AI is, by definition, everything that computing has not yet managed to automate. The term Artificial Intelligence has never had a single meaning. It is a chameleon that mutates with each technological breakthrough, far more than through a linear progression of "intelligence."
A brief journey through time to understand where we stand today:
1950s The cognitive ambition It all begins with Alan Turing and the Dartmouth Conference. AI is then a theoretical quest: can we simulate human reasoning?
1970s-80s The era of expert systems AI becomes a library of business rules. Knowledge is coded (MYCIN, XCON). It is rigid, but revolutionary for medical diagnosis and industrial configuration.
1990s-2000s Statistics and data mining AI becomes discreet but effective: banking scoring, fraud detection, decision support. We no longer simulate the brain; we analyze patterns.
2000s-2010s Applied machine learning Models now learn from data at a larger scale. Recommendation, fraud, optimization: AI becomes industrialized, without being cognitive.
2010s-2020 Deep learning and perception Computer vision, speech recognition, machine translation. Thanks to big data, cloud computing, and the massive use of GPUs for deep learning, AI learns to see and hear (Siri, Google Voice).
2020-2022 The generalist reasoning engine The arrival of Transformers (BERT, GPT-3, ChatGPT). AI becomes conversational, capable of synthesis, translation, and intellectual assistance. This is the era of copilots.
Since 2023 Agentic AI A new paradigm shift. We no longer "chat" with an AI: we deploy systems capable of acting, orchestrating workflows, and executing actions (AutoGPT, DevOps agents, IT agents).
Key point for CIOs In agentic architectures, intelligence is still carried by the model (LLM). The agent itself is primarily a matter of software engineering, control, and governance. The challenge is no longer to "do AI," but to define: 1) What type of AI is actually being leveraged? 2) What level of autonomy are we delegating to our IT systems, especially on critical processes?
What about you? How are you governing these levels of autonomy in your IT environments? Are we ready to massively shift from "Copilot" to autonomous agentic systems? I welcome your opinions and feedback in the comments.