In the age of LLMs programs don't need to be written, only read
COM3 Level 2
MR21, COM3 02-61
Abstract:
Programming languages have always been designed as a compromise between readability and writability. As LLMs become more and more proficient at writing programs, why not make programs maximally readable? While we could design a language optimized for readability, pseudocode already serves this purpose well.
I've been experimenting with providing a short prompt to chatbots such as ChatGPT or Claude that instructs them to only display pseudocode while generating JavaScript or Python behind the scenes. Users don't need to learn the syntax or technical details of a programming language. Instead they can read the programs they are co-creating with an LLM in their preferred natural language.
Bio:
Ken Kahn has been interested in AI, programming languages, and education for 50 years. His 1977 paper “Three interactions between AI and education” In E. Elcock and D. Michie, editors, Machine Intelligence 8: Machine Representations of Knowledge may be among the first publications on the topic. He received his doctorate from the MIT AI Lab in 1979. He designed and implemented ToonTalk, a programming language for children that looks and feels like a video game. He has developed a large collection of AI programming resources for school students (https://ecraft2learn.github.io/ai/). He recently retired from the University of Oxford as a Senior Researcher. He is writing a book about how school students can use chatbots to create apps, adventures, illustrated stories, and more.