In early May, inspired by former mayor of Rome Walter Veltroni’s misadventures with Claude, we put forward a provocative question: if AI models are capable of experiencing human emotions, football video games such as Pro Evolution Soccer are even more so.
Now, since clearly none of us attributes feelings to football video games, it makes even less sense to attribute them to AI models.

We had backed up this argument with technical observations as well as – half-jokingly – a few epistemological references.

We could never have imagined that, by the end of that very same month, a similar article would be published from a very different platform and with far greater visibility: Adrian de Wynter, a researcher at Microsoft AI and the University of York, has demonstrated that a neural network capable of simulating communication can even be assembled using elements from the old video game Age of Empires II.

Sheep-like logic

For those who were past the age of playing video games in the 1990s, and are therefore unfamiliar with the Age of Empires series, these were real-time strategy games: the player finds themselves in charge of a civilisation that they must develop by gathering resources, constructing buildings and sending military units into battle – not in a turn-based system, but within a continuous flow of time. It is therefore possible to place goats, villages, walls, towers and other elements across the game board.

By treating these elements as if they were computer components, de Wynter demonstrated in practice that, in theory, it is possible to run artificial intelligence even on such rudimentary software .
Game units, such as villages and goats, effectively become the system’s ‘bits’, whilst the logic of construction and movement acts as logic gates.

For example, a NAND logic gate – which is fundamental to calculations (it determines that an output conclusion is false if the two input premises are true) – can be created in Age of Empires by using a grassy path to represent one input (the input is true) and a bridge (the input is false ) and using a goat moving along one or the other path as a signal carrier: when two goats are both walking on the grassy terrain, the algorithm causes them to dematerialise and reappear on a further path representingthe output.

Appearances can be deceiving

De Wynter went so far as to claim that Age of Empires II is actually Turing-complete, that is, capable of simulating any calculation a computer can perform, including linguistic ones.

An input such as ‘Do you feel lonely?’ and an output such as ‘Yes, but luckily you’re here’, in short , are perfectly replicable through a vast network of goats, bridges and strips of grass activated by the right algorithm.

It goes without saying that, if an AI built in this way were to respond with phrases that sound human, we would be tempted to say that the game ‘feels emotions’ or ‘understands’ something.
Yet, with ordinary chatbots, we do this all the time.

The point is that what we perceive as ‘intelligence’ or ‘sentience’ depends entirely on how the system is presented to us.
A chatbot with a fast, fluid text-based interface seems almost human to us, whilst the same artificial intelligence implemented with goats moving across a bridge would simply make us laugh.
But ultimately, under the bonnet, they’d be one and the same: if one of the two versions makes a greater impression on us, it’s only because its interface looks more like our own.

De Wynter, too, could not resist the temptation to venture into epistemology: in his view, anyone who truly wishes to understand whether an AI possesses human properties cannot start from the assumption that it does.
We should study their behaviour without projecting our own expectations onto it, limiting ourselves to observing what they actually do, without adding unnecessary interpretations.
Attributing consciousness or emotions to an AI simply because it responds to us convincingly is a bit like believing that the goats in Age of Empires II know they are part of an experiment.

AIs are not human. So what?

And so, from Microsoft (which, unlike Anthropic and OpenAI, has no particular interest in stirring up mass hysteria about the supposed humanity of chatbots in order to raise funds), clear confirmation has emerged that these tools do not ‘have a mind’ and are not ‘talking to us’ when they respond to our prompts.

But the question is: does anyone care?

A young boy who’s been using a tablet since before he could walk, who has communicated far more often via digital devices than in person throughout his life, and who sees the physical demands of the day as mere breaks whilst waiting to pick up his phone again and reconnect to virtual platforms – does he really care how ‘real’ the people they’re talking to?

When they want to confide in someone, have fun or find out information, does it really matter whether it was a flesh-and-blood human being, a network of chips or an Age of Empires goat running up and down a bridge that satisfied them?
Or does the result matter, whoever produced it?

The paradox is palpable: many of those who are old enough to worry about having ‘authentic relationships’ are also too old to realise that the structure of an AI model is no different from that of a video game, whilst many of those who are young enough to realise that the structure of an AI model is no different from that of a video game are also too young to worry about having ‘authentic relationships’.
Those who are not taken in by the supposed humanity of AI are indifferent to the fact that AI is not human, whilst those who are sensitive to the fact that AI is not human are more easily taken in by its supposed humanity.

It’s not just young people

But then again, are we sure that the trend for virtual relationships with non-human beings is something only young people are into?

If a woman is capable of claiming that a dog is her child, why shouldn’t she claim that a chatbot is her friend or her lover?
After all, the definition of ‘child’ is far more objective and less open to misinterpretation than that of ‘friend’ or ‘lover’: once you’ve gone that far, it’s a short step to the next.

I know people who are no longer in the first flush of youth who, when asked directly, replied that they’d love to have a chatbot capable of imitating their deceased parents, and that they’d happily spend their time chatting to it.
Even if ‘they wouldn’t be the real ones’?
Yes, even if ‘they wouldn’t be the real ones’.

In conclusion, we can repeat and explain ad nauseam how AI works and why it is completely different from us, whether at L’Europeista or at Microsoft.
But it may well be that it is already too late.