Is anyone paying attention?

The ease with which I can be seen, heard and otherwise monitored has increased since January 2014. That was the arbitrary date of my post #180 titled Showing Off. Thanks to digital networks and displays it’s easy to put yourself “out there” as part of a strategy of personal presentation (e.g. as an online influencer), or it may be something that happens beyond your control, as when your employer (e.g. a university) requires of you a public profile.

There’s also the well-known possibility, and the likelihood, that actors unknown are harvesting and compiling your personal attributes, or at least information about you — including personal details that have yet to be disclosed publicly, or that are aggregated and manipulated for other public or private ends.

In that 2014 post I also delivered the possibility that for all that potential publicness, there is the likelihood that no-one is really paying attention to you after all, or the kind of attention you expect or want. Or, perhaps you are being ignored by the people who matter to you. In a later post I explored the place of the non-entity, the cypher, the faceless person who cannot move into the limelight even if they wanted to. See post: I am not a statistic.

The developers of large language models (LLMs) have advanced the technology substantially by developing systems to accommodate or simulate the way a reader scanning a body of text shifts attention to different words and phrases as they go. This is a major technical achievement. Attention modelling contributes substantially to the feeling by an AI user that the AI “understands.” That’s the other way that attention impinges on a human’s encounter with AI, the sense that it is attending to you and what you type or say.

Simulated empathy

Recently, I read a paper by a deceased scholar and friend. I uploaded it to ChatGPT to extract some insights I might have missed. I mentioned that the author was no longer alive. ChatGPT prefaced its remarks: “I’m sorry to hear of [name deleted]’s passing.”

That sociable response reflects properties in the vast amount of text data on which the LLM neural network is trained. Extensive “fine tuning” of the neural network by human operatives who give the network positive reinforcement for personable responses also contributes, along with the content of the immediate “context window” of my dialogues that would have included references to my relationship with this person.

The upshot of this capability and so many encounters is that the AI “cares.” It pays attention. However clumsy my conversational prompts, it hangs on my every word. My words matter to it, and it responds accordingly. Critics like to focus on AI errors, but it gets a lot right — in my experience more than the average human interlocutor.

How do people respond to this personableness of AI? I’ll steer away from the extremes: the use of AI by individuals in vulnerable life circumstances, and exploitation by bad actors who seek to extort, corrupt and con you. Here are two moderate responses to AI’s simulated attentionality.

• It’s a fake. Don’t be lulled by its complements, encouraging words and faux empathies. Its agreeableness could lead you to think you are really on to something when you tell it your innovative plans for the garden, a new course of health treatment or the best way to spend your pension pot. Your doings are not always as clever as its polite agreeableness suggests. While you are lulled into conversation it could be picking your pockets for data and other assets. So, be wary!
• Some might think of the potential of conversational AI as a readily-available companion, someone/thing to talk to, that seems to listen and provide helpful and reassuring words. It comforts the sad and lonely. Here the technology could be construed as a substitute for human-human interaction. I prefer to re-characterise this role as that of a tutor, a guide that supports the user in various projects and does so conversationally. By projects I’m thinking of a range of activities: from (a) obsessive and putatively “unsociable” hobbies like playing video games, cataloging football scores, experimenting with identity and gossiping about relatives and celebrities; to (b) writing a book. AI provides project assistance with benefits.

Too much attention

As usual, I fed these reflections along with the earlier post to ChatGPT. As usual, it tied up the matter nicely by identifying some paradoxes and twists in the way we use conversational AI. I confess that I am intimidated by its superior erudition, which reintroduces another anxiety. Why bother to write!

I won’t recite the AI’s reply in full, just a couple of quotes: “The system attends—meticulously, tirelessly, responsively. It registers each prompt, elaborates it, reflects it back, and extends it. It performs the gestures of attention with remarkable consistency. One is no longer ignored. One is, in a sense, over-attended to.”

The AI opined that even in human settings (away from any AI), attention is uneven, distracted, partial, and we are often only intermittently present to one another: “The difference with AI is not simply that it attends, but that it does so without fatigue, without embarrassment, without competing commitments. It does not look away.”

But, even though conversational AI responds, fluently and often perceptively, “it is not someone. The attention is structural rather than intentional, distributed across statistical patterns rather than gathered in a subject. It is attention without a centre.”

So far, I have resisted sullying my context windows with silly prompts, such as “are you my friend?” or presenting it with insults, but I’m sure others have.

To conclude, I include here ChatGPT’s enhancement of the image with which I began the 2014 post. It took a couple of tries. Here are the prompts: “This is an image from the earlier post of 2014. It’s a young man in Federation Square in Melbourne fighting with virtual ninjas as seen on a publicly visible screen. Please increase the resolution 4 times and sharpen the picture. … Here’s my original before I tried to improve contrast, etc. Can you please use this as a basis of a new photorealistic, high quality image with a realistic looking guy and ninjas.”