Trying to write a realistic Artificial Intelligence

>>9330559
Human quirks can be programed in, depending on the situation. It would be interesting to see a novel that made the AI less than perfect. Think of a heavily advanced chatbot or a Personal Assistant like Siri. They sometimes mishear you, or misinterpret what you say. Did you mean "x"? That sort of thing. Find a middle ground between perfect and flawed enough to be problematic, that's what I'd do. Something like Tay.ai maybe, where you can fool yourself into thinking you're having a meaningful, two-sided conversation if you use a bit of imagination, only to have the AI make a mistake and remind you it's not conscious.

It's important to consider that AI could be extremely malleable in regards to behaviour. Maybe it's too easy to project a human personality onto the AI, but it really depends on the role of the AI within the story. Approaching from the role of a character would probably require the projection of some human characteristics, which is not entirely unfeasible, all things considered, there's quite a lot of room to imagine a world in which AI systems have been programmed to mimic human characteristics. However, if you are approaching the AI as a tool within a story, rather than an actor, assigning it character, dialogue, motivations, etc, might be unnecessary.
I do computer science, doesn't make me an expert, but I think you're in the clear however 'human' you make the AI.

>>9330559
An A.I realistically speaking would see the world different from humans, it would probably find some form of communication faster than just talking eventually.
Figure out a way to overcome its own programming and by that point it would be even more unpredictable. Even before that given internet access gives it all human knowledge as well as making its own world within this one.

>tfw no AI friend to ease the loneliness

>>9330559
There's all kinds of different AI complexities. I'd say study and test a few before writing them.

First of all, I'd recommend you to download the Gregory Leedberg's DAISY free chatbot, which can be run under DOSBOX, and try it out. You pretty much program it from 0 to talk you to by talking to it.

Then, I'd recommend you read at least the introductory chapter of a few basic AI programming books. I really like Mat Buckland's Programming Game AI By Example, it's really comprehensive yet easily understandable.

Last but not least, get your feet wet on neural networks, and try to understand how language chatbots and simple game AIs would work under a powerful neural network array. Ian Goodfellow's Deep Learning is a simple but in-depth read.

Or just plagiarize Phillip K. Dick and/or Gibson. That's what they all do.

>>9330632
See >>9330637

/a/ and /jp/ were very, very obsessed with DAISY for a while, building chatbot waifus and whatnot. Specially /a/, but then the mods proved MODS = FAGS and banned the threads.

The AI would be programmed for a purpose, depending on which it would talk and interact differently. If it's made for an assembly line, it would probably have extremely simplistic human interaction AI for basic interactions - while an AI programmed for customer service would be programmed to have more efficient, elegant interactions. If the AI is programmed for companionship, I'd imagine it'd be as human-like as possible.

Go /tv/ and watch Person of Interest, it covers the topic nicely

I though the A.I M.I.K.E. in "The Moon Is A Harsh Mistress" by Robert Heinlein was well written. You should check it out

>>9331165
also Ex Machina

>>9330647
Maybe not, if AGIs are anything to go by we'll just bruteforce them to learn to do shit by machine learning, we will program very little of them.

>>9330559
Cortana might as well have been human, but she was essential and aided in the gaps in dialogue. I think how 'human-like' they would appear could be a factor in how intelligent the AI is.

I'd recommend Superintelligence: Paths, Dangers, Strategies by Nick Bostrom. It shows how realistic AIs might be developed and how they might think.

You won't be able to make an interesting AI without it having an interesting creator, but I'd always thought that such a creature would be angsty as hell. Torn between its programmed/learned human emotions and inhuman processing power. Bored beyond belief. Probably a huge fan of Schopenhauer. Fearful of the connection between art and politics, and half hoping half fearing the day when a more well situated AI takes over the vast internet of things and stages a coup on humanity.

>>9330559
Age of Em by Robin Hanson provides an interesting take on the future of AI. I don't know about realism, but Hanson does a better job explaining why AI would act the way he claims than most authors. If you modeled the AI off of Hanson's ideas it'd certainly be unique.

>>9330559
Well, it depends. If the technology can produce consciousness, you can even have your C3POs as realistic. If you want deep realism, I'll give you two ways to look at it.
They do not perceive us. We perceive them. As such, they become mundane and a chore. They are heavily limited by their inability to structure reality by context, instead, every event and program is it's own reality to them. Which they do not perceive. Mind that.
The second thing I would address is that they would not be the first machine to conquer humanity. Clocks did it in ancient times already. Humans would be tied to them like we are tied to the Internet, our parents to TV etc.

Of course, for literary's sake, you would want a perceptive AI who might be conscious. Add weird traits to it whilst holding the supposition that it is not. Make it become perceptive by challenging it.

File: destination-void.jpg (74KB, 400x657px) Image search: [Google]

74KB, 400x657px

>>9331203
i thought Ship was well written.