The Silent Navigator: UI’s Role in the AI Journey

Why does all the attention go to the GPT when the ‘Chat’ part is equally innovative? ChatGPT launched the GPT-3 model, and it took the world by storm. The secret to its success wasn’t just its advanced technology, which was certainly groundbreaking, but rather how it was used: the model’s integration into a chatbot interface. The ‚Chat‘ in ChatGPT revolutionised AI by making it accessible and engaging to a broad audience. This focus on user interface (UI) played a pivotal role in democratising access to advanced natural language processing technology.

HOW UI CHANGED THE GENERATIVE AI SPACE

Consider the early days of computing, dominated by raw binary codes, streams of 1s and 0s that only those with specialised knowledge could decipher and manipulate. This changed with the creation of graphical user interfaces (GUIs), which transformed lines of code into visual elements like icons and windows, making computers accessible to the general public. Just as GUIs opened up computing to non-specialists, modern conversational interfaces like those used in ChatGPT have made sophisticated AI systems approachable for users worldwide. This transformation highlights the important role of UI in turning advanced technologies from expert-only tools into everyday resources. This shift towards intuitive design in computing paves the way for understanding the revolutionary impact of UI in contemporary AI applications.

For ChatGPT, the UI effectively masked the complexities of its underlying neural networks, presenting users with a simple, engaging conversational model. This conversational part of the technology not only made usage easier but also made the learning curve more fun and less steep. Everyone who has learnt to code knows the long and exhaustive courses that start with simply printing “Hello world” and gradually go into programming logic. It takes a while before you can code things that really interest you, whereas, with ChatGPT, you can learn about prompting by simply talking to ChatGPT about things that do truly interest you. Learning about history or talking about news stories can be super helpful in learning how to talk to a Generative AI model. This chat prompting makes obtaining natural language coding skills easy and fun, which means more people use it, which makes the model better with all the gathered data.

THE OPTIMAL BALANCE OF INFORMATION IN UI DESIGN

Understanding human perception is crucial in designing effective UIs. Our sensory systems, like vision, are optimised not for perceiving all aspects of reality but rather for extracting information that is most relevant for survival (Palmer, 1999). For example, human eyes interpret a 3D image from the light waves that enter them, selectively processing only a fraction of the available electromagnetic spectrum. This selective perception ensures that we focus on what is most useful, avoiding sensory overload, but also means that we can often be fooled by something perceiving itself differently than it perhaps is. This principle has direct implications for UI design in technology. Just as natural selection has shaped our senses to balance between detail and utility, effective UIs must strike a similar balance. They should provide enough information to be useful without overwhelming the user or oversimplifying the interface.

An example of such misperception is the Australian jewel beetle, which mistook shiny, dimpled beer bottles for potential mates. The Australian jewel beetle had evolved in such a way that it adapted its perception of the world around it to only show relevant things. But eventually, this error in perception, highlighted by Hoffman (2009), made the beetle almost go extinct because of its oversimplification. This shows the consequences of an interface that overly simplifies its inputs, leading to errors. Going back to ChatGPT, you can again see how smart they thought about this oversimplification problem. By showing the bot talk word by word, instead of waiting until the bot has finished in a ‘Playground’ design, it shows enough of the magic to mimic thinking. Or the option to show the code it writes to see what the bot is doing. This feature is very impressive for people with coding knowledge but can make things complicated for people with lesser knowledge. For them, there is an option to hide the code with a progress bar. But what happens when there isn’t much of a process to show?

THE BROADER IMPACT OF UI IN AI

Masking the basic underlying technology with a fancy UI is something that happens a lot, especially in the early stages of this but also in other technologies. There are multiple AI companies that wrap AI model API calls (a way for software applications to communicate with each other. It involves sending a request to a remote server and receiving a response) within a newer, more tailored, or modern UI. Most companies, regardless of size, lack the resources to develop their own Large Language Model (LLM, models like ChatGPT or Google Gemini) and typically cannot surpass the capabilities of those developed by leading tech giants such as OpenAI, Google, or Meta. Currently, the major players in high-level text models are OpenAI’s GPT-4, Google’s Gemini, Meta’s LLaMA 3, Mistral’s Mistral 7B and Anthropic’s Claude 3. Each model serves as a base over which almost all solutions are built. Compared to all the companies that offer AI services, this number is very small. But there are companies that innovate the way they use these API calls and build really exciting AI products

EVALUATING COMPANY PRACTICES IN AI

When evaluating these AI products, a key consideration is the transparency of companies about their use of base AI models. Some companies might present UI-enhanced versions of freely available models like ChatGPT-3.5, marketing these as revolutionary solutions and charging a premium. Consumers should be wary of such practices where the whole technology is the UI wrapper rather than the way they use Generative AI. Understanding whether a product offers genuine innovation or simply repackages existing technology helps companies and consumers make good financial decisions instead of being sold as a dream. The main indicator to check if you are dealing with real AI innovations is to look for transparency. The base models from OpenAI, Google, etc. are really cheap or sometimes even free, so if you pay a premium because an organisation found a way to even further improve the useability and quality of these models, they won’t be busy trying to hide it.

In conclusion, the success of AI technologies like ChatGPT depends on both the quality of the underlying models and the interfaces through which they are accessed. As a consumer or user of these technologies, always seek to understand not just how an AI product appears, but also how it works under the hood. This dual approach will ensure you choose products that are not only effective but also trustworthy and correctly priced. Always consider how transparent a company is about its use of AI models because this often indicates the level of innovation of this product. With transparent documentation, you can evaluate the model used itself and what the company does with it ensuring that you will get what you paid for.