Why All Computer Interfaces Still Suck
As I write this, the layout of my QWERTY keyboard is 150 years old. The odd ordering of letters is designed to prevent the mechanical parts of a typewriter from sticking. My MacBook’s graphical interface can trace its origin to 1984. It is intended for point-and-click navigation with a computer mouse. An invention that was debuted 60 years ago.
The reason behind the aging computer interface comes down to a shift in priorities. As the internet reached the mainstream around 1998, the focus shifted from building interfaces to attracting people’s attention. For this, the design principles from the days of magazines and newspapers were used. As bandwidth increased, the psychology of television was implemented to attract attention better.
Today, the most popular form of interface is the 3-second video. This is further proof that the economic incentives are heavily weighted toward attention-seeking. As a result, the graphical interface we use daily, how we interact with data, and the design of our productivity apps have stayed the same for over 20 years.
Rise of Chat
When was the first time you used ChatGPT? You may have seen it as a prototype for a better search. You may have realized it as a tool to save time. It is still far from being a general user interface. However, the chat interface is being adopted as a bolt-on for many popular software services. Something similar to what occurred during the 1980s with the computer mouse.
Before the mid-1980s, most business software used keyboard shortcuts to perform tasks. My dad was in finance and used Lotus 123, a popular spreadsheet. He always had a printed sheet with keyboard commands on his desk. As the mouse gained traction, software vendors began to support it in their existing applications.
During the 1990s, mouse operations became part of every computer operating system, and each platform’s graphical elements were standardized. While improved upon, the core elements of these platforms have remained unchanged. We still use a pointing device and have standard form elements and icons representing our software. Thus the question becomes, will AI-based chat become a main staple of our future computer experiences?
Statements vs. Questions
LLMs and chat interfaces’ most significant limitation is the lack of real-time information flow. For example, ask ChatGPT about the weather, and you will get an error message. Ask Bing or Google about the weather forecast, and you will get a graphical representation of the current conditions. This is still a search but with more options for presenting the question.
However, real problems occur in dealing with numerical datasets. Is chat the best means to query data? For that, we must look into what a question represents. Previously, data was accessed using a computing language based on logic. For example, Standard Query Language (SQL) is the typical means to ask questions about multiple data tables.
Using SQL, one may ask to “SELECT” a collection of data points from the whole. However, it is presented in statement form, not a question. For example, “SELECT COUNT(books) AS num_books FROM publishers WHERE pub_date = 2023.” This is a command. To go from command to question, we can simplify our statement to “How many books were published in 2023?”
Illustrations vs. Correspondence
Educators have long known that complex ideas are best presented in diagrams or illustrations like the model of an atom, the winding vine of DNA, or the periodic table of elements. These representations make storing and organizing large amounts of knowledge into simple structures easy.
For my senior project in Chemistry, I wrote a paper on superconductivity. The ideas of a lattice structure, slowing the speed of atomic collisions, and the notion of absolute zero are challenging to imagine. The literature on the topic revealed mathematical concepts and drawn-out explanations. All of these made it difficult to grasp the phenomenon entirely. Instead, I searched for an illustration.
Discovering a simple drawing of a lattice structure with moving atoms, I could connect the links between scientific explanations and reality. This provided the ability to imagine the occurrences on a molecular level. Now, my mind had a picture of what was occurring.
At Estimand, we have been working to develop the next generation of AI interfaces for working with complex topics such as global financial risk. We have developed an interface based on interactive diagrams instead of written correspondence. The result is an AI that eliminates the need for data mining. Instead, the machine builds the connections using causal relationships and presents them in a diagram that the user manipulates.
The Feynman Technique
We learn from hearing, reading, writing, and doing in school. However, the way most subjects are taught does not lead to mastery. In contrast, Richard Feynman, Nobel laureate and core figure in 20th-century physics, developed a learning technique that creates a deep understanding of any topic.
He believed that any concept is best learned with the following four steps:
1. Select a concept to learn.
2. Teach it to a child.
3. Review and refine your understanding.
4. Organize your notes and revisit them regularly.
These concepts create the foundation for powerful interfaces when applied to human-computer interaction. For example, in step 1, Feynman proposes we write on paper everything we know about a subject. As we learn more, we add to the list. Applied to computer experiences, we can see how each person’s starting point differs.
For some, the best interface is a blank screen with a blinking cursor. For others, it may be a prompt such as “ask me anything.” Yet, for many, it combines graphical elements with different interaction methods. This provides an extra level of sophistication based on current knowledge. For example, the periodic table of elements is an excellent chart for chemistry students. Still, it has to be explained within the first few days of class before it becomes useful.
In step 2, Feynman discusses the importance of teaching the concept to a child. This is powerful because it forces us to break down a complex idea into a simple explanation. For interface design, this is getting close to a diagram or illustration. One can only draw what they fully understand.
In steps 3 and 4, Feynman declares the importance of reviewing and organizing your understanding. This is a technique designed to solidify loose ends. For example, I can explain that friction produces heat, but I need to know more about how atoms collide to do the same. Once I have grasped the concept, I can organize everything into a diagram with simple explanations for each part.
Conclusion
Feynman’s work is the basis for our new interface at Estimand. It must serve the needs of business analysts and data scientists while allowing complex ideas to be expressed in simple interactions.
Founded in April 2023, Estimand Inc. is a Delaware Corporation with principal offices in Raleigh, North Carolina. We began as a paper on causal AI that evolved into the first company to instantly reveal the factors behind global financial risk. You can learn more about Estimand by visiting our website: https://estimand.ai.
References
- Dormehl, L. (2021) The first commercial computer mouse shipped 40 years ago today. Digital Trends. https://www.digitaltrends.com/computing/computer-mouse-40-year-anniversary
- FS Blog (2023) The Feynman Technique: Master the Art of Learning. FS.blog. https://fs.blog/feynman-technique/
- Hanna, K. (2023) QWERTY Keyboard. TechTarget. https://www.techtarget.com/whatis/definition/QWERTY-keyboard
