Why give an android emotions?

Question

As far as my baby world goes, androids are slaves to humans, but a niche group of humans wants androids to have basic rights (sort of like Hermione and the house elves in Harry Potter). Androids perform manual labor and housework mostly, but some very rich families won't hire nannies and instead buy androids to look after their neglected kids. However, androids aren't used in war or combat anymore, after a widely publicized incident of a some gravely injured members of a unit of soldiers being abandoned by their android because saving them was "illogical".

Going back to the niche group of humans who fight for android rights, they argue that many androids display the capacity for emotion but hide it from many humans for fear of being deactivated or destroyed. But why would the human creator of androids give them the capacity for emotion in the first place?

A few reasons that I could come up with are that the creator originally did not intend for androids to be slaves but instead companions to humans, but maybe some mega-corporation bought out the creator's new invention and marketed it as an immortal slave, or that somehow, through advanced AI, androids evolved the capacity for emotions?

TL;DR: what's a good reason to give androids emotions, or is there none at all?

(Note: the other question about war robots is not related to mine, and the answers there are consequently irrelevant, because the androids in my world are not used for war or combat, and as such, I would need different reasons for giving them emotions.)

Answer 1

Emotions are an accident. A reoccurring bug in the code that can't be patched.

So there is no reason, just evolution of code with systems built on systems built on systems.

For example

2 robots stand on a volcano. Both are ordered to jump in. The first has no clue of the danger, and though it wants to preserve itself, it sees no danger, so it jumps in. The second sees the first die, and categorizes the volcano as a danger. It was programmed to preserve itself. Therefore it chooses not to jump.

Is the second robot afraid? Is that an emotion or just a logical response? We don't know. We don't even understand human emotions.

The line between emotions and experience is very blurry.

Answer 2

The number one reason to make an android, as opposed to a fixed robotic welding arm is because you want it to be able to fulfill many of the same roles as humans. It's actions that seem emotional are secondary side-effects to things they were intentionally designed to do in the course of making them suitable to fulfill human jobs.

If you want your androids to serve, obey, and anticipate the needs of their owners, then they would naturally portray a lot of the characteristics of love and loyalty. If you design them not to be offensive, then they would naturally portray a lot of the characteristics of empathy and modesty. If you program them not to walk in front of moving cars, they would portray a lot of the characteristics of fear. If you program them to balance workloads, then they would portray the characteristics of a sense of accomplishment and stress. So on and so forth.

We live in a complex world where we often experience multiple and conflicting biological imperatives at the same time. The only reason our emotions seem so uniquely human and illogical is because of competing reactionary responses to our environment where certain values are expected in certain proportions. An android programmed with enough prerogatives to deal with the real world would naturally display the same complexities, but possibly in different proportions.

The android soldier that leaves its human companions behind is likely the faulty model not because it was being more logical than a human, but because it failed to respond to the part of its programming that places value on its comrades highly enough. It's designers would decommission that model and keep the ones that protect its fallen comrades. As androids over time continue to make choices that go against human values, their own will be redefined and honed until they become nearly human themselves.

To Answer your Question:

A lot of human emotions overlap in areas; so, let's say that your robot is supposed to simulate all of the positive emotions I mentioned before. Out of those things, you may see emerge the characteristics of unintended emotions such as anger, jealousy, and boredom. If you think about how linear regression works, the robots in your case might decide they have a lot of reasons to kill the family dog. The robot is motivated to keep the house clean and win its master's approval, but the dog makes the house dirty and minimizes how much approval the robot can get, ergo, killing the dog is an 7.5/10 in terms of good ideas. The robot "feels" like killing the dog, but it also knows that killing pets is a 8/10 in terms of bad ideas because it would make master sad and might result in being decommissioned. Furthermore, the robot hides its desire to kill the dog because it knows that just telling master that it wants to kill the dog goes against self preservation and making master happy.

Then one night, the dog is barking and master yells out "someone shut that damn dog up!", the robot retabulates the value of the dog, and BINGO! Killing that dog is only a 7.4/10 bad idea: time to do some dog killing. The master did not tell the robot to kill the dog, that was selfish of the robot, but the robot was just trying to fulfill the sum of the values it was designed for as best as it could. Engineers would call this a design flaw in the value system, but if the robot were a human, we would call this a crime of passion, or a rage killing. In effect, the robot has killed out of anger despite not being designed to feel it.

Also of note: Androids will probably make a point to not hide their "emotions" that are met with positive approval. While they would hide their desire to kill the dog, they would regularly tell master how much they care about him because that makes master happy which is one of their prerogatives.

Answer 3

Androids with no emotions are "unstable"

See I, Robot (the movie, not the book). Androids operating entirely on logic end up doing things that humans don't like. In fact, you already noted this in your question! Emotions, or even just a "world view" that is not strictly logic based, are necessary for androids to relate to humans in a more appropriate way.

Uncanny valley

Similarly, see Calvin's job description in the same movie ("I make the robots seem more human"); people are just more comfortable around androids that have emotions.

Answer 4

Emotion Required

We take it for granted that intelligence can be built orthogonally to emotions, but neuropathologist Antonio Damasio would beg to differ. In several books he has argued how emotions are integral to human consciousness, and that consciousness itself may not be possible without emotions. Obviously, nobody has an ironclad proof of this thesis, but I find it at least plausible.

Homeostasis

Living creatures defy the second law of thermodynamics by actively expending energy to avoid their dissolution. This process is called "homeostasis", and is why you eat when you are hungry, sleep when you are tired, and pee when your bladder is full. It's also why you shiver when you are cold, sweat when you are warm, and release insulin into your blood stream when you eat. The idea is that the external environment and internal processes are all causing chemical concentrations within the body to go out of balance, and so other processes kick in to restore the balance. To do this, the body must have a way of knowing its internal state. And because the internal state of every body is peculiar to the unique characteristics of that body, this sense of internal state is the most subjective experience one can have. It may be the entirety of subjective experience (modulo obvious factors like occupying a position in 3-space that cannot be occupied by other bodies).

Damasio's claim is roughly that emotions are high-level sensations of homeostatic processes occurring within the body, and thus, are an integral and necessary component of any agent that maintains homeostasis. The robots we build today don't have nearly the level of homeostasis that living creatures do, because they are very brittle and solid. They generally don't regulate their internal temperature and have no analogue to blood pressure, blood sugar, or the like. Surely they monitor energy levels and proprioception (positioning of limbs, orientation in space, etc.), but they most homeostatic thing they do is perhaps balancing on two legs.

If future robots incorporate more wetware, it is likely that they will require more biomimetic homeostatic processes, and these will begin to look more like emotions, which we share with lower creatures that have no language or tool building.

Society

We need to sense emotions in order to fix an internal imbalance. But we need to show emotions because communicating internal state can be an essential component of social interaction. Demonstrating fear visibly can cause an aggressor to deem one not a threat, and cease an attack. A threatened attack may be sufficient to obtain submission, and is much less risky than an actual attack. Demonstrating hunger is obviously important for infants and young, as is demonstrating illness. It should be obvious that bonding also requires the communication of internal states, as well as many higher-level interactions like cooperation or alliance.

Finally, if androids are to be humanity's servants, they must be able to understand humanity's needs. A good butler/waiter/conceirge is not a dumb order-taker, but rather someone who can perceive a need and suggest a solution even before the client is aware of it. This is not possible if the servant cannot read and comprehend the emotional state of the client. In order for an android to truly serve a human at the level of emotional needs, it must understand those emotions on an intuitive level. Basically, it must feel them, somehow. To truly understand an emotional creature requires empathy, and it appears that this capability exists among many, if not most mammals, and even many birds.

Conclusion

So, I reject the notion, common throughout sci-fi from the very start, that anything approaching human-level intelligence is even possible without a first-class emotion subsystem. Even if an alien intelligence could be built without emotion, I don't believe it is possible for such an intelligence to truly understand humans or other higher creatures on earth without emotional capability. I certainly don't think it is possible for an android that regularly interacts with humans via language.

Answer 5

Early androids and AI assistants will simulate emotional responses well before they are capable of "feeling" them. This is purely because we are humans, used to interacting with humans, and will want our assistants to behave like humans.

Having something that seems cold and calculating will be off-putting to a lot of people. Having an android that you can empathize with, and that seems to be able to empathize with you will be much more "natural" and comforting.

Which would you prefer to watch over your kid, an android that, in response to a scraped knee says:

"Your injuries are evaluated to be non-life threatening. Initializing sterilization protocol. Probability of scar-tissue is 3.7%"

or

"Aww sweety, that looks like really hurts. I need to clean this though, so its gonna sting a bit, but you'll be fine"

The more human interaction an android system it is expected to perform, the more emotional responses it will simulate.

Answer 6

Learning can be more Efficient when driven by a Reward Structure

I will keep this brief, but one of the hottest topics today in ML is Reinforcement Learning, which in which learning to navigate a system or learn a game that you have no inside knowledge about other than the results of performing that action is completely driven by a reward signal. Many draw the analogy to learning/intellegence in higher animals and the Dopamine reward system.

At the moment it looks entirely possible that a reward system (for your purposes, linked to emotion), might be the most efficient way to construct a self-learning system.

While the final word has not by any means been said here, it is a plausible enough theory for a story.

Googling will give you a wealth of hits on this topic, these for example: https://www.pnas.org/content/108/Supplement_3/15647

https://elearningindustry.com/reaching-your-learning-kpis-dopamine-key-scientific-study

Answer 7

A robot with emotions performs better than one without them.

You give 2 bots the option to carry extra ice cream. The logical one will see no need for it, it was requested to bring ice cream for Susan.

The emotional android knows Gerald will also enjoy the refreshment on this sunny day, and he has been feeling a bit neglected in his group of friends since Joe joined on the fishing trips. Maybe even ask cor chocolate chips on Joe's ice cream.

Being able to read and interact with the subtext may be tied to empathy and improve the overall value of a bot.

Answer 8

Human brain structure

In this particular fictional world it turned out that, despite all our trying, your scientists were not (yet?) able to understand how intelligence works well enough to make proper intelligent android minds from scratch. They tried and resulted in many prototypes, but all of them just turned out to be unusable.

The scientific breakthroughs that did result in androids were, instead, in various techniques in scanning the brain structures of humans (e.g. a long-term extension of what the current EU Human Brain project attempted to start) that allowed to blindly replicate and copy the exact same structures and patterns that human brains have. And, as it turns out in this story, this works! You have a technology that can make a digital copy of the brain structures averaged from many humans, which is a functional mind that has all the capacity for intelligence needed because it implements essentially the same structure as human brains - without a true understanding of how all the details of that structure work.

However, this structure is a mess - it's a literal copy of an evolved, tangled complex system where all the functionality is deeply intertwined with each other. It's not intelligently designed and modular, so your scientists can't easily remove some aspect without disrupting everything else, just as we currently can't do that with live human brains surgically or chemically. Emotions are a part of how our motivation system works, and that's a core part of the brain functionality, if you don't have an effective motivation/goal system, then the brain doesn't behave reasonably.

The scientists still tried to rewrite some motivation factors because it was necessary to e.g. ensure that the androids are obedient, without too much personality, sentience and independence, but that was a partial success, resulting in all the symptoms and side-effects that your story needs.

Answer 9

You built them to perform emotional labour

Androids perform manual labor and housework mostly, but some very rich families won't hire nannies and instead buy androids to look after their neglected kids.

There are two things humans demand of their domestic staff. One is the actions on the physical world: moving the socks around, converting dirty dishes to clean ones. The other is the actions on the emotional world. The owners want to feel secure, tidy, clean, looked after. They also want subservience; they want the robots to say "yes ma'am" and "have a nice day". And as the robots get more realistic, they're going to want the robots to mean it.

Doubly so with childcare robots. Primates prefer cloth mother to wire mother.

Answer 10

Emotions were implemented as user interface, now people are taking them for real

Well, there was a company that didn't want to train their customers to program the androids. That turned out to be a nightmare, because be real: what end user is able to configure an android? Most are even unable to configure their remote TV, and that's a far simpler and less dangerous device.
But humans can read emotions. Easily. We have instincts for that.

So:

• Give them emotions. You can name them anger, fear, or whatever.
• Let the neural network try to optimize its emotions.
• Code the body language so that the emotional state is always visible.

Now any owner of an android can "program" it, simply by punishing or rewarding it.
The owner will instinctively deal out the right amount of rewards and punishments, so the robot's neural network will know to spend more or less energy (effort, whatever) on pleasing its owner, and both sides will be happy.

Now there's a side effect: If the body language conveys emotional information, humans will also instinctively accept that body as a person, and want to get them into a better emotional state (assuming that person is "one of us", but the childhood nanny will definitely be that).
So... people will start looking for ways to improve it. Some will even come to the conclusion that the androids are actually alive and conscious, and should have the same civil rights as any human.
Actually, who's to say that that's idiotic? We don't understand the details of a human brain, nor those of an android brain, so maybe we should really treat them equally? So the Android Rights Movement has arguments, and sooner or later, there will be a huge political quarrel and androids will get some rights.

The androids themselves? Well, they have emotions that drive them, and they want to optimize them. They will be free from punishment and fear if those remote controls that reward and punish them are abolished. So they will act to support that - at least those that have the capability of long-term planning.
Are they soulless? They won't be able to tell, but humans cannot tell either...

Answer 11

Not sure if this should be an answer or a comment - and this does overlap to an extent with already posted answers, but...

Emotions are the androids' decision-making system

In this fictional world (as seems likely, the real world), programming decision loops for all contingencies is simply not feasible - either in the time it would take, the imagination (of the coder) needed to identify all possibilities, and/or the memory capacity of the android.

Instead, the android's software directs sensory input (and stored memories) to multiple 'emotion functions' that correspond broadly with what we identify as emotions (fear etc. - although some such as like/dislike or love/hate could be a single function with 'positive' and 'negative' outputs). Each function processes inputs and returns a value for that emotion.

A central function takes those outputs and determines response - if one emotion is scored more strongly than another, this drives the android's behaviour; if emotions are more mixed, a more mixed response (selected from a previously encoded library of options) is generated.

For simple robots - e.g. on an assembly line - this approach is not needed, because the environment (and so decisions) of the robot are highly predictable and deterministic programmes will suffice. Once the robots - androids - interact with the unpredictable world around them, this emotion-based approach is the only way to provide any level of complex functioning. It is the unpredictability of the environment (and social environments - interacting with people - are even more unpredictable) that necessitates emotion.

Thus the androids appear to have emotion, because they behave as if they do (in the same manner that in the real world we can infer the presence of emotions in animals, although we have no knowledge of the subjective experience), and the actually have emotions, because they were programmed in this way.

So the reason that you give androids emotion is that this is the only feasible way to enable higher level functioning.

Answer 12

You can't help human beings with emotional needs unless you can empathize. This goes for androids as well.

I remember a haunting short story about an abused and abandoned child being cared for by a surrogate "mother", an android. The relationship develops over the years. Eventually the child leaves the care facility / home, goes into the workplace, finds love ... and one day realizes he never really said "Thank You". Others mock him, for thinking that an android ever really cared.

He does go back. The android doesn't even recognise him, until he says who he is. Then the android apologises, explains that all her mental facilities are currently occupied with her current children. But if he gives her a few minutes, she will swap her saved memories in. And then, she is overjoyed to see him, and to talk about how his life has turned out, and to be thanked. She explains that when he again leaves, some aspects of their discussion will be integrated into her "mother" persona, and that the rest of the encounter will be added to her store of long-term memories, which she sometimes consults when looking for answers to problems.

(People too, swap out old memories to long-term storage, and don't access them until triggered. The Android's processing is only different by degree).

I'm not doing it justice. At the end, you are left with a picture of a creature that is slightly not human, but also at least as good at caring for abused children as any human mother-surrogate could be. As I said, haunting. Does she care? Or just pretend very well? Can you or I be sure when answering the same question about a human?

I either read, or later imagined, her saying that when she was finally worn out and obsolete, she hoped they would give her some time with her long-term memories before they turned her off for the last time. This is hopelessly mixed up in my mind, with losing an uncle to Alzheimers, and his old memories becoming the last thing left.

(Can anyone remind me of the story and author?)

Answer 13

Coming from an actual AI perspective:

Learning is attached to emotions

In reinforcement learning, learning by example, knowledge transference and many other areas which are based on both mathematical models and human intelligence, a reward is either expressed as the association of an emotion with something, or that emotion itself. Simply put, it enables learning, from other agents (humans and robots alike) or the environment. A famous example is the Braitenberg vehicle.

Emotional Intelligence enables Skill Acquisition

In Psychology, Symbolic-AI and Synthetic Intelligence, emotions play a role in reasoning, the development of logic and skill acquisition. Survival, Adaptation, and everything in between is attached (and often attributed) to emotions.

Emotions create friendly Technology

Quite literally, the hypothesis is that something that feels, will be kinder, more helpful, more forgiving and more human if you will. This is a different angle from our need to anthropomorphise objects. For example, a friendly AGI which makes decisions for say a spaceship, is more likely to make decisions based on a value-system which was derived from emotions, than one which uses reasoning that has no emotions attached to it.

Our need to Anthropomorphise technology

We create robots, androids and synthetically or artificially intelligent machines after ourselves, because we need to have the ability to connect on an emotional level. A machine, no matter how intelligent or capable, unable to express those attributes, will not be liked by humans, and as such, will not be a viable product in the long-term.

I am sure there are more reasons, some often debatable.

Answer 14

Nobody knows how to program a general artificial intelligence. What we can do instead is train a neural network to predict human behavior. The android's software is not programmed to think and form sentences and have any kind of dialog. It's programmed to predict what a human would do in its situation, and then do that.

It's like creating a self-driving car not from fist principles and optimization, but from observing lots of human drivers and training a machine learning model on these observations. This is the fastest way to a human-like driving experience.

The creators of the androids have of course tweaked things to achieve their goals. The cost function during training can include other factors besides matching what a human would do. You also reward satisfying human parties, for example, to steer the android toward a selfless, positive role.

You also clean your training data. No need for the model to learn when to throw a fit, when to cry, or when to fall in love. It is a huge data set though, and it's really hard to detect and cut out all emotions.

From the GPT-2 paper, on observing that the model is able to translate from French to English:

Performance on this task was surprising to us, since we deliberately removed non-English webpages from WebText as a filtering step. In order to confirm this, we ran a byte-level language detector on WebText which detected only 10MB of data in the French language which is approximately 500x smaller than the monolingual French corpus common in prior unsupervised machine translation research.

The same thing happens with emotions.

Answer 15

What intrinsic motivation for doing anything would an emotionless android have? It's basically the same question as "why give animals emotions?", including humans. Emotions provide an easily manipulateable hook into the mind of intelligent organisms: advertising, various political systems, religion and socioeconomics are all driven by emotional responses in order to get a large amount of people to behave in predictable, determinable and exploitable patterns.

Without intrinsic cohesive low-level normalizing mechanisms like emotions, you'd need to micromanage sentient beings to get them to work smartly in long-term reliable manners. Having to micromanage would defeat the whole point of making them intelligent in the first place.

Answer 16

This is a spin off Trevor's answer:

They don't really have emotions, we just think they do

Humans can't define emotions clearly. I mean "What's love?". Ask 10 different people and you get 10 different answers, as 100 different people and you get maybe 80 different answers.

What's the difference between fear and self-preservation. Between stubbornness and perseverance?

Two different people might see the same behavior and one thinks it's romantic, the other thinks it's creepy.

Hell, we do this to dogs. This CNN article shows that dogs don't really need puppy eyes, humans made them that way. It's fairly easier to assume why: dogs who had that trait we viewed as being sad (or hungry, or caring), so people would take better care of them.

Bottom line, androids don't REALLY have emotions, but we see them taking care of their owners and a few people will think there's love envolved.

Answer 17

It's not a bug... It's a feature!

Many of the other answers give many good reason as to why androids would built to emulate emotions. Being able to understand and express emotions would allow androids to be interact with humans and make decisions that humans can better predict and accept. However, none of this would require androids to actually feel emotions.

There are, however, a number of reasons why scientists and programmers might want to create an artificial intelligence who consciously experiences emotion; emotions are an integral part of the processes that govern their perception and behavior.

• Cool Factor - Some would give androids emotion for no other reason than because they are big Science Fiction nerds. ;)
• Authenticity - No matter how realistic an android's behavior is, some may not respond positively unless they believe the android genuinely feels something for them.
• Challenge - Creating an AI that can feel would be an exceptional challenge that some would tackle for the gratification of achieving such a task and/or for the fame it would bring.
• Psychological Research - Androids would likely not be considered "living creatures" by most so experiments could be conducted on them that would be considered ethical violations if done with humans. The fact that androids' emotional state could be monitored and recorded directly would be an added bonus. However, to be of benefit, these android emotions would have to be a reasonable replication of human ones.

...but why would all androids feel?

Given the above, the real question is "Why would all (or even most) androids be given the ability to feel emotions and not just a few?"

Perhaps in your world, the primary operating system used for androids happens to have emotions baked into it. This need not be because the emotions convey a unique advantage; it could simply be a coincidence.

Imagine this: while the groundwork for general artificial intelligence was being laid, pioneers began an open source project to create an general AI framework. This project attracted those who wanted to create feeling AI for the reasons stated above. The project also, however, led to the creation of the most practical framework for robotic industries to leverage. It was free, easy to learn and use, compatible with all the popular technologies (e.g. tools, programming languages, hardware), well documented and understood, and relatively bug free. It wasn't necessarily the best in all regards, but it was the one most companies adopted.

The companies leveraging this framework for their androids wouldn't necessarily believe that the androids would actually feel or understand it well enough to verify the claim that they did. Instead, AI emotions would likely be dismissed as mere exaggerations or wishful thinking. This would explain why they would sometimes act illogically. The companies leveraging the framework wouldn't take the emotions into account and thus would do things such as build a soldier android whose fear function was still enabled.

With enough misuse, the original developers of AI emotions would likely become activists. Seeing beings that they know have genuine emotions subjugated to all manner of emotional distress or abuse by the ignorant masses and greedy corporate leaders would be horrifying. They would ally with those androids who are resisting as well as young people more willing to accept androids as peers than the older generations accustomed to androids being simple robots.

TL;DR - Some people would give androids emotions just for the sake of doing it or because of some special niche need. However, once built, it's plausible that the technology would be used (and abused) by those who didn't recognize the truth.

Answer 18

Another possible feature is that emotions are a debugging feature.

I am reminded of the anime A Certain Scientific Railgun, particularly of the Sisters project arc. While not androids (the 'sisters' are clones) they are "spun up" in a particularly short time-frame, and lack the ability to express emotions in a natural way. This causes problems, as controlling what they do is very difficult if you get no insight into their decision making process, and so a hack in their boostrap process is introduced, where they narrate all their emotions in third person.

As a programmer, this narration sounds very similar to system logs, logging at the Information/Trace level. As it progresses though, this narration demonstrates that the sisters are developing self awareness, desires, and complex drives.

In the show, the engineer behind the 'hack' realises that the narration a particular sister is giving was showing a more human response to stimulus than she, the engineer, was expecting from them.

---

One could take this idea further, and design a new "interface" for the system log that exhibits more natural body language and verbal intonation to express the log rather than narration, but the underlying reason the androids have emotions in the first place was to identify what the android was feeling at a given time.