Why data without profound human interpretation, is meaningless, even in the era of AI

October 16, 2023
Data News
Victoriano Izquierdo
Victoriano Izquierdo

Many people are saying that ChatGPT is going to replace all data analysts and data scientists. Well, that might never happen… Data, by nature, is always subjective and meaningless without human interpretation.

I believe there are four significant challenges that AI cannot address (at least for now, maybe never?) without a human providing context. Here are my bets, from easiest to hardest to resolve:

1 - Measurable External Variables - a competitor campaign, the weather, the celebration of a soccer match, a pandemic, normalizing data by a country’s population… All these variables, often not included in your internal datasets, are usually crucial for explaining why something has occurred. Among the challenges I'll mention, this is the "simplest" one, and AI, especially helping find alternative data from companies like Cybersyn, will assist. However, the range of external data variables is so vast that we probably still need humans to decide which data to look for in each context, with expertise in the field.

2 - Difficult/Impossible to Measure Internal Variables. The most important and predictive variables are often extremely expensive, difficult, or directly impossible to measure. Even measuring socio-demographic or firmographic factors of customers tends to be challenging. Asking for age in a form creates friction, and asking for gender raises suspicions about how that data will be used. Psychographics are even more challenging to measure. They might be inferred from a clickstream of behavior events when using a digital product. But often, these variables can only be measured through human-to-human interviews, extracting information through sales artistry and human complicity - commonly called qualitative data. These conversations often inspire new variables that no one had thought of.

3 - Humans Can Identify and Adapt to Emerging Trends That Pre-existing Data Models Do Not Recognize. Learning from past data is possible, but humans continuously adapt their behavior. For instance, you may offer discounts to customers who bought certain products to incentivize them to purchase others. However, your customers will learn, and the coupon might cannibalize future purchases of other products without you realizing it. Read the Book of Why to learn more about this.

4 - Understanding the Data Analysis Process Is Critical for Building the Necessary Trust for Decision-making. Many people distrust their doctors and science, even if these professionals know more about medicine. If they don't explain the diagnosis well, individuals may turn to pseudotherapies that seem logical to them. Any analysis involves numerous micro decisions and assumptions about the data. This includes how the data is sampled and filtered, which variables are ignored, which are considered, and how it is visualized. Small changes in the criteria can lead to entirely opposite conclusions. This uncertainty means that virtually no one, no matter how much they believe in data, can be entirely sure about the conclusions drawn.

Without tools that allow for an intuitive and straightforward understanding of why specific analysis decisions were made, it is highly likely that decision-makers will not gain sufficient confidence and conviction to act boldly and decisively. These interfaces must guide and instill confidence in the analysis process. This is a big part of mission of Graphext. 


The Data

Explore Yourself

Key Variables

Type of Analysis

Relevant Industries

Other stories

Ready To Get Started?

Ready To Get Started?

Let's dive into your data with Graphext. It's super simple, and you'll get your project ready in a few minutes.