“How should we make generative language artificial intelligence a part of our business strategy?”

‍

If someone hasn’t already typed the question into ChatGPT, it’s almost bound to happen before long. Unfortunately, this is one area where the answers you get back probably won’t help. That’s because those who first ask the question are often not the ones working directly with the technology, such as a CEO or board of directors.

Stakeholders want specific, contextual, relevant solutions that can be implemented across large organizations and implemented with very little adoption risk. This is the kind of problem, in other words, that calls for the domain expertise and comprehensive experience of human beings who understand the nuances of their specific organization and the needs of its end customers. It also demands an ability to transcend from the urge to quickly jump on the moving bandwagon.

‍

The race to catch up with generative AI opportunities

Projections that generative AI solutions based upon generative pre-trained transformer (GPT) language generation will become a more than $422 billion market by 2028 may create an understandable urgency to weave the technology into existing product roadmaps. Challenging economic conditions that may continue through 2023 and beyond could create even greater pressure to join the current gold rush.

Another big catalyst for exploration, of course, is the sheer range of activities where generative AI tools could be applied within companies. Already, well-known technology companies have launched ChatGPT integrations for everything from contact center suites to HR recruiting platforms and everyday productivity applications.

One of the big risks – beyond delivering a product that fails to meet expectations – is becoming seen as a copycat solution. Much like companies were once accused of simply rebranding solutions to align with buzzwords like “big data,” generative AI could suffer from short-term thinking and execution. Instead of asking what ChatGPT could do for your business, a better path might be to ask what your business could do to create greater value for customers leveraging generative AI technologies. Companies need to open up more transformative possibilities based on their unique use of data, training models and the user experience they can deliver.

The last area might be the most critical – if often overlooked – pillar of the strategy businesses are trying to develop. User-centered design approaches and innovative UX design will be on the forefront of making sure that generative AI systems are adopted, work well, deliver on their promises and futureproof the capital investments that are needed to build a successful tech stack.

‍

The UX design imperative for generative AI

At first glance, UX didn’t seem to be much of a consideration for ChatGPT as developed by Open AI. However, you could almost argue that the stripped-down UI (user interface) in fact  – that is almost akin to the simplicity of the “I’m feeling lucky” button that once graced Google’s search engine is a central part of the genius of its user experience. Generative AI systems have been around for years, but the ease of access of ChatGPT’s call-and-response interface is a big part of its runaway success.

This bare-bones approach has worked well in galvanizing consumer interest in experimenting with the generative AI to produce everything from college essays to Shakespearian sonnets. For other companies that want to capitalize on the technology, however, UX design will help ensure the solutions they develop are highly differentiated and contribute to their overall value proposition to customers and partners.

At Loft, we’ve spent the past four years working on complex natural language generation (NLG) software-as-a-service (SaaS) products. That means we’ve seen first-hand how these technologies need to be tailored to deliver the right experience for those who might harness them for drug discovery, financial service processes or public safety.

Based on our experiences, we recommend taking the following steps, when considering incorporating generative AI in your product roadmap.

‍

1. Develop a data set and design process with customers at the center

As innovative as generative AI appears, it’s important to remember what’s not changing, such as best practices that have guided solution development for decades. The most successful efforts will likely start with a proof of concept (POC) first, for example, followed by iterative, incremental improvements over time.

Crowdsourced testing can help to gauge initial product-market fit, but working with more specific users in the right environment (and with the relevant industry-specific ontologies) will be essential to this work. The initial data will not be perfect. Human oversight will likely be necessary to recognize where the inputs and outputs need to change.

This may require going beyond the kind of user testing that is a common practice in UX design. It will take more of a startup mindset, where A/B testing, pivoting and adjusting will be the norm.

‍

As innovative as generative AI appears, it’s important to remember what’s not changing, such as best practices that have guided solution development for decades.

‍

2. Identify the learning model that will teach generative AI to deliver business value

Using generative AI is a two-way street: you need people to ask smart questions or prompts, and you also need solutions that will be able to understand the correct meaning of those questions and prompts. Training and experience will help address the former, but the latter requires making the right choices in terms of learning model.

Reinforcement learning (RL) – a form of artificial intelligence in which systems are designed to reward desired behaviors – is already being used to detect insurance fraud and assist in retail merchandising, for example. Supervised learning, meanwhile, has been applied in areas such as stock trading and price prediction. RL with Human Feedback (RLHF), can greatly improve the output of NLG systems, but it requires an efficient process to make the system viable for your application.

‍

3. Determine your organizational risk appetite in terms of quality, accuracy

The New York Times story about a ChatGPT conversation gone awry could become just one in a series of generative AI cautionary tales. A recent study showed searches for “Is ChatGPT safe” have risen 614%.  

The reality is that language models can offer surprisingly intelligent results with little effort, but also embarrassingly catastrophic failures. Businesses will need to define what level of quality they accept and where they draw the line in order to manage the expectations of their employees, customers and other ecosystem stakeholders.

‍

4. Calibrate the application for a variety of users and levels of expertise

Don't design an expert system where the UX design is too narrowly focused, such as data scientists who might have a PhD. AI systems can have a massive impact on elevating the skill level of your employees.

ChatGPT has gained traction in part because it is highly intuitive. Businesses need to work with experts (like Loft) who can strike the right balance in terms of designing for accessibility as well as specialization and domain complexity. This takes time, effort and iterative development to get right.

As generative AI becomes core to a wider range of applications and platforms, companies will also have to navigate the best way to frame reinforcement learning models to align with regulatory requirements and protect their intellectual property and data, so that it is not used inappropriately (to train third party models, for example).

‍

‍

5. Be prepared to evolve the metrics used to evaluate success with generative AI

Return on investment (ROI) calculations for incorporating generative AI into products and solutions will naturally vary depending on an organization’s business objectives. They will also likely change over time.

Initially, for instance, generative AI could lead to substantial increases in productivity in areas that require significant manual effort today. Business leaders need to couple their use of these technologies with a human-centered approach that also monitors softer metrics such as job satisfaction, and whether they contribute to a more engaged workforce.

The UX design and UI of more external-facing applications might boost a company’s Net Promoter Score (NPS), reduce churn and even lower customer acquisition costs. Again, testing and iterating will improve the ability to move the needle on the most desired or relevant metrics.

Conclusion: The role of empathy in generative AI UX design

People can be resistant to new technologies even during the best of times, but generative AI solutions are coming into prominence at a time of mass layoffs and corporate cutbacks. This makes it even more important that product solutions are designed with a human-centered approach that empathetically recognizes potential barriers to adoption.

The smartest organizations will pair their learning models and data with a design process that captures user sentiment at the outset, including their biggest fears, misperceptions and areas of concern. The UX solutions should not only minimize any learning curve but align with workflows based on higher-value work that generative AI should free employees to do.

Instead of worrying about whether tools like ChatGPT will replace their job, designing the right solution will require showing employees how these technologies will help them do their jobs better, or address more expansive organizational needs.

The full scope of business opportunities opened by generative AI and NLG may still be unknown, but the path to best-in-class UX design to support them is not.

Connect with us to learn about Loft’s experience in this area. We can tailor a product development roadmap based on your business goals! Our years of experience can help you reimagine how generative AI can accelerate your growth.