Sulphur on 24/12/2019 at 15:25

---

Eliza is an immediately interesting idea for a visual novel: what if you gave the old Eliza program from the 60s emotional reasoning and natural language processing capabilities, then sold it as a sanitised, vacuum-packed, corporate-branded wellness package?

So: AI counselling, with humans delivering a teleprompted script as Eliza 'proxies', or glorified chatbot voices (if you've seen Her, think back to the sexual surrogate scene, but replace sex with therapy). But it's a bit of an oxymoron, surely. Why would anyone listen to a machine's take on their lives, especially when they know they're talking to one? One of the signs of how observant the writing in Eliza is, is that the very first session forces you to confront that idea, and the solution is exactly as disquieting as you expect it to be.

It's not about a dystopian near-future though. Unless that's how you choose to look at it. The narrative presents a few different perspectives on the technology alternatives of the future, and while there are biases inherent to the story's framework, it does a fair job of presenting each option with its own pros and cons. It's a thoughtful experience; if I had only one adjective to choose to describe this visual novel, it's that: 'thoughtful'.

Take, for example, the perspective from a c-suite executive about how you can quantify everything. And he's not wrong: when he talks about numbers describing emotions, it's hard not to think back to the counselling sessions where each patient's heart rate, mood/affect and receptivity are measured (affect gets a baseline at 0, with negative and positive affect swinging either way respectively), and it observes things like motor agitation (theoretically: eye movement, hand gestures, general body language) to get a read on the patient's current emotional state. You could crunch all that down to digital scales - that's what the technology in the game is doing already.

And yet: surely that isn't the whole story. What about everything that numbers don't describe? That they can't? The contexts we chain together in our heads, all our unspoken histories and subconscious triggers?

You don't get to ask all those questions through the game. But it remains open to questioning your stance, thinking, and deciding whether to go this way or the other. No judgment. Just your own.

Thoughtful, like I said.

Does it have flaws? Yes, it does. The character writing for the main character is oddly ambivalent. She doesn't feel strongly one way or the other about anything - perhaps because she functions a bit as a tabula rasa for the player to control, but she's very much constructed with a history. It's an odd conflict of personality and game intent. The characters you play off of are interesting, but none are really compelling. In fact, some of the most well-observed personalities are the clients you face: specifically, a girl who wants to be recognised for her art and achieves a path to self-realisation through therapy, and a self-absorbed twit (literature faculty, mais oui - no, I'm not singling out Thirith anyone, I just know the type seeing as I myself am technically a literature student) I wanted to stomp into the carpet until I could hear the *crack* *crack* of splintering vertebrae and ego bounce off the walls. The game lets you do that figuratively in the final act, and good gravy it is SO REWARDING when you're given the option to break free from the script.

There are multiple endings. I'm happy with the one I got that involves just saying 'fuck it' to all the competing enterprises and kicking off to make hot electrosynth music with a girl. (There's also another motivation involving the future of AI, but I won't go into it because [spoiler].)

As I watched the end credits roll, there was something in my mind that kept turning itself over. In one early conversation in the game, that CEO character remarks that we'll be able to tell we've made an actual, general intelligence when it can one day write its own poem. The funny thing is, Elon Musk's OpenAI program unleashed GPT-2 on the world recently, and (https://twitter.com/rossgoodwin/status/1131442974812823552) here is a poem* its algorithm recently wrote. It's not very good, but in a way it's also human, somehow, because of that.

And that's the kicker: maybe in the future, not too far away, we'll have a game like this, very good but not perfect, and for all we know -- it would have been written by a machine.



*Don't be scared, though: in function it's really only a very sophisticated evolution of the Markov chain, with natural language inferencing and a vast amount of source data to play with. This is not general AI - not yet.

Pyrian on 24/12/2019 at 23:45

---

Might try Subsurface Circular; not exactly that but kind of similar.

twisty on 27/12/2019 at 15:04

---

Whether AI sentience is mathematically possible or not, I suspect it may only be a few decades at most before it is beyond our human abilities to tell the difference between a person and a robot. At present, the uncanny valley creepiness of even the most advanced robots leaves something to be desired, however considering that voice and chat based automations already have the ability to mimic humans quite effectively in a limited range of functions, it's not much of a stretch to envisage the possibilities of a near future state where an AI with a split microsecond recall of entire libraries of best practice knowledge at their disposal could be utilised to provide cost effective (and possibly better) treatment. Of course, the current economic realities probably makes this sound a little naive.