11

According to the 'Mario Lives!' video, researchers have been able to develop an AI unit that is able to experience emotional states, such as greed, hunger, and curiosity. If the AI is currently experiencing an emotion, it will engage in certain behaviors. For example, if Mario is very greedy, Mario will look for coins. This approach is very similar to the Sims video game series, where emotional states are also represented, and AI characters will attempt to fulfill those drives.

Now, I can concede that the Mario AI is effective at simulating emotions. The media, however, has gone beyond this. Two articles, in particular, troubles me.

I can accept the idea of Mario 'learning'. I can even accept the idea that Mario is 'self-aware' within his own environment. But I cannot intuitively accept the idea that Mario is feeling any emotions. It seems weird to think that EA programmers and German AI researchers were both able to accomplish the dreams of science fiction writers all over the world, and without anybody ever noticing.

Maybe my intuition is wrong though, and maybe Mario is indeed feeling emotions (even if they are not the same emotions as humans feel).

3
  • 2
    To me, if you are asking "Is X actually a Y?" you need to look into more detail about what criteria you would apply do decide the question. The phrasing in the title question is "equivalent to actually experiencing" -- what aspects of behavior and conition are allowed to differ and which aspects are required to be constant?
    – Dave
    Jun 6, 2016 at 13:46
  • The negation follows directly from the Chinese Room experiment. Jun 7, 2016 at 1:17
  • I made a painting of a large fig tree and I'm very confident that the leaves or undergoing photosynthesis. Let's hope no one ring-bark's it as I would hate to see it die. Jun 7, 2016 at 23:06

8 Answers 8

9

Different schools of thought within the philosophy of mind would answer your question differently. I will try to describe the answer that each position implies, based on the information from the two links you provided. The various positions with regards to the mind-body problem are richer and more diverse than the 4 I describe below (In each case I refer you to the SEP article on the topic - you can find further references there), but I believe these 4 basic positions are enough to cover your question:

  • Behaviorism is based on the idea that a scientific theory can only describe observable phenomena, and we can never observe emotions and beliefs directly, we can only observe behavior. So all mental and emotional states are reducible to behavior, and to say anything more about them is to make unscientific statements. From this point of view the "Mario Lives!" program you describe does experience emotions, because it behaves according to those emotions, and there is nothing more to emotions other than the corresponding behavior. Behaviorism was popular in the 1930s and 1940s but has since falling out of favor because of a serious challenge: It fails to properly explain situations where someone has an emotion but is hiding it, and conversely where the same behavior can be explained by more than one emotion (is that person crying tears of sadness or tears of joy?).
  • Functionalism appeared in the 1950s and attempts to solve the problem that behaviorism faces as mentioned above. Functionalism, associates mental states with internal states similar to the internal states of a computer or a Turing machine. These functional states might or might not lead to specific behaviors, but they are in a causal relationship with sensory inputs and outputs and with other functional states. Thus learning that a loved one has died will lead to a functional state of sadness, and functional state of sadness can eventually lead to a functional state of depression or to certain beliefs, etc...A functionalist will agree that the "Mario lives!" program does experience emotions, provided its designers equipped it with the proper functional states.
  • Type-Identity Physicalism or Type Physicalism makes the very strong claim that emotions and beliefs correspond to exactly the physical brain states of the person experiencing them, based on the idea that only the physical/material world has any concrete existence. The usual example is that of the sensation of pain: Pain is exactly the firing of the specific C-fiber nervous tissue in the brain and nothing else. Therefore "Mario Lives!" cannot experience emotions, because it doesn't have the corresponding neural structures that these emotions are identified with. Type-idendtity physicalism is not very popular these days, because it is considered too strong. Per this approach, even other living beings which have different neural structure than that of humans cannot experience pain, since the definition of pain is such a specific one. See the question of multiple realizability.

  • Dualism is the position that mental states are of a different nature altogether than physical/material states (possibly a different substance or part of a different realm). DesCartes was one of the first to provide a serious philosophical argument for (Substance) Dualism with his famous "I think, therefore I am". Although DesCartes version of the argument is no longer popular, there are modern variations of it, such as Saul Kripke's and Thomas Nagel's, which have received considerable attention. Per a dualist, the "Mario Lives!" program cannot experience emotions, since it lacks the specific properties or substances that make something mental as opposed to physical.

Going through the article and looking at the youtube video provided in the OP, it seems that the authors are using purely behavioral criteria to make the claim that their program experiences emotions. In particular, this line stands out:

Fabian Schrodt, one of the researchers involved in the project, tells The Verge that as far as artificial intelligence goes, "Mario Lives!" is comparable to a regular video game opponent. However, he says, it's the combination of AI programming that learns and adapts as well as the addition of principles from psychology that's interesting.

Based on this statement from the authors, I would say that contemporary philosophers of mind would say that "Mario Lives!" doesn't experience emotions, since behaviorism has been pretty much refuted as a solution to the mind-body problem.

In your title question you use the word "experience emotions". The way I read it is: Does the "Mario Live!" program have subjective first person experience. Closely related to the question of first person experience, and a topic that might help you more with your research, are the questions of Qualia and the hard problem of consciousness.

2
  • 1
    Behaviorism has nothing to do with philosophy. Associating science with philosophy is oxymoronic.
    – D J Sims
    Jun 9, 2016 at 10:43
  • 3
    @DJSims There are two types of behaviorism: Behaviorism in psychology and behaviorism in philosophy (see the link in my answer). Philosophical behaviorism's main inspiration was logical positivism, who truly thought that all philosophy should be reduced to scientific (i.e. experimentally verifiable) propositions and those parts of philosophy that couldn't conform to this criteria were literally nonsense. Behaviorism was just applying that idea to phil.specifically. Both Behaviorism and L.P have been throughly refuted since then. Jun 9, 2016 at 17:48
2

It's a matter of amount and quality of conscious experience. The simulation may have some degree of sentience, but it may be many orders of magnitude lower than what humans are able to achieve.

Fundamentally, any claim about consciousness is untestable, and it is a moot point. But it is theoretically possible to imitate human qualia so exactly in a computer that an outside observer cannot tell the difference. The example you gave is just a very minor step in this direction.

2

In Germany the psychologist Dietrich Doerner some decades ago has set out to incorporate emotions and feelings into artificial beings. See

  • Doerner, Dietrich: Bauplan fuer eine Seele. (1998) (In German) and
  • Doerner, Dietrich et al.: Die Mechanik des Seelenwages (2002) (In German)

Doerner has also developed a corresponding computer program named PSI, see https://en.wikipedia.org/wiki/Psi-Theory

Some people consider Doerners work an approach to naturalize the soul (in the sense of psychology.)

On the opposite, the philosopher Searle does not tire to insist "Simulation is not duplication."

2

From a computational standpoint, emotions represent global state that influences a lot of other processing. Hormones etc. are basically just implementation. A sentient or sapient computer certainly could experience emotions, if it was structured in such a way as to have such global states affecting its thinking.

Simulating emotions would apply to simpler systems, say something like a chatbot where "current emotion" is just a register somewhere that switches to a different bank of responses.

2

I can accept the idea of Mario 'learning'. I can even accept the idea that Mario is 'self-aware' within his own environment. But I cannot intuitively accept the idea that Mario is feeling any emotions.

Correct me if I am wrong, but you seem to be getting at the question of whether or not Mario has any first-person subjective experiences at all. The computer program that Mario is a part of can learn to say "Ouch!" in response to getting poked with a stick, but does the computer program actually experience anything? What are these first-person subjective experiences that we call feelings, anyway?

@Alexander S. King's answer hints at this at the end, where he mentions the concept of qualia, and I thought I'd expand on that just a little, since I think it lies at the heart of the issue.

To quote from the SEP article on qualia that I linked to above:

Feelings and experiences vary widely. For example, I run my fingers over sandpaper, smell a skunk, feel a sharp pain in my finger, seem to see bright purple, become extremely angry. In each of these cases, I am the subject of a mental state with a very distinctive subjective character. There is something it is like for me to undergo each state, some phenomenology that it has. Philosophers often use the term ‘qualia’ (singular ‘quale’) to refer to the introspectively accessible, phenomenal aspects of our mental lives.

The statement "Computer Scientists Generate A Self-Aware Mario That Can Learn And Feel" on its face implies that Mario experiences qualia. The concept of "feeling" seems utterly meaningless unless it includes some kind of subjective, first-person experience that would qualify as a quale. That Mario actually experiences qualia is a very strong claim - I'm not even sure that the scientists who created Mario would claim that their computer program actually undergoes any kind of a first-person, subjective experience. When Mario feels sad, what does it feel like? Is that question even intelligible?

The role that qualia play in consciousness is hotly debated. Two of the most famous papers are Thomas Nagel's, “What is it like to be a Bat?” and Frank Jackson's "What Mary Didn't Know". In different ways, both papers argue that qualia demonstrate that consciousness can't be reduced to purely physical processes, because even a total knowledge of all the physical processes involved in producing a conscious state would leave something out - namely, "what it is like" to be in that mental state.

Daniel Dennett's "Quining Qualia" takes an altogether different tack, arguing that qualia aren't really special in the way that we assume that they are, and that in fact the concept of qualia itself is really just a product of folk psychology that should be done away with altogether. "Far better, tactically," he writes, "to declare that there simply are no qualia at all."

But regardless, it is impossible to address this question of whether or not Mario really "feels" anything, without first taking a look at what it means to feel at all, and the role that these first-person subjective experiences or feelings play in consciousness.

In the Mario example, I don't see any evidence that the computer program actually experiences qualia, and I think the idea of "feelings" is incoherent without qualia.

2
  • 1
    How would you distinguish a "subjective, first person experience" from, say, setting the "sad" boolean in a computer program to true (and then having the program do actions relating to sadness)? Though, I guess, a functionalist would argue that they're essentially the same thing. Jul 17, 2016 at 14:41
  • 1
    To me, that is like asking how to distinguish a rain storm from setting the "isRaining" boolean to true in a computer program that simulates weather systems. I would not suddenly feel the need to place towels on the floor under the computer to prevent water damage in the lab :) Jul 17, 2016 at 15:07
1

Dave H. wrote:

From a computational standpoint, emotions represent global state that influences a lot of other processing. Hormones etc. are basically just implementation. A sentient or sapient computer certainly could experience emotions, if it was structured in such a way as to have such global states affecting its thinking.

This description of emotions gives a hint of the nature of emotions, and why they are needed. Emotions are global states, which are normally communicated to all involved parts of the body by hormones. They are needed to get the different parts onto the same page. I wouldn't dismiss this global communication aspect as an implementation detail, because it is closely related to why emotions are needed in the first place. And emotions have meaning exactly because they are needed. Having meaning is one important aspect of emotions that should be kept by any proposed simulation of emotions.

For me, a jet aircraft with global states like "normal on ground operation", "starting", "normal in air operation", and "landing" is much closer related to a true simulation of emotions than that artificial Mario AI described in the question. Those different global states of the aircraft are needed, because the engines and wings operate at different working-points there, and use different calibrations for control and feedback in those working-points. So the global states are needed and do have meaning, and they affect everything (each part of the plane) and not just the central processing unit. And because of the different calibrations, the thinking (or behavior) of the central processing unit in each of these states is also different.

4
  • 1
    I find your answer completely obscure, as if you were telling a private joke that only you can understand. would it be possible for you to clarify it?
    – nir
    Jul 24, 2016 at 12:24
  • 1
    @nir Is it clearer now? Jul 24, 2016 at 13:43
  • so according to you, if that mario was not running around inside a virtual world, but was a robotic doll running around in a miniature nintendo playground, his would have been candidate for having real emotions?
    – nir
    Jul 24, 2016 at 15:53
  • @nir Even in the miniature nintendo playground, I still don't see the need for those fake emotions. Maybe I should ask instead whether the global states of a simulated aircraft could count as real emotions. Maybe, if the simulation is detailed enough. Jul 25, 2016 at 7:23
0

I can't tell you if Mario subjectively feels. I can tell you that emotions are not magic. Everything is data. Everything that exists or can be imagined is data. All of its attributes are data. That data can be simulated by a computer. The only difference between a properly done simulation and physical reality is the amount of resources needed, such as matter, space, energy, and time. Computer simulations will eventually be more efficient at resource utilization at equal to or better quality of life then is possible in physical reality. The big limit in physical reality is that magic isn't real,but in the virtual it is.

Tldr emotions can be simulated and subjectively experienced.

-1

I have considered much of the responses here, and I would suggest that most people here have missed the point when answering the question about emotions.

The problems is, scientists keep looking for a single solution as to what emotions are. This is akin to looking for a single shape that will fit all different shaped slots.

Also, what is ignored is that animals are just as capable of emotions and emotional states as we are:

When looking on Youtube for insects fighting each other, or competing or courting, it should be clear that simple creatures experience them too!

When I challenge people about emotions, I suggest to them to go to Corinthians 13 - which describes the attributes of love. If you consider all those attributes, one should notice that an actual "feeling" is not required for fulfilling any of them.

Therefore, the suggestion that a psychopath lacks emotions, and so he commits crimes or other pursuits outside of "normal" boundaries is far from true, especially when one considers the various records left to us from court cases and perhaps psychological evaluation - which show us that they do act out of "strong" emotions.

It should be considered that a psychopath's behaviour is motivated out of negative emotions and emotional states with a distinct lack of or disregard of morality and a disregard of conscience. Psychopaths "enjoy" what they do.

I am strongly suggesting to all that we are blinded by our reasoning, and by the reasoning of others.

Though I do agree with the following quote mentioned before: -

Dave H. wrote:

From a computational standpoint, emotions represent global state that influences a lot of other processing. Hormones etc. are basically just implementation. A sentient or sapient computer certainly could experience emotions, if it was structured in such a way as to have such global states affecting its thinking.

However, his reasoning below it (that quote) is also seriously flawed.

Emotions are both active and passive: They are triggered by thoughts and they trigger our thoughts; Emotions are a mental state and a behaviourial quality; Emotions react to stimuli or measure our responses to them; Emotions are independant regulators and moderators; Yet they provoke our focus and attention to specific criteria; and they help us when intuition and emotion agree or they hinder us when conscience or will clash.

A computer has the same potential as us to feel emotions, but the skill of implementing emotions is much more sophisticated than the one solution fits all answer people are seeking here.

Also, if anyone argues that emotions are simply "states" where a response or responses can be designed around it, really does not understand the complexity of emotions; the "freedom" emotions and thoughts have independently of each other; or what constitutes true thought!

Programmers and scientists are notorious for "simulating" the real experiences of emotions or intelligence, without understanding the intimate complexities; Thinking that in finding the perfect simulation they have "discovered" the real experience.

The Psi-theory seems to adequately give a proper understanding of the matter: https://en.wikipedia.org/wiki/Psi-theory

So I would say that the simulation of emotional states "is" equivalent to experiencing emotions, but those emotional states are far more complex than what most realise.

1
  • 1
    First off, welcome to philosophy.SE. There's quite a bit going on in this answer, and clearly it reflects your views on the question, but there's not much related to philosophy here. Our goal on the SE is to answer people's questions within the domain of philosophy roughly understood around the academic discipline.
    – virmaior
    Nov 24, 2016 at 1:22

You must log in to answer this question.

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