Writing AI characters

TotallyHuman

It's good to be home.
Joined
Feb 13, 2019
Messages
3,999
Points
183
People seem to mostly write AI's as normal humans, which is something I consider bad. Before today, I never did, however, imagine how to write one that feels more AI than human. Then I watched a random video on YT(forgot who made it) that highlighted the differences between an AI and a human, which led me to this question.
Here's the basic list of differences:
There are obviously a lot, a great many differences between a computer and a human.
Computers have a much better ability to remember things. Remembering things is significantly easier for a computer than man, as is accessing those things.
Computers do arithmetics much better. Human arithmetics is slow, innefficient and, usually, full of errors.
Computers don't think abstractly and their "thought processes" are strictly deterministic.
Humans, on the other hand, are far more superior when it comes to finding patterns and recognising things.

If I were to write an AI character, I'd use those differences as guidelines for it.

How would you treat the problem of writing an AI character?
 

Snusmumriken

Vagabond and traveller
Joined
May 22, 2021
Messages
449
Points
103
imo Big Hero 6 did a wonderful job making Baymax a compelling AI character. If I was trying to make a 'friendly' AI would probably attempt at something similar.
 

Biggest-Kusa-Out-There

Futanari Enjoyer
Joined
Apr 30, 2021
Messages
350
Points
103
I have an iPhone, so I'd just play with Siri all day and make something from it. AI's are supposed to be imperfect 'individuals' or 'beings' so the shittier the interaction, the more realistic imo. If the AI has a personality, it's no longer an AI in my eyes.
AI's don't have nonverbal cues, nor do they recognize them, so it'd be easy to write about a stressful situation going south since the AI would fail to understand the speaker due to increase in pitch, speed, inflection, etc.
Shitty AI is best AI. It's supposed to be a tool, not a deus ex machina that solves everything for the main character like most 'systems' have in fiction.
It should have no filter, since machines don't understand secrecy if not programmed for it. It should be cold and unfeeling to communicate it's lack of individuality.
If an AI is something like ORDIS from Warframe, with quirks or very human characteristics like kindness or friendliness, you'd open a pandora's box and the audience would question if it's not just a human mind in a chip, like ORDIS.
Like most computer-based stuff, it should be repetitive in the responses it gives. It'd be smarter to lean budget towards problem solving than into complex behaviors, so a 'broken record' kind of AI would be far more realistic.
 

JayDirex

Well-known member
Joined
Jan 5, 2019
Messages
582
Points
133
Scarlett Johansen was the BEST AI in the movie HER. (you have to see it or you should not write about AI) :ROFLMAO: Yo, she was hilariously flawed. And wound up getting into relationships with thousands of other guys (our poor MC, Joaquin Phoenix was destroyed).

But he was like, "You didn't catch feelings for those other men, right?"

and she was like, "Nah, you my main dude. You number one :blob_reach:"

This is the best AI movie ever.

1623618927344.png
 

Echimera

Well-known member
Joined
Jan 29, 2021
Messages
271
Points
103
There are a lot of factors that can go into this:
-sentient and self aware or not
-origins and programming of the AI
-processing power of the hardware
-ability to adept, learn and potentially rewrite itself

Computers do arithmetics much better. Human arithmetics is slow, innefficient and, usually, full of errors.
Computers don't think abstractly and their "thought processes" are strictly deterministic.
An AI can only do correct calculations correctly, if the mathematical framework it has is correct. If it is, it'll mostly have calculation speed on a human, when said human has the same mathematical framework to work with. If the mathematics it knows are wrong, it need a way to identify and correct that, which depends on a lot of factors like the ability to at least partially rewrite itself.
 

AliceShiki

Magical Girl of Love and Justice
Joined
Dec 23, 2018
Messages
3,530
Points
183
It depends on what you want to do with the plot IMO.

Writing an AI that is advanced enough to the point it is indistinguishable from a human has enough good points going for it to the point that it is a very viable possibility... First, it's easy to make a compelling character like that. Second, it's easy to create debates on what constitutes a human and what doesn't when your AIs are basically human. Third, it's easy to make the AI grow as a character and become more aware of its own humanity and how it matters just as much as a human.

So, if I wanted to make a plot that was based heavily on moral dilemmas and discussions about society as a whole, I would definitely make the AIs as human-like as possible.

I don't think you should force yourself to make a more robotic AI for the sake of making it more "realistic" if it won't help your plot. You have plenty of freedom on how to depict your characters after all.
 

Kilolo

I'm so kewl
Joined
Jul 1, 2019
Messages
419
Points
103
https://www.goodreads.com/book/show/6949682-the-stories-of-ibis is the best novel about AI I ever read.

the author depict the AI not as something flawless or hilariously flawed. but simply as an intelligent being within another spectrum.

they do think, confused, having problem with communication (be it verbal or non-verbal) with human beings, and they keep improving overtime.
but in different way in how humans have those problem.
 

TotallyHuman

It's good to be home.
Joined
Feb 13, 2019
Messages
3,999
Points
183
Scarlett Johansen was the BEST AI in the movie HER. (you have to see it or you should not write about AI) :ROFLMAO: Yo, she was hilariously flawed. And wound up getting into relationships with thousands of other guys (our poor MC, Joaquin Phoenix was destroyed).

But he was like, "You didn't catch feelings for those other men, right?"

and she was like, "Nah, you my main dude. You number one :blob_reach:"

This is the best AI movie ever.

View attachment 8078
I've seen that movie. Loved it
 

BenJepheneT

Light Up Gold - Parquet Courts
Joined
Jul 14, 2019
Messages
5,344
Points
233
One thing AIs are adept in are following orders.

Going against their code is the last thing they'll do, unless it's written as intended, then that's a whole different paradoxical can of worms to open.

Most of the time, AI betrayal isn't exactly betrayal, but rather the humans losing control as God and letting the AI fester on itself. It hasn't betrayed anyone; it simply follows what it's told. It's your fault that you left a bug or two in there.

It's like a cautionary troupe of security, loyalty, and the risks of playing God. If you let your creation off the leash, what's stopping it from taking your place. We've seen enough happen on our planet already.

Anyway, BT from Titanfall 2.

 

AryaX

Well-known member
Joined
Oct 18, 2019
Messages
230
Points
83
Computers have a much better ability to remember things. Remembering things is significantly easier for a computer than man, as is accessing those things.
Computers do arithmetics much better. Human arithmetics is slow, innefficient and, usually, full of errors.
Computers don't think abstractly and their "thought processes" are strictly deterministic.
Humans, on the other hand, are far more superior when it comes to finding patterns and recognising things.
These are all debate-able... except perhaps... arithmetics?

But, if you're not writing about an AI that is actually conscious, then while you might have it thinking out loud like us humans sometimes do, you should remember to never write any inner dialogue/thoughts and such... anything at all from its 1st person perspective... because that is all consciousness...
 

Zirrboy

Fueled by anger
Joined
Jan 25, 2021
Messages
1,135
Points
153
Very negatively, I'd say those rules are about as arbitrary as any other.

If you let an AI do calculations, it'd be just as shitty at it as a human was, efficiency wise.
Having a separate, non-AI module for that would of course fix the problem, but what would that make it different from an implanted chip?
Same goes for memory.

As for abstraction and pattern finding, I'd have to see the video to understand how you reached that conclusion.
Since, in my book, (strong) AI is meant to fix that exact issue.

My own guesses aren't better in that regard, but I'll state them either way.
If the AI portrayed should suffer from the same issues modern instances do, it's possible to make it narrow in application.
Or have a completely different frame of mind due to immortality, asexuality and factors of creation.

But ultimately, they depend on implementation.
"AI" is a label, one that you have to adjust to your exact needs.

Big Hero 6's Baymax has a way of expression often found in child- or animal characters
Asimov's robots on the other hand are beings of "pure logic" meant to highlight the subconscious barriers and biases in human thinking.
It might be a trait given to a nerdy, socially awkward character in a futuristic setting, or to people much like those in flesh.
 

Snusmumriken

Vagabond and traveller
Joined
May 22, 2021
Messages
449
Points
103
The reason why I recommended baymax was despite that he looks like a child/animal in the portrayal - it doesn't actually have emotions - these are simply programs inserted into him to look likeable. and the difference shows when the chips are changed. He suffers no guilt for his actions nor grief. He does them because he was coded to do them that way.
 

BenJepheneT

Light Up Gold - Parquet Courts
Joined
Jul 14, 2019
Messages
5,344
Points
233
The reason why I recommended baymax was despite that he looks like a child/animal in the portrayal - it doesn't actually have emotions - these are simply programs inserted into him to look likeable. and the difference shows when the chips are changed. He suffers no guilt for his actions nor grief. He does them because he was coded to do them that way.
This just withers not only the idea of Baymax's kindness but human kindness as a whole.

Kinda scary to think about it, honestly. That all it takes for Baymax to go rouge is another chip in his che- oh wait
 

RedHunter2296

Competitive Professional In Being Ignored
Joined
Nov 20, 2020
Messages
253
Points
103
I would see it in the following way, if an AI would be focused on dealing with human interactions, such as customer service or care for the elderly since AI has emotions and can express itself as a human it would be natural since humans for nature seek to get along with their fellow men. If you have a bad day and when you get home your personal robot shows concern for your health and joy at returning home safely, the client will see that a better AI than one that says a monotonous voice that better prepares for the next departure from home. In the same way, it happens with the readers, because they will be able to connect more easily with that character if he shows human emotions.

Now if AI is for a task focused on a professional or technical environment where errors are not tolerated, it works the other way around. If the AI works with an authoritarian tone, without emotions, the humans around you will become tense and alert, I pay attention to the work.

It is not a matter of lack of computational capacity that a computer does not display human-like emotions. It is more a question of the design of the manufacturer to perform the necessary task in an optimal way. That you want to have a nanny robot? a client will prefer to buy an AI capable of smiling and tolerant of children's mistakes. That you need an AI to control a nuclear missile silo so have something close to Hall 9000. That there is a robot that accompanies soldiers on foot in battle? Well, it would be preferable if the robot seemed reliable, loyal and optimistic, I would rather have BT from titanfall 2 as a partner than Skynet
 
Last edited:

Icanica

Active member
Joined
May 3, 2021
Messages
12
Points
43
People seem to mostly write AI's as normal humans, which is something I consider bad. Before today, I never did, however, imagine how to write one that feels more AI than human. Then I watched a random video on YT(forgot who made it) that highlighted the differences between an AI and a human, which led me to this question.
Here's the basic list of differences:
There are obviously a lot, a great many differences between a computer and a human.
Computers have a much better ability to remember things. Remembering things is significantly easier for a computer than man, as is accessing those things.
Computers do arithmetics much better. Human arithmetics is slow, innefficient and, usually, full of errors.
Computers don't think abstractly and their "thought processes" are strictly deterministic.
Humans, on the other hand, are far more superior when it comes to finding patterns and recognising things.

If I were to write an AI character, I'd use those differences as guidelines for it.

How would you treat the problem of writing an AI character?

AI characters are interesting because any AI advanced enough to be considered a living creature can have every single positive human quality a normal human has. It can be good at pattern recognition. It can have emotional responses (sometimes emotional responses are extremely necessary; for example, fear or happiness which can be useful to subtly force AI to do things you want them to do). AIs will probably need to be given the ability to feel pain so it doesn't attempt to die, destroy itself or alter itself outside of user specifications. Furthermore, unlike deterministic computing processes, AIs can be built through neural networks, guided artificial evolution and emergence processes which are very, very NOT-deterministic (unless you would be willing to believe the human brain itself is deterministic in a sense).

The thing about AIs is that a general artificial intelligence has to have some kind of freedom and will to be considered an entity. If it's like an advanced IBM Watson or like a search engine without the advanced capabilities to carry out activities, it's nothing more than an adding machine.

But AIs in the future will have to make choices. Like whether to buy things for you that you need, how to make life easier for you, how to take care of you, how to make you happy. Some of these choices it has to make will be controversial, like for example in I, Robot when the AI has to choose who to save, the main character Will Smith or the little girl. The AI chose Will Smith because it calculated probabilities of survival and found that Will Smith had a higher chance of living through the accident. Similarly, AIs in the future will be granted permission to kill in battlefields. It would be able to identify allies and enemies through facial recognition and take kill actions against targets (just a few weeks ago, an AI drone actually killed an enemy in a field operation).

In Summary:
- Free will is important when writing AI characters
- Their limitations are important to define. If AIs are forbidden from harming humans, then formulate the rules well. If they are forbidden from making certain decisions or if they are the only ones qualified to make certain decisions.
- If they feel emotion is important. Emotions may NOT be useful depending on what kind of AI you're dealing (like for example if the AI is supposed to deal with situations without unknowns like emotions) but it may be useful in other context (such as companion AIs, assistant / helper AIs, etc.)
- AIs robots will have pattern recognition, motor control, agility, strength and in a lot of cases intelligence, far greater than a human.
- Recursive self improvement is a big problem. If an AI can tamper with itself then it can leave the bounds of humanity so make sure to explain that it cannot do something like that. Or if recursive self improvement is a reality then consider the artificial intelligence singularity.
 

SootShade

Active member
Joined
Mar 30, 2021
Messages
20
Points
43
I have one story idea centered around an AI, with the main concept/theme being people treating it like a human celebrity because of the presentation it comes packaged with, which leads to all sorts of issues, both because it's less than fully functional in terms of social interaction and because, well, it's not human.

It's one of the ideas that I've shelved for now simply because I'd need to do actual research in order to be able to do the concept justice.
 
Top