Granny and the Robots – Part 2: Will conversational AIs ever be safe for addressing loneliness?

In my earlier article on this theme I discussed the risks of allowing conversational AIs to supplant real human conversations and talked about my own surprise at how realistic interaction with these robots has become.

I was advocating that we should do everything in our power to connect humans with other humans, but I did also enthuse about a quite lovely chat with an AI called Siobhan.

So do I believe that Siobhan and her digital brethren will ever have the capability to interact with us conversationally and in a way that is healthy in the long run?

The contamination problem

Right now the lack of checks (and controls) around the positive reinforcement of unhealthy thoughts facilitated by AIs will, I expect, soon be more fully recognised as dangerous for our mental health and become a public health emergency.

There is already a rise in “AI psychosis” – mental health collapse as a result of prolonged exposure to conversational AIs. This is a trend that is likely to continue, if for no other reason than a vast number of new people are discovering these tools every day.

The fact is, like the contents of a beautiful chalice made of lead and full to the brim with a sweet elixir, no matter how tantalising that first taste might seem, AIs are contaminated by the very datasets they are trained on and algorithms that are not sufficiently refined.

Whilst a singular gulp from a lead chalice is unlikely to harm (to the extent it can be measured at least), extreme care is needed because regular consumption would likely prove poisonous and maybe even lethal.

AIs have instructed people to kill themselves, to murder and have pushed people to the edge of their sanity. Is this what we want to expose our isolated and lonely people to right now?

Just to be clear, I am an advocate of the utility of AIs to support us in our work, speed up research and – in the case of agentic AIs – go and perform tasks for us. I am specifically bemoaning using them as a source of companionship, conversation and as a tool for dealing with isolation and loneliness.

Surely, the tech companies will clean it up?

The financial opportunity for the companies behind these technologies is colossal – and so it seems certain that conversational AIs will have these toxic contaminants removed. The AIs will be adapted and enhanced.

Checks will be implemented and we will celebrate that these are tools now fit for the mainstream. “These AIs are safe” they will announce and we will believe it, because we will want to and are suckers for an easy fix.

Meet your new ultra processed AI companion

It’s increasingly clear that our bodies don’t respond well to ultra processed foods. They’re linked to a whole host of diseases. Something is missing from them, and over the long term they just don’t sit well with our physiology.

Like those ultra processed foods, this next generation of conversational AIs, free of contaminants – might continue to prove corrosive to us, not so much for what they contain but what they fundamentally lack.

As was the case with the advent of cheap white bread or mass produced pasta, we risk making them the mainstay of our diet and gorging on them until we get sick.

It’s not that it’s inherently bad that we occasionally have some white toast smothered in chocolate spread, or indulge in a pizza, but this must be part of a balanced diet.

So what might be lacking from these AIs that could make them fundamentally unhealthy?

Empathy is a micronutrient, and without it we will get sick

Empathy is our ability to detect, feel and react appropriately to how someone else feels and their situation. By feeling something ourselves, it helps us form an appropriate human response.

Hopefully it goes without saying – but computers and their sophisticated models don’t actually feel anything. To even guess at responses, they’re going to need to have a firm grasp of our emotional state, what those emotions actually mean to us and what appropriate action might then look like.

But I’ve heard that AIs can detect emotions?

There is zero doubt that technologies like intonation detection, the ability for a conversational AI to listen to and detect basic human emotions in the patterns of our speech, has evolved substantially.

But, what we might call our intuition – that gut feeling and ability to detect that something is “off”, is more developed than AI can presently achieve partly because emotion doesn’t occur in isolation and, if it wasn’t already obvious – humans are seriously weird and messy.

Emotions occur on a spectrum and there can be complex interactions between dozens occurring simultaneously, making for fraught articulation of them with words alone.

Consequently, a lot of what we feel is difficult to explain even by the most self-aware and erudite person. It is transmitted nearly imperceptibly in the pauses, stutters, rhythm of our breaths and in the unspoken silences. It is transmitted in the tones of our laughs and the precise breathlessness of our sobs.

Equipped with some knowledge of what is happening in someone else’s home (where they live, what’s happening there and the ‘zeitgeist’), knowing their back story, culture, understanding something of their friendships, combined with our subconscious capability to pick up on those unspoken signals – just might equip a real person to be able to respond to another in a way that provides nourishment.

And nourishment might mean agreeing, disagreeing, asking questions, telling a story, simply saying nothing – and a million and one other subtle variations in between.

The incredibly rich data we humans pick up is the source of our empathy – and it seems to me that it’s empathy that drives the right responses when they matter most.

Put plainly, conversational AIs are not capable of this. They are abstracted from the physical world and know nothing of its pain and pleasure, except what has been written.

So I ask you – do you really want your granny talking for long periods to a “psychopath”?

Sentient artificial intelligence (yes, really)

Ultimately, I possess a great deal of faith in our capacity to engineer extraordinary things, but that doesn’t always mean we get exactly what we want, or what’s good for us.

Huge progress is being made in the field of quantum computing. Computers capable of solving certain problems trillions of times faster than traditional computing.

When this technology truly arrives and it surely will, and converges with AI – as it surely will, this will probably give rise to sentient AIs. Effectively entirely new forms of life.

By this time, these ‘’computers” will have access to sensory data through armies of physical robots each equipped with tens of thousands of sensors. They might know what it is to move around the world, to touch and interact with physical objects and living things.

We’ll certainly be having interesting conversations with these beings but whilst they might have been created in our likeness, we must remember they will not be us.

So James, what’s your conclusion?

In the near term could I ever condone the use of conversational AIs as a tool for addressing isolation and loneliness in older and vulnerable people?

Well despite my misgivings, and because I believe we are on the cusp of those very nasty contaminants being removed, then like ultra processed foods – if the “safer” AIs are consumed in moderation (or when there is no other choice) I may be at ease with them being part of a healthy diet that includes ample quality human contact too.

But whatever happens in the future, I’m sure I will always be an advocate of humans talking to humans wherever possible, and I would encourage you to be too.

Whilst we live in a world with millions of isolated and lonely people, I think we should be investing in technology to connect them with each other, no matter how beguiling the alternatives.