4
$\begingroup$

Nowadays, robots or artificial agents often only perform the specific task they have been programmed to do.

Will we be able to build an artificial intelligence that feels empathy, that understands the emotions and feelings of humans, and, based on that, act accordingly?

$\endgroup$
2
  • $\begingroup$ A very related question (if not a duplicate): Could an AI feel emotions?. $\endgroup$
    – nbro
    Apr 21, 2019 at 19:04
  • $\begingroup$ This could maybe be considered an extreme version of the value alignment problem. $\endgroup$ Apr 21, 2019 at 19:17

3 Answers 3

6
$\begingroup$

Let us describe a very simple system that does something we could label as empathyc.

A chatbot answers "I am sorry to hear that. What happened?" when we type "I feel bad", and it replies "I am glad to hear that. Fancy some music?" when we type "I feel good".

Somehow, it perceives a human emotion, and acts accordingly.

Planes fly but they do not fly as birds. Similarly, we can expect artificial empathy from an AI, not necessarily natural empathy as we feel it.

Here is a related paper, Shared Laughter Generation for Empathetic Spoken Language

And here is an actual example:

enter image description here

The system correctly maps tears to sadness and outputs a proper answer, while also explaining its non-sentience. Not bad!

$\endgroup$
1
  • $\begingroup$ Comments are not for extended discussion; this conversation has been moved to chat. $\endgroup$
    – nbro
    Mar 9, 2020 at 20:18
2
$\begingroup$

It seems to me that empathy is based on understanding the experience of another entity:

originally Psychology. The ability to understand and appreciate another person's feelings, experience, etc.
SOURCE: OED

Using this definition, the AI would have to understand human experience. (There may be a "Chinese Room" issue in terms of whether one considers the algorithm to truly "understand". But, if it can classify the input sufficiently to produce an appropriate response, that can constitute understanding.)

The underlying problem is that the algorithm likely doesn't "feel" in the same way humans experience emotions, in that the human experience is colored by chemical response. So while the algorithm might be able to demonstrate sufficient "understanding" of a human's experience, and act in an empathetic manner, the degree of understanding may always be limited.

$\endgroup$
1
  • $\begingroup$ Would you agree that the example I just provided classifies the input sufficiently and produces an appropriately aligned response? $\endgroup$ Mar 1 at 22:28
1
$\begingroup$

Empathy is about living and values, just as someone can be a good person, or a bad person, based on their education, culture, past experience and values, just as I who recently became a father became much more empathic about my parents.

If it's really a semi-programmed artificial intelligence that captures around you the experience with your programmers, with people, with the environment, etc., and somehow is set to learn from it, I believe you can develop some kind of “ empathy ”, but as Jaume said, it won't be“ natural ”

We have to remember that real artificial intelligence is the one that absorbs and learns, not a set of IF's and ELSE's.

$\endgroup$
2
  • $\begingroup$ I don't see a contradiction between learning and using a set of IF-ELSE to "classify the input properly" (from DukeZhou's answer), especially if conditions can change as more information is available. Were you thinking about predefined, static conditions? $\endgroup$ Sep 19, 2019 at 19:59
  • $\begingroup$ I also see no problem, after all for presets would have to have some IF-ELSE as input, just as you said. But I was thinking of setting empathy as something "fixed" to having a predefined "forced" empathy $\endgroup$ Sep 19, 2019 at 20:01

You must log in to answer this question.

Not the answer you're looking for? Browse other questions tagged .