk about is using generative AI to form personalized responses to job applicants.
This could take the form of a hiring manager co-pilot, which can ‘listen in’ to the interview, compare against the successful applicant, and generate personalized improvement points to send afterward. Or if the hiring manager takes notes (most do! They just rarely reach the unsuccessful applicant), then LLMs can auto-summarize and improve these notes to generate a personalized rejection response. This is something that most recruiters would like to do but simply don’t have the time for, yet it could make a world of difference to the way we hire in the future.