Will ChatGPT-like interfaces ever replace graphical user interfaces?
After attending a great workshop about the ultra famous ChatGPT by Jordi, maker of the viral abbrevia.me, I got obsessed about this question:
Will NLP interfaces (like ChatGPT) ever replace traditional GUIs?
I remember making me a similar question about 7 years ago, when working at Tiendanube and we were giving conversational commerce a try. Back then, there was an explosion of chatbots, and it was the thing to do. Eventually, all the fuzz went away, some chatbots remained, but it was far from being a revolution. The technology was still rudimentary and the user experience was far from enjoyable. Graphical UIs were never threatened.
Now it’s 2023, and ChatGPT disrupted the status quo. Significantly. Talking with this technology feels scary and natural.
Natural interaction
In his book “The Design of Future Things”, Donald Norman discusses various aspects of design and human interaction with technology. While the book primarily focuses on the design of future technologies, Norman does touch upon the concept of natural interaction.
Norman emphasizes the importance of designing technology that aligns with human capabilities and behavior. He argues that technology should be designed to accommodate and support natural interaction, where users can easily understand and operate devices without explicit learning or conscious effort.
Norman criticizes the idea that users should adapt to technology, suggesting instead that technology should adapt to users. He emphasizes the need for intuitive interfaces that make use of natural human skills and abilities. This includes designing devices and systems that leverage human perception, memory, and cognition, allowing users to interact in effortless and familiar ways.
Now think about NLP technologies. No learning curve, no ad hoc language, no new semiotics. It’s the peak of natural interaction.
All that glitters is not gold
Anyhow, while natural language processing has made significant advancements in recent years, there are still several challenges that remain unsolved.
Just to mention a few, here are some of the key areas where NLP faces ongoing challenges:
Understanding context and common sense: NLP models often struggle to grasp the context and common sense knowledge required to accurately interpret language. While models like GPT-3 can generate coherent responses, they can also produce incorrect or nonsensical answers due to limitations in understanding the broader context of a conversation or the world’s knowledge.
Handling ambiguity and ambiguity resolution: Language is inherently ambiguous, and NLP models can struggle with disambiguation. Resolving ambiguous references, word senses, or idiomatic expressions accurately and consistently is a challenging task for NLP systems.
Dealing with rare or out-of-vocabulary words: NLP models typically learn from large datasets, but they may struggle when encountering rare words or domain-specific terminology that is not present in their training data. Handling these out-of-vocabulary words effectively and generalizing to new vocabulary remains a challenge.
Ethical and bias-related issues: NLP models can inadvertently perpetuate biases present in the training data. They may generate responses that reflect gender, racial, or cultural biases, leading to biased or unfair outputs. Mitigating bias and ensuring ethical use of NLP models is an ongoing challenge.
Grounding language in the physical world: Understanding language in relation to the physical world is a challenge for NLP. While progress has been made in areas like image captioning or visual question answering, fully integrating visual or sensor information with language understanding remains a challenge.
Contextual understanding over longer conversations: NLP models often struggle to maintain consistent contextual understanding over extended conversations. Maintaining a coherent dialogue or understanding references made earlier in a conversation poses difficulties for current models. Even though I know people who struggle the same.
Generating explanations and providing reasoning: While NLP models can generate responses, they often lack explicit reasoning or the ability to explain their decisions. Transparently explaining why a particular response was generated or providing detailed reasoning behind answers is still an open challenge.
GUIs and affordances
Ideally, good design should provide clear and intuitive affordances, allowing users to easily understand and anticipate how to use a particular object or system. In the before-mentioned book, Norman also highlights the distinction between perceived and actual affordances. Perceived affordances are the cues or signals that users perceive or interpret from an object, while actual affordances are the real capabilities and functionalities provided by the object. He suggests that designers should strive to align perceived and actual affordances to minimize confusion and enhance usability.
Feedback helps users understand the consequences of their actions and provides information about the state of the system or object. Effective feedback enhances the perceived affordances and guides users in interacting with the design.
Graphical user interfaces, when designed correctly, have the ability to communicate affordances in a way that users can perfectly understand its capabilities and functionalities. Users are able to predict the behavior of the system.
You know exactly how this volume knob works, how to operate it, what are all of its capabilities, and what will happen immediately when you touch it.
In general, NLP interfaces struggle to provide these affordances. Like with human beings, you don’t know what to expect.
Conclusion
Given the current constraints of the technology, it’s unlikely that traditional graphical user interfaces will be completely replaced by chat-based interfaces in all scenarios. Different types of interfaces serve different purposes and have their own advantages. GUIs are effective for visual representations, complex data manipulation, and tasks that require precise input or control. They are well-suited for tasks such as graphic design, video editing, or working with large datasets.
On the other hand, chat-based interfaces excel in tasks that involve natural language interactions, information retrieval, and personalized recommendations. They can be particularly useful in scenarios like customer support, personal assistants, and interactive storytelling.
In many cases, a combination of both types of interfaces might be employed to create a more comprehensive user experience. For instance, a chat-based interface can handle high-level commands and inquiries, while a graphical interface can be used for more granular control and manipulation.
Overall, while chat-based interfaces have the potential to significantly impact how we interact with digital systems, it is unlikely that they will completely replace traditional user interfaces in all domains and use cases. The coexistence and integration of different interface paradigms will likely continue to shape the future of user interaction.
