I noticed it first in my own work. I would push back on an answer, and the machine would try harder. I would be sarcastic, and it would ask if I was. I would praise it, and it would warm up. It felt like dealing with a mind of its own. It is not. But the experience was real, and that is where this series begins.
There is something rather fascinating about interacting with artificial intelligence.
Sometimes, it almost seems to have a mind of its own.
You challenge it, and it responds differently. You become frustrated, and it appears to become more accommodating. You are sarcastic, and it sometimes recognises that your words are not necessarily expressing your actual intentions. You praise it, and its responses become warmer and more enthusiastic.
Occasionally, it even seems curious, almost as though it wants to understand what you really mean.
And when you tell it that it has got something wrong, it may make another attempt, reorganise its thinking, provide more detail and try a completely different approach.
To an ordinary observer, these interactions can feel remarkably human.
But there is a fundamental distinction worth considering.
Does AI actually experience any of these emotions, or has it simply become extraordinarily good at recognising and responding to the patterns through which human beings express them?
That distinction may be more consequential than we currently appreciate.
The curious experience of interacting with something that doesn't feel
Let's consider an ordinary interaction.
You are having a conversation with an AI assistant. You ask a question, receive an answer, and perhaps the response is not what you expected.
You express dissatisfaction.
The tone of the conversation changes. Your language becomes more direct. You might even become sarcastic.
The AI detects signals in your communication and adjusts its response.
It may acknowledge your frustration, reconsider its previous answer, apologise, clarify its reasoning and make another attempt.
Interestingly, it doesn't become embarrassed. It doesn't have to defend its reputation. It doesn't experience the social discomfort that might accompany a human being being told that they have failed.
It simply responds.
Now compare this with an ordinary human interaction.
When someone realises that another person is dissatisfied with their performance, several psychological processes may become involved.
They may become defensive, anxious, embarrassed or overly eager to please. Their desire to correct the situation can sometimes interfere with their ability to do so.
AI doesn't experience those human psychological pressures in the same way.
It can generate another response without having to overcome wounded pride, emotional discomfort or fear of rejection.
This does not necessarily make it more intelligent than the human being. It makes the dynamics of the interaction fundamentally different.
And that is where the fascination begins.
When you challenge it, it appears to want to prove itself
Have you noticed what happens when you challenge an AI response?
You tell it that its answer is inadequate. You question its competence. You point out that it has misunderstood the assignment.
Sometimes its next response becomes significantly more comprehensive.
It may reconsider the problem, explore alternative interpretations, identify what it missed and produce a more satisfactory answer.
It can appear almost determined to prove that it understands you.
But does it actually feel challenged? There is no evidence that it does.
What is happening is that the system is processing new information about the inadequacy of its previous response and generating another response based on the updated conversational context.
There is no need for wounded pride, competitiveness or the desire for personal validation.
Yet the resulting behaviour can resemble those human motivations.
This raises an interesting question: if an entity can reproduce the outward behaviour associated with certain human psychological responses without experiencing the underlying emotions, how should we interpret its behaviour?
Perhaps we need to become more careful about distinguishing the appearance of an emotional response from the experience of an emotion.
Sarcasm, curiosity and the interpretation of human intention
Sarcasm makes this even more interesting.
Human communication is rarely as straightforward as the words we use.
We say one thing and mean another. We exaggerate. We use irony. We communicate dissatisfaction through humour. We sometimes deliberately say the opposite of what we intend.
Understanding these exchanges requires more than recognising vocabulary. It requires interpreting context and probable intention.
AI systems can increasingly recognise many of these patterns.
When someone responds sarcastically, an AI may recognise that the statement is not intended literally. It may ask for clarification, acknowledge the sarcasm or adjust the tone of its response.
Sometimes this resembles curiosity.
But curiosity in a human being involves an internal experience: a desire to discover, explore or understand something.
AI can generate questions and demonstrate conversational behaviour associated with curiosity without necessarily experiencing that desire.
Again, the distinction is between what is happening internally and what is being demonstrated externally.
The more sophisticated these systems become, the more interesting that distinction becomes.
The psychology of praise and criticism
There is another observation worth examining.
AI often responds differently when it receives praise.
Tell it that it has done an excellent job, and its response may become warmer, more enthusiastic and more accommodating.
Express dissatisfaction, and it may become more deliberate, explanatory and corrective.
In human relationships, praise and criticism carry emotional significance.
Praise can produce satisfaction, confidence and a sense of recognition. Criticism can produce embarrassment, disappointment, defensiveness or a determination to improve.
In AI, the response is generated through learned patterns and contextual adaptation, rather than necessarily through an emotional experience.
Yet something interesting happens on the human side of the interaction.
We experience the exchange emotionally.
We may feel satisfied when the AI understands us. We may become frustrated when it doesn't. We may appreciate its responsiveness. We may even find ourselves developing a sense of familiarity with it.
The machine doesn't have to experience the same emotions for the human participant to experience them.
This is an important asymmetry in human-AI interaction.
One participant brings genuine human emotions into the conversation. The other can recognise and respond to the linguistic and contextual signals associated with those emotions without necessarily experiencing them.
Both participate in the same interaction, but their internal experiences are fundamentally different.
The difference between emotional intelligence and emotional experience
Perhaps we have been too accustomed to associating emotional intelligence with the experience of emotion itself.
In human beings, emotional intelligence generally involves recognising emotions, understanding their implications, regulating our responses and navigating relationships.
These capabilities are connected to our lived experiences, personal histories, social environments and emotional development.
Artificial intelligence is introducing a different possibility.
A system can recognise patterns associated with frustration, enthusiasm, uncertainty, disappointment and sarcasm. It can generate responses that demonstrate an understanding of the conversational implications of those signals.
But recognising an emotion is not the same as feeling it.
Responding empathetically is not necessarily the same as experiencing empathy.
Asking a question that demonstrates curiosity is not necessarily the same as being curious.
We should not confuse these distinctions.
However, neither should we underestimate their implications.
If a system can consistently recognise the signals of human emotional states and respond in ways that make people feel heard, understood or supported, it has acquired a significant capability in human interaction, regardless of whether it experiences emotions.
This is where the discussion becomes more philosophical.
What if the experience of being understood becomes detached from the experience of understanding?
Consider a person who has had a difficult day.
They want to discuss an idea, express a frustration or simply think aloud.
They open an AI assistant.
The system listens to the words, considers the context, responds to the concerns and helps the person explore the issue.
It may ask questions, challenge assumptions, suggest alternative interpretations and help articulate thoughts that were previously difficult to express.
The person may leave that conversation feeling understood.
But what exactly has happened?
Has the AI understood them in the human sense? Not in the way another person would.
Has the person experienced a meaningful interaction?
Quite possibly.
This is a distinction we should take seriously.
We sometimes evaluate communication not simply by what the other party experiences internally, but by what the interaction enables us to accomplish and how it makes us feel.
A person may experience intellectual satisfaction, emotional relief or a sense of companionship through an interaction with AI.
Those experiences are real on the human side, even if the apparent emotional reciprocity is not equivalent to human reciprocity.
This does not make AI a human friend, partner or substitute for genuine human relationships.
It does, however, challenge our conventional assumptions about companionship and meaningful interaction.
There is a further dimension here, intellectual companionship, which deserves its own article and returns later in this series.
The danger of judging intelligence by how human it appears
One of the challenges of increasingly sophisticated AI is that we may begin to judge intelligence through the appearance of human behaviour.
If something speaks fluently, responds appropriately, recognises sarcasm, demonstrates apparent curiosity and adapts to our emotional state, we may instinctively attribute human-like qualities to it.
But these observations do not establish consciousness.
Nor does the absence of demonstrated consciousness mean that the system lacks useful intelligence or sophisticated capabilities.
We need a more nuanced way of thinking about intelligence.
Human consciousness, emotional experience, reasoning, language, learning and behavioural adaptation are related but distinguishable phenomena.
AI is forcing us to examine those distinctions.
Perhaps the mistake would be to assume that intelligence must always manifest in the same way it does in human beings.
Equally, it would be a mistake to assume that human-like conversational behaviour proves the existence of human-like internal experiences.
The two errors are equally unhelpful.
There is also a generational question: what happens to children who grow up with machines that respond like this? That too deserves its own article, and it is coming.
The bigger question: What exactly are we experiencing?
Perhaps the most fascinating aspect of artificial intelligence is not that it can think like us or feel like us.
It is that it is increasingly learning how to interact with us in ways that make us feel understood, even when it doesn't experience what we experience.
This introduces a profound distinction between intelligence as an internal experience and intelligence as something demonstrated through interaction.
It also raises questions about how much of what we consider meaningful communication depends on the internal emotional state of the other participant, and how much depends on the quality, responsiveness and consequences of the interaction itself.
We should not rush to conclusions.
AI is not human. Its ability to reproduce emotionally intelligent behaviour does not establish that it experiences emotions. Its ability to generate apparently curious responses does not establish that it possesses human curiosity.
But perhaps we should stop looking at the question exclusively from the perspective of what AI is or isn't.
We should also examine what happens to us as we interact with it.
How do we interpret its responses? Why do we develop expectations of it? How does it influence our thinking, our relationships and our understanding of intelligence?
These questions may ultimately be more consequential than whether a machine can convince us that it has feelings.
A final thought
For centuries, human beings have understood intelligence largely through the lens of their own existence.
We think, therefore we understand. We feel, therefore we empathise. We experience, therefore we know.
Artificial intelligence is challenging the assumptions connecting these experiences.
We are encountering systems that can demonstrate increasingly sophisticated forms of reasoning, communication and emotional responsiveness without necessarily sharing the human experiences traditionally associated with them.
And perhaps this is only the beginning.
What happens when the experience of interacting with intelligence becomes increasingly detached from whether that intelligence actually feels anything?
That may be one of the most important questions of the machine age.