Using Natural Language Prompts With AI Models for Low-Cost Assistive Software Design: Exploratory Comparative...

Area of Science:

• Assistive Technology Development
• Artificial Intelligence Applications
• Human-Computer Interaction

Background:

• Investigating artificial intelligence (AI) model capabilities for creating low-cost, personalized assistive products.
• Assessing 7 AI models (5 free, 2 paid) for functional software generation.
• Focus on AI's role in empowering nontechnical professionals.

Purpose of the Study:

• Determine the most effective, accessible, and consistent AI models for developing inclusive digital solutions.
• Evaluate AI's ability to generate code from natural language for assistive technology design.
• Compare performance of commercial and free AI models.

Main Methods:

• Prompting AI models with natural language to generate Python code for an adapted controller.
• Requesting 16 progressively complex functions without technical input or feedback.
• Evaluating performance based on implemented functions and prompts needed.

Main Results:

• Paid models significantly outperformed free models.
• Gemini Pro (Google) achieved 14/16 functions (1.25 prompts); ChatGPT Plus (GPT-5) achieved 11/16 functions (1.31 prompts).
• Free models generated 0-4 functions; DeepSeek and Gemini Free led their category.

Conclusions:

• Paid AI models (Gemini Pro, ChatGPT Plus) show strong potential for assistive technology development by nontechnical professionals.
• AI tools can bridge the gap between professionals and software creation for affordable, user-centered solutions.
• Human oversight and prompt design literacy are essential for reliable and ethical AI use.