AI Won't Take Data Scientists' Jobs: Here's Why

The Illusion of the AI Threat

In today's digital world, it's not uncommon to hear voices raised, whether online or in face-to-face discussions, claiming that artificial intelligence (AI) is about to replace data scientists. This fear, while widespread, is often fueled by those not directly involved in the field. It can deter potential talent from pursuing this promising career. However, after five years of experience in this domain, I am convinced that AI will not take the place of data scientists, at least not in the next ten years. This article aims to dispel this unfounded fear.

AI as an Ally, Not a Rival

My approach to AI is not that of a skeptic. On the contrary, I integrate this technology into my daily professional life, constantly seeking to improve my AI skills. This tool proves to be a valuable ally for various tasks such as writing standard code, generating technical ideas, creating documents, and quickly producing data visualizations. AI is here to stay, and it is crucial to embrace it to avoid being left behind. Mastery of these tools will become as essential as using email or Microsoft Word today. AI will not replace data scientists, but those who can use it effectively may surpass those who cannot.

As a data scientist, it is imperative to master tools such as Python, R, and SQL. These skills will become cornerstones of our industry, just as Python has become indispensable in machine learning. This evolution is inevitable, and it is essential to prepare for it now.

Challenges AI Cannot Overcome

For AI to completely replace data scientists one day, it would need to acquire complex skills such as the ability to transform vague business problems into mathematical or algorithmic systems, communicate effectively with non-technical stakeholders, and write production code without errors. Furthermore, it would need to be capable of making trade-offs between complexity, architectural design, and development processes while establishing trust within a team or organization.

If AI were to master all these skills better than a human data scientist, it would mean that most jobs would also be at risk. In such a scenario, we would face far more serious issues, almost at the level of technological singularity, rendering concerns about a career in data science trivial.

The Limits of AI's Mathematical Reasoning

One area where AI still shows notable weaknesses is mathematical reasoning. While AI can perform basic calculations like finding the gradient of a function or calculating the determinant of a matrix, it is incapable of solving unsolved mathematical problems. For instance, the Riemann Hypothesis, a famous unsolved conjecture, remains beyond the reach of AI. This hypothesis suggests a hidden order in the distribution of prime numbers but requires a creativity and conceptual reasoning that AI does not possess.

Persistent Errors of AI

In using these tools daily, it is evident that AI still makes many mistakes. Language models, or LLMs, tend to "hallucinate," producing results that seem plausible but are often incorrect. These errors stem from the probabilistic nature of these models, which can generate sequences of words devoid of meaning to meet user expectations.

While humans also make mistakes, they generally have the ability to recognize and correct them. In contrast, AI often displays excessive confidence in its responses, which can mislead users.

The Limits of AI Model Improvement

Interestingly, AI models have not shown significant improvements over time. This can be attributed to two main reasons: on one hand, the underlying algorithm remains unchanged, as all these models use the Transformer architecture. Thus, each "new" model is not truly innovative. On the other hand, the amount of data available for training is limited, as there is only a finite amount of information in the world.

For example, OpenAI's GPT models have been trained on almost the entirety of the Internet, meaning there is little new data to leverage. Therefore, there is a ceiling on their potential for improvement.

The Absence of Human Relationships in AI

Despite the emotional attachment that some people may develop towards robots, AI is incapable of forming genuine relationships. Humans are social beings, and most business interactions rely on human relationships. People prefer to work with individuals they like, even if they are not the most technically qualified.

A stakeholder will trust a data scientist who has proven their ability to deliver consistent results. Even if an AI proposes a technically "better" solution, the human relationship will likely prevail.

The Real Impact of AI

One day, a former manager of mine asked me what had really changed since the advent of AI. While we now have better tools to solve certain problems and productivity has increased in some aspects of our work, the role of data scientist has not fundamentally changed.

Take a moment to reflect on what has truly changed in your daily life thanks to AI. You may find that you can name very few things, if any at all. AI, in its current form, has existed for over four years, and society as a whole has not been significantly transformed from my perspective.

If, after reading this article, you wish to deepen your knowledge of AI, I recommend checking out my previous article, which offers a comprehensive roadmap for mastering AI.