AI doesn’t really ‘learn’ – and knowing why will help you use it more responsibly

 What if we told you that artificial intelligence (AI) systems such as ChatGPT don’t actually learn? Many people we talk to are genuinely surprised to hear this.

Even AI systems themselves will often tell you confidently that they are learning systems. Many reports and even academic papers say the same. But this is due to a misconception – or rather a loose understanding of what we mean by “learning” in AI.

Yet, understanding more precisely how and when AI systems learn (and when they don’t) will make you a more productive and more responsible user of AI.

AI does not learn – at least not like humans do

Many misconceptions around AI stem from using words that have a certain meaning when applied to humans, such as learning. We know how humans learn, because we do it all the time. We have experiences; we do something that fails; we encounter something new; we read something surprising; and thus we remember, we update or change the way we do things.

This is not how AI systems learn. There are two main differences.

Firstly, AI systems do not learn from any specific experiences, which would allow them to understand things the way we humans do. Rather they “learn” by encoding patterns from vast amounts data – using mathematics alone. This happens during the training process, when they are built.

Understand how AI is changing society

Take large language models, such as GPT-4, the technology that powers ChatGPT. In a nutshell, it learns by encoding mathematical relationships between words (actually, tokens), with the aim to make predictions about what text goes with what other text. These relationships are extracted from vast amounts of data and encoded during a computationally intensive training phase.

This form of “learning” is obviously very different to how humans learn.

It has certain downsides in that AI often struggles with simple commonsense knowledge about the world that humans naturally learn by just living in the world.

But AI training is also incredibly powerful, because large language models have “seen” text at a scale far beyond what any human can comprehend. That’s why these systems are so useful with language-based tasks, such as writing, summarising, coding, or conversing. The fact these systems don’t learn like us, but at a vast scale, makes them all-rounders in the kinds of things they do excel at.

AI systems do not learn from any specific experiences, which would allow them to understand things the way we humans do. Rido/Shutterstock

Once trained, the learning stops

Most AI systems that most people use, such as ChatGPT, also do not learn once they are built. You could say AI systems don’t learn at all – training is just how they’re built, it’s not how they work. The “P” in GPT literally stands for “pre-trained”.

In technical terms, AI systems such as ChatGPT only engage in “training-time learning”, as part of their development, not in “run-time learning”. Systems that learn as they go do exist. But they are typically confined to a single task, for example your Netflix algorithm recommending what to watch. Once it’s done, it’s done, as the saying goes.

Being “pre-trained” means large language models are always stuck in time. Any updates to their training data require highly costly retraining, or at least so-called fine-tuning for smaller adjustments.

That means ChatGPT does not learn from your prompts on an ongoing basis. And out of the box, a large language model does not remember anything. It holds in its memory only whatever occurs in a single chat session. Close the window, or start a new session, and it’s a clean sheet every time.

There are ways around this, such as storing information about the user, but they are achieved at the application level; the AI model itself does not learn and remains unchanged until retrained (more on that in a moment).

Most AI systems that most people use, such as ChatGPT, also do not learn once they are built. Ascannio/Shutterstock

What does this mean for users?

First, be aware of what you get from your AI assistant.

Learning from text data means systems such as ChatGPT are language models, not knowledge models. While it is truly amazing how much knowledge gets encoded via the mathematical training process, these models are not always reliable when asked knowledge questions.

Their real strength is working with language. And don’t be surprised when responses contain outdated information given they are frozen in time, or that ChatGPT does not remember any facts you tell it.

The good news is AI developers have come up with some clever workarounds. For example, some versions of ChatGPT are now connected to the internet. To provide you with more timely information they might perform a web search and insert the result into your prompt before generating the response.

Another workaround is that AI systems can now remember things about you to personalise their responses. But this is done with a trick. It is not that the large language model itself learns or updates itself in real time. The information about you is stored in a separate database and is inserted into the prompt each time in ways that remain invisible.

But it still means that you can’t correct the model when it gets something wrong (or teach it a fact), which it would remember to correct its answers for other users. The model can be personalised to an extent, but it still does not learn on the fly.

Users who understand how exactly AI learns – or doesn’t – will invest more in developing effective prompting strategies, and treat the AI as an assistant – one that always needs checking.

Let the AI assist you. But make sure you do the learning, prompt by prompt.This article addresses a critical misconception about artificial intelligence, particularly systems like ChatGPT, and how they “learn.” To clarify, AI doesn’t “learn” in the same way humans do, and this distinction is key to understanding how AI systems function. Let’s break it down further, emphasizing the nuances and implications for users.

The Human Concept of Learning vs. AI "Learning"

When we say a human is "learning," we’re referring to the process of adapting over time through experiences. Humans learn from their environment, adjusting behaviors, remembering lessons from successes and failures, and acquiring new knowledge continuously. We also update our mental models based on new information and change our approach as we gain more understanding. This dynamic process is rooted in biology and cognition, allowing us to learn in a deeply contextual and personalized way.

AI, particularly systems like ChatGPT, does not operate in this manner. AI systems don’t accumulate knowledge or adjust their behavior based on direct experiences after their creation. Instead, they are "trained" on large datasets using mathematical models to recognize patterns and relationships between data points. During training, AI systems like GPT-4 analyze vast amounts of text data to learn the statistical relationships between words, phrases, and concepts. This training helps AI make predictions about which words or pieces of information are likely to come next in a given context, but it does not involve experiential learning.

The Nature of AI's "Learning" Process

In AI, the term “learning” is more akin to statistical pattern recognition than the experiential learning humans engage in. For example, when GPT-4 is being trained, it ingests a huge corpus of text data and adjusts its internal parameters to predict which word should follow another in a sequence. This involves adjusting its model based on statistical probabilities rather than any understanding or direct experience with the content. The AI is essentially learning mathematical patterns in the data, not gaining knowledge or insights in the way humans do.

AI’s "learning" phase happens during training, which is a process that requires vast computational resources and occurs before the AI is deployed for actual use. Once the system has been trained and is ready for interaction, its learning effectively stops. It doesn’t continue to learn in real-time based on new inputs or interactions unless it undergoes another training phase, which can be both costly and time-consuming.

AI Systems and “Training-Time Learning” vs. “Run-Time Learning”

To clarify further, there are two primary types of learning:

• Training-Time Learning: This refers to the learning that happens during the model's development phase, where the system is exposed to large amounts of data and adjusts its internal parameters to encode patterns. For GPT-4, this is the phase when it is being trained on text data to generate responses.

• Run-Time Learning: This would be the kind of learning where an AI system adapts and improves itself based on individual interactions after it has been deployed. This is the kind of learning humans engage in daily—our brains continuously process information, update beliefs, and modify behaviors based on new experiences.

Most AI systems, like ChatGPT, do not engage in run-time learning. Once they are deployed, they do not modify their behavior or understanding based on interactions with users unless they are retrained by developers with updated data.

The Importance of the Pre-trained Nature of AI Models

As the article explains, AI systems like ChatGPT are described as "pre-trained" models, meaning that their ability to generate responses is based on a fixed set of data and patterns they learned during the training process. Once deployed, the model doesn't evolve or learn more by simply interacting with users. The model remains "frozen in time" until it is explicitly retrained with new data or fine-tuned with additional adjustments.

This is why AI tools like ChatGPT don’t retain memory of past interactions between sessions. Each time you interact with the system, it starts with a clean slate. It doesn't remember who you are, what you asked last time, or anything you may have shared during a previous conversation. Any personalization you might experience is typically due to external systems, such as user databases that store your preferences, which can be incorporated into the interaction.

Implications for Users

For users, understanding that AI systems don’t learn in real-time has significant practical implications:

1. Limitations in Knowledge: AI systems, particularly those like GPT-4, are trained on vast datasets, but that knowledge is static and frozen in time. If you ask an AI for current events or recent developments, the model might not have the most up-to-date information unless connected to live sources or updated periodically. This makes the AI more suitable for general tasks or well-established knowledge, but less reliable for tasks requiring the latest data.

2. The Importance of Prompting: Since AI systems like ChatGPT do not learn interactively, users should approach them as assistants rather than autonomous learners. Effective prompting becomes critical. You need to know how to phrase your questions, direct the model, and work with its limitations. AI can generate useful, high-quality content, but it still requires human oversight to ensure that it’s accurate and appropriate.

3. Personalization Through External Systems: While the AI itself doesn’t adapt to individual users’ needs automatically, developers can create external systems that track user preferences and incorporate those into the interaction. These personalization features, however, rely on databases and algorithms that sit outside the AI’s core learning process.

4. Quality Control: Given that AI doesn’t “learn” in the way humans do, it’s crucial for users to maintain quality control over AI-generated content. Always verify facts, double-check outputs for accuracy, and treat the AI as an assistant—one that doesn’t learn from its mistakes but can be immensely useful when used correctly.

Wrapping Up: What Does It Mean for the Future of AI?

As AI systems become more integrated into daily life, it’s important to understand both their capabilities and limitations. While the idea of AI "learning" in real time might sound exciting, it’s crucial to recognize that current systems like ChatGPT rely on fixed models that are powerful but static.

The future of AI could evolve with models capable of real-time learning or adapting over time. But for now, AI is primarily a tool that excels at certain tasks, like natural language processing, without the human-like ability to learn and adapt from everyday experiences.

For users, this means treating AI as a powerful assistant that can help with language-based tasks, but with the understanding that it doesn't “grow” or "improve" after deployment—unless actively retrained by developers. The responsibility for accuracy, nuance, and adapting to context still lies with the human operator, who must guide and supervise AI interactions with knowledge and care.