Much like the early days of aviation, when pioneers dared to defy gravity and soar into the unknown, the dawn of artificial intelligence presents a similar blend of wonder and challenge. Just as meticulous engineering and stringent regulations transformed air travel from a daring feat into an indispensable global connector, AI is on a trajectory to become an integral force in our digital landscape.

The journey of integrating AI, especially with the recent introduction of data analysis through the ChatGPT plugin code interpreter, mirrors the evolution of aviation—filled with potential, yet demanding caution and precision. To help ensure data analysis AI achieves its potential and becomes as trusted as modern air travel, we outline some of our recently discovered strategies that help us fly safely through the early trials of AI data analysis.

Replicate and plan

We have taken full deliverables and recreated them using code interpreter to serve as an AI benchmark. Use your project plan to craft prompts that execute on each step. This method will show you how and specifically where the AI makes decisions, and where the AI may fail. Ask the AI to explain its choices or instruct it very specifically to adhere to your past choices.

This process will educate your team on how the AI likes to work, but is also a wonderful way of learning by doing. While this may feel slower than doing it without AI, it is an investment in a future workflow that can be scaled.

When replicating a regression for a client, our regression analysis was effectively replicated using seven prompts. Here’s a generalized version as an example:

1. Upload data set (with context about what it represents) and request AI to report back on all fields and their formats.
2. Request AI to clean any data, remove outliers, address missing values, and explain the methodology for creating a clean and complete dataset.
3. Apply a linear regression model and check for multicollinearity. Instruct the AI which columns to avoid, and which variables are dependent/independent. After setting up and running the model, provide a detailed summary, highlighting coefficients, significance levels, and other relevant metrics.
4. Compile your findings into a well-organized table. This table will showcase the variables used, their coefficients, and the associated p-values from the regression model. For ease of access and further analysis, make this table available for export as an Excel file.

An AI intern

Think of AI as a diligent intern and use it for lower-risk tasks that you will eventually check and double-check. Assign it tasks with clear instructions, and assume the AI needs as clear instructions as you would give a new intern starting on their first day. Some time-saving tasks that are appropriate for an AI intern include:

• Data categorization using NLP libraries
• Data cleaning and merging
• Text parsing to generate new fields
• Handling missing value

Utilize multiple AI platforms: While the code interpreter plugin is unique, seek methodology and code from various sources, akin to getting a second medical opinion. Use platforms like Claude.AI to cross-check ChatGPT’s suggestions and build confidence in the output.

Enterprise platforms: ChatGPT’s upcoming enterprise platform, a secure SOC2-compliant solution, promises enhanced trust in AI. It offers an admin console, data safety integrations, a transparent usage dashboard, customizable workflow templates, and advanced data analysis capabilities. Keep an eye out for its launch later this year.

Just as jetliners transformed travel, AI holds the potential to revolutionize various sectors. With careful navigation, informed decisions, and a clear vision, data analysis is poised to soar to new heights in the digital realm, promising a future where its trajectory is both purposeful and pioneering.