Sunday, July 08, 2007

Could a robot have feelings?

One of the great pleasures of giving public lectures is the questions that come from the audience, and my talk last week to the Bath U3A group was no different. A great question (and one I've been challenged with before) was "could a robot have feelings - emotions like fear, sadness or love?"

My questioner (and I guess most people) would be hoping for a comforting answer: No, only people and maybe some animals (your dog) can have feelings. In one sense that answer would be correct, because if we define feelings simply and subjectively as 'the feelings we experience when we are afraid, sad, or in love', such a definition would make those emotions uniquely human and something robots could not have by definition.

For the sake of argument let's get over that by asking a slightly different question: could a robot have artificial emotions that - in some way - are an analogue of human feelings. At this point I would normally say to my questioner, well, yes, I can conceive of a robot that behaves as if it is experiencing emotions. A humanoid robot, in other words, that acts is if it were afraid, or sad. The better the technology, then the more convincing and subtle would the (artificial) body language be. Actually we don't have to try very hard to imagine such a robot - colleagues in the lab are working on robot (heads) with artificial empathy. Here a picture.

At this point my questioner says "ah, but that robot doesn't really have feelings. It's just pretending to have feelings".

True! But don't human beings do that all the time. Isn't it the case, in fact, that we value the ability to be able to pretend to express feelings very highly, providing it's the honest deception of the actor's trade? We also value people who can make us feel in different ways: musicians, artists, poets or storytellers. Ok, I admit I'm being a little tricky here, but the point I'm trying to make is that humans can be very good at pretending to have feelings or at deceiving others into having feelings, and some (me included) have a hard time telling the difference.

Back to robots though. Yes, a robot that is programmed to behave as if it has feelings cannot, I grant, be said to actually have feelings. At best that robot is a good actor. This might seem to be the end of the issue but it's not, because the possibility of a robot that is very good at behaving as if it has feelings raises some pretty interesting issues. Why? Because in human robot interaction such a robot could have quite some power over the human. What interests me here is how such a robot would make humans feel towards it. Could a human, for instance, fall in love with an deeply (albeit artificially) empathic robot? I suspect the answer is yes.

This is a subject I'll return to in future blogs: not only the social implications of robots with artificial feelings, but also the question of whether robots could be designed that really do have feelings.


  1. "What interests me here is how such a robot would make humans feel towards it. Could a human, for instance, fall in love with an deeply (albeit artificially) empathic robot? I suspect the answer is yes."

    A very interesting post you have there but i'm just going to quote this part. It really ponders my imagination. How about the other way around where robots have affections of love towards a human being over time?

  2. We do have feelings, why not robots as well?

    Think about us. We are a set of muscles, bones... we have a cell, a heart, that pumps forever until our death.
    In the other hand we have created robots that can do many things if we program them...

    If we put conditions for orders, why not for feelings? It seems easy... and it is :p

  3. I would say that humans arnt much more that robots.Our brains being extreamly high tech computers that can actually be hooked up to computer.theres is a man with no power in his limbs but is able to move a mouse on a screen with the power of his mind because he is hooked up to the controls by wires.
    i feel like the key emotion artifical inteligence would have to experiance to be considered real emotion is order to be consider real emotions it would have to question its programing and alter it on its own by feeling shame for what it has done there.

  4. We are conscious beings. We have five material senses plus the mind. We can see, hear, smell, touch, taste & think, feel and will, because we are conscious. We have to go to the root of consciousness (Can modern science really help?). Where does this consciousness come from? We have to get answers to this question first.

    Second, is a robot conscious or can a robot actually ever be conscious? We have to get answers to these questions.
    Suppose I and a robot are given the same book to read. What happens when a person reads a book? When a person reads, he becomes aware of various thoughts and ideas corresponding to higher-order abstract properties of the arrangement of ink on the pages. Yet none of these abstract properties actually exists in the book itself, nor would we imagine that the book is conscious of what it records. I may find the content interesting/boring/thrilling/amusing/horrifying whatever... I may enjoy reading or I may dislike what I read. Will a robot (equipped with extremely sophisticated cameras, sensors etc.) who scans each and every letter, word, or sentence ever be able to experience the book the way i did?

    Another example: can robots smell? (in response to "Publunch would like to know a little bit about artificial olfaction, i.e. robots that can smell. Any good leads? Thanks --Publunch 16:50, 8 January 2006 (UTC)" (Wikipedia talk))

    What does one mean by artificial olfaction? First, just think about the sensation of smell. There are good odours (rose, sandalwood, camphor, jasmine, musk etc.), bad odours (hydrogen sulphide, rotting flesh, skunk spray), some are in between and there are many other types. The nature of these smells may be attributed to the chemicals, but what about the sensation of smell itself. We cannot describe any particular smell (the actual perception of the smell) using any amount of descriptive words, pictures or sound or anything else.

    Suppose I wish to describe the smell of sandalwood paste (paste made by grinding the wood of the sandalwood tree with water), using words to someone who has never smelt it. I may tell you the exact chemical composition of the paste, I may be able to describe the response of various sensory organs, neurons and how the brain recognizes it etc. in elaborate scientific language or at most I may compare the smell to some substance that you already know: but what about the experience of the odor itself? The odor has to be personally experienced to know how exactly sandalwood paste smells. Now when we smell something, we say "I smell", or "I like this smell" etc. I may find a particular smell "pleasant", "not so pleasant" or "unpleasant"

    When we smell something, who exactly is the perceiver of the smell, who actually experiences (feel the exact sensation) this smell (that is "This is rose", "This is phenol" etc.)? Is it the nose? is it the olfactory epithelium? olfactory sensory neurons? The brain? The mind (what is the mind in the first place?)? Our nose, olfactory epithelium, brain etc. are all made of organic matter. Can matter experience ("feel") smells?

    If a very complicated robot were to "smell" a substance, it may be able to determine the exact chemical composition of the substance down to the last molecule. But is it possible for a robot, however advanced it may be, to actually experience odours as human beings (or animals) do. Can the robot experience smells as "pleasant" or "unpleasant"? (Of course the terms "pleasant" and "unpleasant" are relative. What one person finds pleasant may not be pleasant to another.)

    Modern science has not been able to get to the root of consciousness. There should be serious research about the origin of consciousness before assuming that one day ultimately robots can become conscious.

  5. Thank you Anon for your very interesting comment.

    You ask some very deep questions about the nature of consciousness and in this short response I could not possibly do you questions justice. (But I am working on a new blog post about robot consciousness.)

    However I will say this. If we were for the sake of argument to accept that robots could experience subjective consciousness - in other words have some degree of self-awareness (recognition of the difference between me and not-me) - then I think it will be nothing at all like what we humans experience. Thus, if a robot could experience some artificial emotional response to smells, I think it's safe to say that it would be quite different to our experience.

    I think imagining 'what it is like to be a robot' (to paraphrase Daniel Dennett) is even more impossible that imagining what it is like to be a cat, or a bat, or a fish.

    Thank you again for your great comments.

  6. Here's a great talk by an eminent MIT scientist who works on "affective computing," Ros Picard.

    She spoke on "Robots, Autism, and God" at The Veritas Forum at Rice.

    Here's a short clip called, "Can Robots Have Feelings?"

  7. Rosalind Picard is an eminent scientist at MIT who gave a talk at Rice entitled, "Robots, Autism, and God." Pretty cool stuff.

    Here's a clip on whether or not Robots can have feelings.

  8. I do not know how old this is but I would like to bring up the idea that maybe there is such a thing as robot emotions. Humans have their own set of emotions. Robots could easily have their own emotions that humans could not possible understand.