Table of Links
Abstract and 1 Introduction
2. Prior conceptualisations of intelligent assistance for programmers
3. A brief overview of large language models for code generation
4. Commercial programming tools that use large language models
5. Reliability, safety, and security implications of code-generating AI models
6. Usability and design studies of AI-assisted programming
7. Experience reports and 7.1. Writing effective prompts is hard
7.2. The activity of programming shifts towards checking and unfamiliar debugging
7.3. These tools are useful for boilerplate and code reuse
8. The inadequacy of existing metaphors for AI-assisted programming
8.1. AI assistance as search
8.2. AI assistance as compilation
8.3. AI assistance as pair programming
8.4. A distinct way of programming
9. Issues with application to end-user programming
9.1. Issue 1: Intent specification, problem decomposition and computational thinking
9.2. Issue 2: Code correctness, quality and (over)confidence
9.3. Issue 3: Code comprehension and maintenance
9.4. Issue 4: Consequences of automation in end-user programming
9.5. Issue 5: No code, and the dilemma of the direct answer
10. Conclusion
A. Experience report sources
References
9. Issues with application to end-user programming
The benefits and challenges of programming with LLMs discussed so far concern the professional programmer, or a novice programmer in training. They have formal training in programming and, often, some understanding of the imperfect nature of AI-generated code. But the majority of people who program do not fall into this category. Instead, they are ordinary end users of computers who program to an end. Such end-user programmers often lack knowledge of programming, or the workings of AI. They also lack the inclination to acquire those skills.
It is reasonable to say that such end-user programmers (e.g., accountants, journalists, scientists, business owners) stand to benefit the most from AI assistance, such as LLMs. In one ideal world, an end-user wanting to accomplish a task could do so by simply specifying their intent in familiar natural language without prior knowledge of the underlying programming model, or its syntax and semantics. The code will get generated and even automatically run to produce the desired output.
However, as we have seen so far, the world is not ideal and even trained programmers face various challenges when programming with AI. These challenges are only exacerbated for end-user programmers, as a study by Srinivasa Ragavan et al. (2022) observes.
Participants in the study were data analysts (n=20) conducting exploratory data analysis in GridBook, a natural-language augmented spreadsheet system. In GridBook (Figure 6, adopted from Srinivasa Ragavan et al. (2022)) users can write spreadsheet formulas using the natural language (Figure 6: a-f); a formal formula is then synthesized from the natural language utterance. GridBook also infers the context of an utterance; for example, in Figure 6, the query in label 4 is a follow-up from label 3. Both the natural language utterance and the synthesized formula are persisted for users to edit and manipulate.
