Want a free comprehensive digital product evaluation? Find out more

The Music of Chance


Embracing uncertainty when designing generative AI-powered products and services

In the world of technology, we're accustomed to a certain level of predictability. We push a button, we get a result. We type in a search, and an algorithm delivers a list of relevant links. However, the rise of generative AI is bringing a layer of uncertainty – and with it, a new “AI-first” approach to designing highly personalised and conversational products

Image From Deterministic to Probabilistic

Generative AI models like GPT-3 or Gemini and their successors don't deliver precise, pre-programmed responses. They offer probabilistic outcomes. They create different outputs every single time they're queried, based on huge datasets of information and the parameters of the prompt itself.

This shift can feel both unsettling and thrilling. It means we can no longer guarantee specific outcomes when a consumer interacts with an AI-powered product or service. Responses may be brilliant, average, or occasionally nonsensical. Yet, this is where the power of generative AI lies – we tap into a vast wellspring of potential when used by a creative human being.

Rethinking Friction: The Value of Slowing Down

For decades within the world of user experience (UX) design, the mantra has been to "remove friction." We streamline processes, reduce the number of clicks, and provide immediate gratification. However, when working with generative AI, a new paradigm emerges. Adding a degree of thoughtful friction back into the experience becomes essential. It's through this slowing down that we gain the opportunity to deeply understand a user's true intent, guiding the AI toward more meaningful and valuable responses. This might involve the system asking clarifying questions or requesting additional context before offering a comprehensive response.

Managing Hallucinations: RAG and Other Techniques

Generative AI models can sometimes produce entirely false statements – the infamous AI "hallucinations." Techniques like Retrieval Augmented Generation (RAG), where the AI links its output to specific source documents, can help. Human-in-the-loop oversight may also be key in early rollouts to catch nonsensical or dangerous outputs before they reach the consumer. But there are instances where these "hallucinations" are not harmful, and may even be desirable. Knowing how to guide users to get creative responses from these tools is a big part of prompt engineering and AI model training and safety guardrails.

The Uncertain Future is Promising

This transition toward embracing uncertainty may not be easy. Companies may have to rethink how they build and test AI-infused products. The end users of these products might need to adjust their expectations of AI capabilities.

At Potato, we’re still wary of how AI is being used by most companies experimenting with ChatGPT and Gemini and other AI tools. It’s hard to see through the hype and how to build carefully to create safe products powered by AI.

However, we see potential in several areas that we’d love to talk to you about

  • Personalized experiences: How generative AI and your first party customer data can provide a highly personalised experience that’s aimed at increasing customer lifetime value and better brand experiences.
  • Extending the capabilities of your team with creative workflows that are AI-enabled. With our Fictioneers platform, we don’t see AI as a tool for replacing artists or storytellers. Instead we see AI helping to orchestrate highly interactive and choice-based XR experiences that blend the real and virtual worlds
  • Back office tools and orchestration of customer systems within your existing martech stack: Using frameworks like LangChain or Semantic Kernal to connect your content management tools with your 1st party customer databases. AI/ML Operations so that you use the right models for your business needs effectively and safely.
  • Building the right cloud environment to take advantage of AI in your services in a safe and secure way

The process for using generative AI in consumer products is just beginning. The most successful companies will be those willing to embrace uncertainty, experiment thoughtfully, and prioritize responsible implementation alongside the thrilling, creative possibilities.

Get in touch if you want to talk further about how you can augment your products and services with AI. Or if AI is even the right technology to use to help you meet your goals.

Photocredit: Marius Masalar