No, AI is not sentient – and that's a good thing

AI: A new life form or just a fluke?

Artificial Intelligence – the one topic everyone seems to talk about these days. People have turned to this seemingly novel technology as a beacon of hope, believing it will someday solve all of humanity's problems. At the same time, businesses large and small claim – sometimes disingenuously – that their products use AI, have been improved by AI, or simply are AI. The media further amplifies the trend through misinformed press articles and commentary, from established outlets and influential online personalities alike. Stock valuations are rumored to be inflated by AI hype, and talks of a bubble bursting are ever-increasing. Others view AI as a threat to our very existence – at best widespread unemployment, at worst human extinction. And all this from a 2022 chatbot.

But how did we get here?

"My belief is that biological, intelligent life is essentially a caterpillar making a cocoon. And that cocoon is going to give birth to artificial life, digital life. It's going to give birth to a new life form." – Joe Rogan

A short history of AI

Ignoring fictional notions like the Greek mythological automaton Talos, the history of AI arguably begins in the 20th century. The first official use of the term is widely attributed to the 1956 Dartmouth Summer Research Project on Artificial Intelligence in New Hampshire. Before that, two key milestones laid the groundwork: the McCulloch–Pitts Neuron (1943), describing an artificial neuron that is either activated or not – the foundation of modern neural networks – and Alan Turing's famous Turing Test (1950), asking whether a machine could converse with a human so convincingly that the human couldn't tell the difference.

AI enjoyed considerable attention at the time, with institutes founded and funding secured. But many promises couldn't be kept. Public optimism collapsed in the 1970s, leading to what is now known as the first AI winter – a period marked not just by skepticism, but by a sharp drop in funding from bodies like DARPA.

It wasn't the last. The 1980s revived the craze under a new label: expert systems. These emulated human decision-making through algorithmic if-then rules, for instance decision trees. With advances in computing hardware, expert systems attracted not just researchers, but industry as well. Yet their viability in real applications remained limited, ushering in the second AI winter in the 1990s. Interestingly, the term "AI winter" itself was coined just before this downturn – by AI researchers Marvin Minsky and Roger Schank, warning businesses about exactly what was about to happen.

Researchers kept working nonetheless. Through the 2000s and 2010s, significant progress was made – quietly. Increased computing power and vast amounts of data made concepts like the artificial neuron viable again. IBM's chess computer Deep Blue beat world champion Garry Kasparov in 1997. Convolutional Neural Networks, first introduced by Yann LeCun in 1995, eventually outperformed all traditional approaches to image recognition. Natural Language Processing and Computer Vision research thrived, giving us facial recognition, object detection for autonomous vehicles, and much more.

MNIST dataset showing handwritten digits 0-9 used for training image recognition models — **Figure 1:** An excerpt of the MNIST dataset — used for optical character recognition and image classification.

These advances, however, were neither perceived nor marketed as AI at the time. Then came 2022 and ChatGPT – and everything changed.

What AI actually is

There is no single agreed-upon definition of artificial intelligence. Common descriptions include "a technology that enables computers to simulate human learning, comprehension, problem solving and decision making" (IBM), or "computer systems capable of performing tasks that typically require human intelligence" (ISO). The term is deliberately broad.

The main subdiscipline driving today's AI is Machine Learning – the process of training a model to make predictions or generate content from data. As MIT professor Sara Brown puts it: machine learning gives computers the ability to learn without being explicitly programmed.

Within Machine Learning, Deep Learning uses multi-layered neural networks that train autonomously on large datasets. Each layer processes inputs in a predetermined way – for example, a convolutional layer in an image recognition model reduces an image to its most important features: edges, shapes, patterns. The model initially guesses, measures its own accuracy, and adjusts its internal parameters accordingly.

Deep Neural Network diagram showing input layer, hidden layers, and output layer with connected nodes — **Figure 2:** An example of a minimalistic Deep Neural Network with hidden layers.

Over many iterations, it learns. Once trained, a model can transfer its skills to similar tasks without needing much additional data – hence the name Generative Pre-trained Transformer, or GPT.

This is impressive. But it is not intelligence.

For the most sophisticated models – Transformers, which power ChatGPT, Claude, Gemini and others – the core task is simply this: predict the next most likely word in a sentence.

Three types of AI predictions: classification of cat, dog and human; object detection with bounding boxes; regression predicting age 35 — **Figure 3:** An example of different predictions: (a) classification, (b) detection, and (c) regression.

That's it. The goal is to sound as human as possible.

Why people believe that AI is intelligent

Deep Learning models are predictors. Philosopher John Searle illustrated why that doesn't equal understanding with his famous Chinese Room Experiment (1980): imagine a machine that takes Chinese characters as input, follows step-by-step rules, and produces correct Chinese output – without understanding a single character. Searle argued that computation alone does not constitute genuine intelligence or consciousness.

So why do so many people think AI is intelligent? I believe the answer is threefold.

First, a chatbot just feels more human than a chess engine or an image classifier. Language is how we connect with other people. When an AI speaks to you in natural, fluent conversation – increasingly with a convincing voice – it triggers the same instincts we use to relate to other humans.

Second, AI is complex, math-heavy, and opaque by nature. Models are black boxes even to many researchers. And the word "intelligence" is literally in the name. It's only logical that people unfamiliar with the field assume the output must be intelligent.

Third – and this is what motivated this article most – the term AI is simply a great sales pitch. It sounds impressive and means whatever you want it to mean. Tech companies have leaned into this deliberately, because it makes their products more appealing. The result: people use Large Language Models for political advice, medical decisions, or emotional support – as if they were objective authorities. They are not. And researchers who know better are often financially disincentivized to say so publicly.

Where do we go from here?

AI is not a fluke. It has survived two winters and will be around for decades to come. The advances of recent years are real and remarkable.

But as AI becomes part of everyday life, we urgently need better public understanding of what it actually is – and what it is not. Large Language Models are fallible, not objective. They are not conscious, not all-knowing. What they produce is a prediction, not the truth.

That's not discouraging – it's reassuring. It means we as humans remain essential. The dystopian visions of AI critics are not imminent. What we should do is treat these models for what they are: powerful tools, excellent as translators, writing assistants, or coding aids.

But ultimately, for now, they are just that – a sophisticated simulation of intelligence. Nothing more. Nothing less.

Original Paper

This article is based on an academic paper by the author.

About the author

Dr.-Ing. Jérôme Rutinowski is a researcher at TU Dortmund University, where he serves as Head of Research & Operations at the Chair of Material Handling and Warehousing. He holds a PhD in Mechanical Engineering and is a member of the Lamarr Institute for Machine Learning and Artificial Intelligence — one of Germany's leading AI research centers. His work focuses on Deep Learning, Computer Vision, and the trustworthiness of AI systems.