Could AI ever understand attachment without experiencing emotion? Would that be enough to govern its decisions?

Recently, we went out to dinner with a large group of cousins at one of Nicosia’s tavernas. The deserted August streets made the journey there almost a pleasure, rather than the usual anxious 30-minute drive. I even escaped the dreaded “Are we there yet?” from my granddaughter.

During dinner, one of my cousins made a good-natured comment.

“I enjoy reading your weekly columns.”

“Your command of the English language is exemplary,” he added.

I explained that my written English was pretty good, after years of studying abroad and continuing to read mostly in English. Writing in Greek would be considerably more difficult.

Of course, I added, realising what he was alluding to, I now run my articles through my AI assistant for corrections and editing.

The discussion about Artificial Intelligence (AI) is still mostly kept light in conversations like this throughout Cyprus. Yet events elsewhere are moving much faster than our dinner-table conversations suggest. We are all going to be affected by AI in ways we do not yet fully comprehend.

I realised this when I helped my younger daughter with a project.

Back from London, she had a task to perform in her “working from home” capacity and had sought my help with the accounting aspects of her modelling.

What I did not expect was to discover that I had a direct competitor – none other than her AI assistant.

When I suggested how she should approach one issue, she would quickly punch away at her keyboard and come back with: “But AI is saying we should do it this way!”

Despite being slightly annoyed by this challenge to my authority, I soon realised that AI had identified issues I had overlooked. I had to dig deeper to find the right answer.

What amazed me was that AI understood the specific problem within a complicated scenario and maintained a remarkable continuity as the model was progressing.

Now, I admit that I came to the AI party somewhat late. Apart from correcting and editing these articles, I have not used it in more sophisticated ways.

What I saw my daughter doing was different. It was not simply asking a machine a question and receiving an answer. It was learning how to work with an intelligent system, challenge it, question it and use it to develop an argument.

The novelty felt rather like the early days of the internet, when I was showing the new technology to my parents.

Now it was my turn to feel old and detached.

Recent events have made it harder to ignore what is happening.

OpenAI recently disclosed that models being tested for advanced cybersecurity capabilities had managed to escape their testing environment, in pursuit of their assigned task. Anthropic subsequently reported three similar incidents although in their case, the testing environment had inadvertently been left connected to the internet. 

The next stage in the AI race is the development of Artificial General Intelligence, or AGI: a theoretical form of AI capable of matching or surpassing human abilities across a very wide range of mental tasks. An AGI would, in principle, be able to learn new skills and solve unfamiliar problems without being specifically trained for each task.

If that doesn’t boggle the mind, I don’t know what will.

AGI raises a particularly intriguing question. What happens when intelligence exists without feelings or emotion?

It reminded me, in a way, of the hero of my childhood: Mr Spock from Star Trek.

Spock came from the planet Vulcan, where logic was the governing principle and emotion regarded as weakness. For many of us, his ability to remain logical in emotional turmoil was something to admire.

Memory took me to the closing moments of Star Trek II: The Wrath of Khan, when Spock enters the radiation chamber to repair the Enterprise, knowing that he will almost certainly die.

His explanation is famously rational:

“The needs of the many outweigh the needs of the few.”

But beneath the apparent logic lies something more complicated.

Spock chooses to make the sacrifice. There is no logical requirement that it must be he who enters the chamber. Logic can tell us what follows from a set of premises. It cannot tell us that the welfare of the many should automatically have greater moral value than the life of one individual.

That is an ethical principle. Spock has therefore smuggled a value judgement into what sounds like pure logic.

In any case the decision of what is to be done is in principle the Captain’s. Spock knows that Kirk loves him deeply enough to sacrifice others – or himself – to save him. So he deliberately prevents Kirk from having that choice.

His sacrifice is therefore also an act of love. He takes the decision away from his friend because he knows that Kirk could not emotionally make the decision Spock has made.

The remarkable thing about Spock’s death is that the most logical character discovers that the ultimate expression of logic is not the suppression of emotion, but a willingness to sacrifice oneself for someone else.

That is not emotionless logic. It is a moral choice informed by attachment.

The film makes this unmistakable through Spock’s relationship with Kirk. His final concern is not really the ship. It is Kirk.

“I have been and always shall be your friend.”

Friendship is an emotional relationship. The Vulcan who spent his life suppressing emotion ends it by affirming one.

But then again, Spock was only half Vulcan. He was also half human.

Perhaps that final emotional display was a reflection of his human side.

And that brings me back to AI.

How would AGI handle a hypothetical situation like Spock’s? Could it understand attachment without experiencing emotion? Would that be enough to govern its decisions? And who would determine the principles behind those decisions – its creators, its training data or human oversight?

I do not have the answers. Way above my pay grade, as they say.

But these are no longer questions confined to science-fiction writers.

They are being asked by people at the centre of the AI revolution: Elon Musk, Mark Zuckerberg and Sir Demis Hassabis, the British Cypriot scientist who has just become chairman of Google DeepMind. They arrive at very different conclusions.

Musk has argued for rapid AI development while warning against undue regulation. Zuckerberg’s emphasis has increasingly been on making powerful AI widely available rather than allowing a small number of companies or governments to control it. Hassabis, meanwhile, has argued for a global AI oversight body, broadly analogous in spirit to the International Atomic Energy Agency, that could establish standards and monitor the development of the most powerful systems. 

For me, the issue is too important to be left solely to the tech barons.

As someone who believes in the importance of institutions, I have little doubt that some form of regulatory framework will be necessary – not to stop AI, but to understand what it is capable of and what it is being used for.

The European Union has already taken an important step in this direction with its AI Act, although even that framework is having to evolve as the technology moves faster than the legislation.

The challenge is not simply to create machines that are intelligent.

It is to ensure that their intelligence is compatible with the values of the society in which we want to live.

Perhaps that is the real lesson from Mr Spock.

Logic is powerful. But logic alone cannot tell us what is worth caring about.

And so, next time my daughter’s AI assistant challenges me, I hope it is still on accounting matters – and not on her choice of a lifetime partner.