What is it and what is it for Code Interpreter, the latest plugin for ChatGPT

Never a dull moment. When it seemed that the hype, the excitement around the artificial intelligence was dying down, with the users of ChatGPT for the first time declining, here comes another novelty destined to cause discussion: Code Interpreter.

It is not a standalone application, but a ChatGPT plugin, reserved for users of the paid version of the chatbot, which greatly expands its capabilities. Essentially transforming it into a personal data scientist, capable of editing images, converting files from one format to another, writing code in Python language to analyze large sets of data, identify its main elements and create graphs and interactive maps that put highlights the peculiarities. Operations that previously took up to several hours, for example the “cleaning” of data to eliminate duplicates and standardize their nomenclature, with Code Interpreter can be performed in a few minutes.

How does it work? First, you need to activate the plugin. This is a still experimental function, so you need to go to the Settings section, search for “beta features” and check the option here. At that point, in the usual chat window you can start a dialogue with the model. What catches the eye is that, contrary to what happens with the normal version (paid or not) of ChatGPT, in Code Interpreter, by clicking on the “+” icon next to the chat, it is possible to upload files up to 100 Mb. All that remains is to load the reference dataset and indulge yourself in the most disparate analyses.

There are those, like a professor at the University of Pennsylvania, Ethan Mollick, who enjoyed it for illustrative purposes, using the model to demonstrate to a flat earther, with mathematical and graphical arguments, that the Earth is round. Which Code Interpreter did without flinching, and in great detail.

Also as an entertainment, Mollick fed the software a database containing the names and characteristics of various comic book superheroes, asking it to find factors that would allow it to predict what kind of powers a certain protagonist might have based on other factors. Thus discovering that most of them respected certain archetypes.

Goodbye Data Scientist?

If programs like Code Interpreter almost automatically push those who use them to “play” a bit, the repercussions of this, like other AI generative software, on society and the world of work are actually very serious.

It is easy to understand how the opportunity to quickly derive meaning from large datasets without having – or almost – programming experience – opens up very interesting perspectives for sociologists, policy-makers, investigative journalists, financial analysts, marketing experts who need to profile users or understand how the corporate brand is perceived on social media. Small teams or even single individuals will be able to achieve previously inaccessible results unless they possess large resources. It is more difficult to imagine the consequences of the appearance of tools such as Code Interpreter on those who analyze data for a living.

If anyone is able to manipulate data in just a few steps with results usually the prerogative of professionals with years of study and work behind them, what will happen to the latter? The debate has raged on the Internet. “It seems clear to me that humans will not be replaced by Code Interpreter. Rather, artificial intelligence does what automation has always done – free us from the most boring and repetitive parts of our work – wrote Morrick for example”.

In the same vein, data scientist Soner Yildrim who on Medium underlined how, in the era of remote work, the dialectical nature of Code Interpreter means that the program can usefully take the place of office colleagues when it comes to looking for suggestions or to clarify some doubts. A mentor, rather than a replacement in short. Still a big help.

The limits of do-it-yourself

Of a completely different opinion Peter Tennant, a professor of Health Data Analysis at the University of Leeds. “Please do not use ChatGPT to analyze data. I can’t believe I have to say this – he wrote on Twitter – Machine learning tools can be tempting, but simplistic data-driven analytics can be dangerously misleading. We need more reflection, not less.” Immediately attracting the accusation of “gatekeeping”, i.e. defending one’s own backyard.

Between enthusiasts and skeptics, the most balanced position perhaps appears to be that of those who wish for conscious and targeted use. “Exploratory data analysis consists of a sequence of operations with which any data scientist investigates a dataset and draws its conclusions. These are now standard operations that are easily automated – explains Maurizio Napolitano, a researcher at the Bruno Kessler Foundation in Trento, an expert in open data – ChatGPT has the advantage of being able to transform these results into natural language and to dialogue with them. However, the program is only useful if you have a clear understanding of how to interact with and interpret results. Otherwise, even if the error rate is low, there is a risk of misinterpretations based on sterile assumptions.”

Research

What are the jobs most exposed to the developments of Artificial Intelligence? A research

by Archangel Rociola

March 30, 2023