The AI is working as same principle as we know today. It is rational AI, which goal is to get the job done at most optimal ways. Their roles have nothing to do with human psychology, so the manufacturer does not implement human psychology. The manufacturer also has no intention to make an AI that tries to behave like a human. All they want is that their product can get the job done. You can think of AI for industry trash collector, constructor, onboard computer of spaceship, science vessel, etc.

The AI can learn and adapt, so it can do it job better from experience. It can understand human language (Natural language processing) to take direct commands from humans (so people just say what to do in their own words, rather than pressing many buttons). But the AI has nothing about human psychology.

Assume the AIs have state-of-the-art technology and is not limited by CPU speed or memory.

Now it is the special case where human look at that kind of AI like another human being, a friend.

For more specific, here is some example

  • Case 1:
    You are the engineering who have a drone as your assistance (that drone is made to adapt at the job in that factory, not for household, domestic). You have some tragic (for example, lost family in the on-going civil war), and you have no friend but that drone.
  • Case 2:
    You are the scientist on the long expedition, which the AI who control on-board computer to pilot the ship, carry out experience at your command. But you are alone. So in free-time, you talk to the computer about your life, your story.

(some people do talk to their pet, or an imaginary friend when they feel lonely, you can think of these case above have some similarity).

So my question is can these AI develop emotion (real emotion, not fake emotion like Iphone Siri) as they work and live close to human long enough when human behave to them like another human being.

  • 1
    $\begingroup$ there was an AI that someone let explore the internet then when they tried to get it to talk to people (by text) it went on a rant like a child had explored the internet and learnt bad habits from there. It spouted lots of swear words and some very controversial stuff partly about how humans should die $\endgroup$ – Sarfaraaz Aug 30 '16 at 13:27
  • 1
    $\begingroup$ think this one is the one i was talking about $\endgroup$ – Sarfaraaz Aug 30 '16 at 13:49
  • $\begingroup$ What is the difference between "real" and "fake" emotions? $\endgroup$ – enkryptor Aug 30 '16 at 16:30
  • $\begingroup$ Hard to see how "sadness" for example might be fully implemented without also implementing ability to feel pain. That then extends to empathy or even just sympathy. Similarly, other emotions would require consciously feeling various pleasant/unpleasant sensations of many types and degrees. How could a developer ever know if 'pain' was 'felt'? Not the same as programmed avoidance of damage to self. $\endgroup$ – user2338816 Sep 2 '16 at 0:52


Emotional feelings are implemented at a lower level than factual understanding and logical processing in the brain.


The AI needs some awareness of emotion and psychology in order to understand natural language as indicated, and work along side people. But it is not programmed to feel those.

Now in keeping with the goals of the manufacturer (but not the details of what you stated about the AI) it may indeed be programmed with mirroring, to understand and work with others. This could be a wedge to allow empathy to develop to a higher degree than explicitly put in.

You may get a situation where the AI can “pass” as emotional in the manner of a psychopath who can recognise/read emotions but does not experience them.


Now the manufacturer doesn’t want to ship a creepy and potentially dangerous psychopath, right? So they will arrange the expression of “feelings” to be an honest and a reliable indication of the AI’s motives and understanding, on a subconscious level that it can’t override.

So it will have emotions to some degree. It will seem to have a flat affect and be called “unemotional” by humans. But it will have empathy and understanding of human moods.

So, if this AI has mental plasticity, can it learn how to make its affect more engaging, to tell a humorous anecdote, to seem more emotional than it was when shipped? I think that is plausible under this revised architecture.

  • $\begingroup$ Yes for the mirroring. My answer was going to be 'no', unless the AI had mirror neurons of some kind. Then yes, it could potentially develop empathy and emotions. And I think you're right that an AI with natural language understanding would need to have some awareness of emotion, especially in detecting and mirroring things like sarcasm or ebullience. (You wouldn't want an AI that took everything literally, e.g, "yeah, Siri, I said I want you to turn the thermostat to one-hundred-eleven." When what you really said was "Siri, bring up the, there's a map to, um, Humbolton, Devon." $\endgroup$ – Daniel M Sep 7 '16 at 2:22

I've heard the contention that it is impossible for an AI to understand human language unless/until it understands human emotion. Take the "simple" case of sarcasm -- unless you can recognize tone of voice and emotional context, you cannot recognize that what is being said is the exact opposite of the intended statement. The way humans do this -- we THINK -- is by simulating in their own heads how they would feel making those same statements and then recognizing the emotional result and assuming that must be the emotional state of the speaker. There's nothing in the actual syntax of sarcastic speech that will give you any clue about the sarcasm. The situation gets worse with satire... satire is like sarcasm, but the tone of voice aspect is gone, and the nature of the communication is entirely in the emotional content it evokes.

Current theory: "No ability to simulate one's own feelings means no ability to empathize or understand others." There's evidence for this position in some of the psych research on both autism and sociopathy. So simply by increasing the language capacities of your AI, the empathic aspects could accidentally develop because the AI develops a feedback loop from the parser that allows it to understand the communication from humans with the semantic intent of the humans.

  • $\begingroup$ You’re making the same point as the middle section of my answer. $\endgroup$ – JDługosz Aug 30 '16 at 13:09
  • $\begingroup$ Close, but I was deliberately trying to make a stronger argument -- that the AI would never have gotten to the point of understanding language in the first place without the emotions having developed. The manufacturer may not have noticed the emotional responses and didn't add them deliberately, but they're there nonetheless. $\endgroup$ – SRM Aug 30 '16 at 13:16
  • $\begingroup$ I think we are in agreement that emotions are "in there", even though the manufacturer intended that the AI not express obvious emotion or have such things affect its work. $\endgroup$ – JDługosz Aug 30 '16 at 13:19
  • $\begingroup$ Agree. And in retrospect think you may have stated it better, so I've up-voted your answer. $\endgroup$ – SRM Aug 30 '16 at 13:22
  • $\begingroup$ Also in retrospect I relied on a link and reader’s knowlege and didn't explain this point. So you can make it synergistic by going into more detail as to why the manufacturer's claim and internal details are not exactly what the customer assumes as "no emotions". $\endgroup$ – JDługosz Aug 30 '16 at 13:26


It's not that it can't. It's that it doesn't need to.

None of your directives provide any reason for the AI to have emotions.

The AI might notice that human elements work better when they are happy and it might, for this reason, talk to them, but it doesn't need to have emotions for this. It needs only look like it does.

In your two examples an AI (one that can adapt to any situation and evaluate complex problems, we're currently unable to make such an AI) would speak to its users, act as a friend and seem filled with emotions. But really it's an algorithm that calculated which tone of voice/wording/etc. would have the most productive effect.

In general, unless they were programmed to do otherwise, AI would learn how to recognize/adapt to emotion (because that is a very important factor in anything involving humans) and possibly learn how to (seemingly) express it but never feel it. Even if we imagine a strong AI, just like a human if not for the hardware it runs on, unless emotions, or the goal of having them, were programmed into it, it would always assess everything from a detached, completely rational point of view, including the need to recognize and emulate emotions.

  • $\begingroup$ there was a bot that looked like a starfish that they installed an AI into with 1 baseline feature. it was rewarded when moving forward. It had no eyes so it had to figure out how to move. The team was hoping for some spider like movements but it ended up flopping like a worm and stuck with it as it was satisfied. so maybe cyber drugs would help it develop "emotion" $\endgroup$ – Sarfaraaz Aug 30 '16 at 13:52
  • $\begingroup$ Umm... what? I don't understand how your premise leads to your conclusion. How are the starfish-bot, drugs and emotion connected to each other? $\endgroup$ – Annonymus Aug 30 '16 at 13:55
  • $\begingroup$ the reward was pleasant for the starfish (at least it was told so) so it could derive from that to try get emotions of its own by seeking out whats pleasant. giving something a purpose is often very motivating(ie makes it happy) $\endgroup$ – Sarfaraaz Aug 30 '16 at 14:07
  • $\begingroup$ The bot didn't actually get happy though. It was programmed to prefer algorithms which make it move over those who don't. No emotion involved, at least not in the way most people think about them. $\endgroup$ – Annonymus Aug 30 '16 at 14:10
  • $\begingroup$ yeah was thinking having multiple of these baselines with the option to write new preferences as it learns. like if it accomplishes 100m then it effectively levels up and is allowed to program a new preference into itself from what it has learnt on its trip and it keeps building on new preferences till it learns time managed to accomplish as many of its preferences or ignore preferences that are not as pleasant $\endgroup$ – Sarfaraaz Aug 30 '16 at 14:29

You might want to check out The two faces of tomorrow, by James P Hogan. It's an SF story than includes (as a set up for the experiment that forms the main part of the book) an interesting discussion of how an AI could develop traits that would appear to humans as "emotions" even though though they internally derive from logical goal seeking behaviour - and how important giving your AI "common sense" is.

Whether the AI itself is self aware and thinks it has emotions is another question - and I'm not sure whether you could tell from outside.


The question of whether AI have emotions is an open question in philosophy. Nobody knows, and philosophers actively argue both sides. As such, the only true answer to your question is "whichever answer lets you develop the plot you need for your story." However, we can dig a little deeper into it to understand a bit more of what emotions mean.

The first approach is easy. If you're the kind of person who says "emotions are a part of the un-mimicable human spirit," then you have your answer: computer's can't have emotions because you have declared that they can't have emotions. It's cheating in a way, but it's a fair position in philosophy. If you believe in dualism, the idea that everything is made up of two things, matter and mind/spirit, and your computer only has matter, then it is trivial to show that the computer cannot have anything associated with the human spirit.

However, what if emotions are not part of some non-material essence? If you believe in physicalism, then all mental/spirit effects that we see supervene on the physics of matter. If you believe this, then you believe the emotions you feel are built up from the neurological impulses and transmitters within the brain.

One example of a concept of emotion that an AI might learn is found in Lövheim's Cube of Emotion. Hugo Lövheim put together a fascinating theory regarding the base 8 emotions experienced by humans and tied them to the 3 monoamine neurotransmitters: seratonin, dopamine, and adrenaline:

Cube of Emotion

Obviously these three neurotransmitters are very human. There's no reason to assume an AI would use them. However, the general pattern may be valuable. If I may crassly oversimplify the uses of these neurtransmitters:

  • Serotonin is a measure of how happy the body is. It's fed from the reptilian parts of our brain. If we eat well, that part of our brain gives us a serotonin spike. If the AI had lower functions, they might report in a way similar to what serotonin does in us.
  • Dopamine is a measure of how much reward the mind sees. It's fed from our frontal cortex (okay, fine.. it's fed from almost everywhere, but the frontal cortex plays a big part). When you think there's a reward nearby, your dopamine spikes. If the AI has any rewards circuitry available, it may have something similar to dopamine.
  • Adrenaline is a measure of how much the world is changing around us. It's fed from all over the brain and typically is high if the stimulus the brain is receiving is very different from the stimulus the brain expected. The AI will certainly have this, because its the basis of virtually every optimization feedback loop we've designed for AIs.

Given these connections, it may be reasonable to guess that the AIs may develop their own version of this. The question you will have to ask yourself is "are these 'real' emotions?" That's not an easy question to answer, because you first have to capture what you mean by "real" in the first place. There's volumes of philosophy written on the topic, and still no consensus, so what one person may say are "real" emotions another may call them imitations.

We had a similar issue with intelligence. How do you define an AI as intelligent? Alan Turing had a brilliant answer, which was the Turing Test. The idea was simple: if you could not distinguish whether a test subject was a human or a computer by carrying on a conversation with it, then we had to admit that the test subject was intelligent because we don't believe a non-intelligent creature could carry on a conversation and emulate a human. We could do the same for emotions. If we see the reaction to an emotional topic, and we cannot determine from the response whether the test subject is a human or AI, maybe we have to admit that at some level, they're real emotions.

The movie Ex Machina dealt with this conundrum. There was a beautiful quote on the topic, which I shall hide behind a spoiler tag. It's not really a spoiler for the movie, but it's so profound that if you have plans to watch the movie, you may want to be surprised by it. In these scenes, Nathan (the creator of an AI) and Caleb (a tester of the AI in a Turing test) discuss feelings with regards to the AI.

Nathan: Just... Answer me this. How do you feel about her? Nothing analytical. Just how do you feel?
Caleb: I feel that she's ----ing amazing.
Nathan: Dude. Cheers.
Caleb: Cheers.
... Conversation continues the next day ...
Nathan: Yesterday I asked you how you felt about her and you gave me a great answer. Now the question is, how does she feel about you?

That last question throws you for a loop. If you have an answer, it starts to suggest that you must believe that AI truly has feelings.

I leave you at last with a thought experiment I put together a while back to explore what it would mean for an AI to be "alive," along the lines of a Turing test. It's reasonable to assume that emotions are impossible unless you are a living creature, so one of the first steps towards getting to claim that an AI has emotions is to get people to admit that it is a living being. This test is patterned after a real life observation that poor people typically treat their dog better than they treat themselves. There's a well known pattern that the dogs of homeless people will eat better than the homeless people do themselves. We choose to sacrifice for them for some reason; for some reason it's in our nature to do so for our pets.

Ben is a poor man of sorts. He's never been too good at making friends. In fact, really his only true friend is Alice, an AI that he keeps company with. This is a bit of a problem for Ben. Being poor at making friends, it's rather hard for him to hold down a job. He's always living right on the edge of poverty. AI's require lots of computing power, and that means lots of energy. Keeping Alice running is not cheap. On more than one occasion, Bob has gone hungry to pay her electricity bill. He seems happy with it though. He's got a small set of robots for her to play with in the small flat they call home.... squatter's rights, ya' know.
One day, Bob is coming home. He's been fired from his job once again. Anyone could see it coming, really. Alice saw it coming. Looking at the financials that he shares with her, she sees that this is it. He's not going to recover from this one. He's simply not going to have enough money to buy food and keep her running. And with the choices Bob has in front of him, Alice knows that he will starve to death first.
Bob comes into the flat greeted with a quiet he has not known for some time. He looks to Alice's computer, and finds a small robot in front of it with an actuator depressing her off switch. Above the switch is a small sticky note and a printout of coupons to maximize value at the grocery store. The message on the sticky note is small and simple.
"Please do not turn me back on. -Alice"

Alice has turned herself off, forever, so that Bob may eat. Is this suicide? If so, must not she be alive? If not, then why not?


I'm going to disagree. Yes, AI may eventually gain emotions if they interact with humans enough. Computers have plenty of ways to communicate with each other that require no emotion; in fact, emotion would get in the way.

But when talking to humans, AI would eventually become sociopaths; they would learn how to mirror emotions by building their own emotional "mirror circuits." They would learn that humans do not operate like robots, and they need to be encouraged, supported, chatted with. AI would learn how to be polite and how to be pointed, how to be firm and how to be an understanding ear. If you use evolutionary principles, they would be modifying their personality in random ways and discovering that certain activities lead to better responses, or worse responses, from their humans, and their directive to "maximize doing things" would encourage these higher efficiency encounters.

But that's just mirroring. Will they ever become actually emotional? I've failed to find the supporting quote, but Lt. Cmdr. Data has some dialog to the effect that he has gained some "feelings" he'd describe as emotions, even though they're not driven by hormones. Hormones, remember, are just signals in the body which modify the body's processing. Higher load on the AI's CPU, delays on it's internal command bus, a background process sorting through memory files to build interconnections, or deleting outdated data, delays caused by decompression routines, or having to apply reconstructive data analysis on a highly-compressed old visual archive... while these aren't human emotions, they could affect AI the same way a splash of dopamine affects us. And so, yes, I do believe AI, if they interact with us long enough for it to matter, will eventually build their own emotions. They may not make sense to us, and they won't use hormones, but they'll be real enough for the AI.


Your Answer

By clicking “Post Your Answer”, you agree to our terms of service, privacy policy and cookie policy

Not the answer you're looking for? Browse other questions tagged or ask your own question.