Simplicity in interfaces has always appealed to me. This is perhaps why I have a deep affinity for working with the command line. However, my recent discovery of a new user interface has piqued my interest: the interface of conversational AI [1] or generative AI models [2]. It's elegantly simple yet remarkably powerful. You only need to type your question or input into a single text box and hit 'Enter.' Just like that, you're in a conversation. Whether you receive a response to your query or embark on a new discussion is up to you.
Using search engines as an example, a single input box for user queries simplifies the user experience compared to a multitude of boxes for advanced filters. However, as users become more familiar with the system, the addition of multiple boxes or options can greatly benefit power users in swiftly obtaining their answers. As we find ourselves in the early stages of the release of conversational AI agents such as Bard [3], ChatGPT [4], Claude [5], and others, it's akin to drawing a parallel with the initial versions of search engines.
However, maintaining a single input box is not as straightforward as it may sound. Behind this seemingly simple interface, complex processes are at work to extract relevant information. Whether it's through querying indexed web pages or utilizing transformer models [6], the underlying mechanisms are complex. Before the era of conversational AI and even search engines, we had the command line interface in terminals. The concept is similar: users input dedicated commands for distinct tasks. Yet, there was a significant limitation—users had to remember numerous commands and, at times, an array of options in order to achieve specific results or perform precise subtasks, resembling the current version of advanced filters or options in a graphical user interface (GUI).
It's intriguing that despite our numerous technological advancements, the command line interface has not faded into obsolescence. Perhaps it's been immortalized in many movies or simply because simplicity remains a prevailing principle. Nevertheless, we cannot equate the conversational AI interface with a terminal or command line, as it diverges from traditional methods of 'commanding' machines to produce specific results. Instead, every text input is regarded as a 'prompt,' ushering in a new era of roles like prompt engineering [7].
From Command Line to Conversational AI: A Multilingual Revolution
What's truly remarkable is the ability of people to engage in conversations with these conversational AI systems in multiple languages. However, it's important to clarify that our understanding of their underlying mechanisms remains incomplete, beyond the knowledge that they are built upon the transformer architecture. These systems are often likened to stochastic parrots [8] or advanced auto-completion systems. We're not entirely certain whether these systems genuinely 'comprehend' our languages. Nevertheless, they have introduced an entirely new computing paradigm: the multilingual paradigm, where we can interact with these systems in a multitude of human languages. The command line lives on, albeit under a new name—'prompt,' but now with a multilingual twist.
References
- What is conversational AI?
- Generative artificial intelligence
- Bard
- ChatGPT
- Anthropic \ Introducing Claude
- Large language model
- Tech’s hottest new job: AI whisperer. No coding required.
- Bender, Emily M., et al. “On the Dangers of Stochastic Parrots: Can Language Models Be Too Big? 🦜” Proceedings of the 2021 ACM Conference on Fairness, Accountability, and Transparency, Association for Computing Machinery, 2021, pp. 610–23. (PDF)