Carrying on the series about human-machine relationships, today’s topic is intimacy. I’m not proposing to talk about sex specifically – nor do I in Far from the Spaceports – but a much wider spectrum of close relationships.
In the book, Mitnash has a long term human partner back on Earth. She’s called Shayna, and we only actually meet her in one scene near the start, though she is a regular background presence throughout.
The main relationship that we see is with Slate, his working partner, who also happens to be female gendered. She has no physical form that would distinguish her from any other virtual persona, and with a bit of preparation can adapt herself to a wide range of available hardware.
So their relationship is not on the basis of bodily shape – I didn’t want to write an android book, and the difficulty of getting Slate close enough to the action to be useful is an important narrative ploy. But clearly they are a close-knit couple. As Slate comments to Mitnash about a particular data file she has intercepted,
“there’s actually more about me in the packet than Shayna.”
To which Mitnash replies,
“best not to tell her that, if you don’t mind.”
Their intimacy, the way I see it, rests on two things. Firstly, they share intense and difficult experiences together, supporting one another in them to the best of their ability. But secondly, they communicate with one another in a direct, constant and intense manner. Use of a cochlea implant and subvocal transmitter – originally simply to avoid having to speak out loud in situations where this would be awkward – means that Mitnash communicates not only what he is consciously framing in thoughts, but also a whole other level of half-framed thoughts and ideas.
“Slate, how much do I talk to you without knowing it?”
“All the time, Mit. You murmur to yourself while you’re thinking, and you subvocalise throughout the day. There’s very little about your thought life I don’t know. Or your fantasy life. You’re whispering to me almost all the time.”
“I suppose that means you know all sorts of things I have never told Shayna.”
We are clearly a very long way from this level of artificial intelligence just now. All of the major players in today’s online world have been working on this – Apple’s Siri is probably the best known, but there are many others. At the moment they are all quite gimmicky – after asking Siri what the meaning of life is, and showing your mates that you can send messages and be reminded about events, most people get bored with him (or her in some countries) and the level of interaction drops. Siri and that whole current generation of virtual assistants are just not interesting enough.
Sounding relational, as opposed to encyclopaedic, is a really hard problem in machine intelligence. I think most people remember with dislike Microsoft’s Office Assistant, with its cheerful chatter like, “it looks like you’re writing a letter… can I help?”. I actually thought it was a brave effort back then, but obviously I was in a minority and the whole idea was quietly dropped for another day.
The single best known benchmark for all this is the Turing test – basically you are allowed to chat without being able to see the other person, and have to decide if you are talking with a person or machine… without asking leading questions like “what are you made of?” Part of the test – certainly in the way it is conducted nowadays – is seeing how the entity at the other end deals with abrupt changes of direction in conversation, with ambiguous or poorly defined statements, and with questions where the speaker cannot possibly know the answer.
To date, nothing yet built does very well at the Turing test, despite massive improvements and changes in recent years. As I said, it is a really hard problem, and the numerous “digital assistants” already in use, succeed primarily because they are operating in a very limited domain, with a very constrained set of questions. Do I think we will get there one day? Yes indeed, but I don’t think it will be for a few years yet.
Next time… anticipation and context