Could a sentient machine suffer?

Question

I was considering this closed question very intently, and I found that I'm not at all fluent in the idea of modern slavery. Many philosophers have spoken on slavery. On this forum, someone has already asked of Plato's ideas of slavery. It is well-known that Hobbes and Locke considered slavery and the state of nature in their development of their conceptions of social contract theory.

But what really got me thinking was the fact that even some classic 'Utopian' philosophers, such as Thomas More, thought slavery had a place in their world. More, for example, thought that temporary slavery could be used as a punishment (rather, that it was used on his island of Utopia). Then, sometime within the following century, inalienable human rights began to take hold. And there are plenty of writings on that topic, and I don't care to name them.

But we got to the next step. People like Bentham argued that animals should have inalienable rights too. His famous quote involved asking not whether things talk or reason, but rather can they suffer? Although this hasn't taken hold, it made me wonder. Could a sentient machine suffer? Supposing we have a human-level artificial intelligence, what would it mean for it to suffer?

Answer 1

On the one hand, one might naturally understand that to to be sentient means to have an inner mental life that is, in broad strokes, largely like our own. Since the capacity to suffer and feel pain is an important part of human mentality, one might expect that any animal or machine that is fully sentient would have the capacity to suffer. The ability to feel pain is surely a part of the full experience of our mentality.

But perhaps sentience does not require the full experience of our mentality, but only a part of it. In this case, I would separate your question from the machine aspect of your question, and reduce the issue merely to the question of whether the ability to suffer is a necessary condition for sentience.

Even in the case of human beings, this doesn't seem to be the case. The reason I believe so is that we can easily imagine a person that has suffered some brain damage to the parts of their brains controlling the pain response, so that they are no longer able to experience pain or suffer in any way. Of course, such a condition would be debilitating, since it is precisely our ability to feel pain that protects us from and alerts us to many dangers in the world (burning, cutting, other injuries, etc.). Indeed, our ability to feel pain is surely a critical component of human survival (and there seems similar good reason to think that the pain capacity would be critical for the success of a thinking machine). Nevertheless, I think we can also easily imagine that a person might have such a pain disability without losing their ability to reason, think, write, experience joy, love, appreciate music and so on, having in all other respects a full mental life despite their disability. And in such a case, I think we would say that such a person is still sentient, even if unable to suffer pain. If this is correct, then the ability to suffer would not seem to be a necessary condition for sentience.

In the case of a machine, if machines could be said to be sentient at all, even with pain, then if a particular machine is able to experience all those mental phenomena in a fully similar manner to our pain-disabled person, but be unable to suffer, then I think that we would also say that this machine would be sentient to the extent that we would say that any machine can be sentient.

Answer 2

First, let's suppose we have such a human-level artificial intelligence. Depending on the constitution of the machine, it may have 'transparent access' to its own cognitive structure. In such a situation, the machine would have a choice as to whether or not to experience the pain, or indeed attenuate the intensity experienced (by, say, 17%) before the pain-experiences even begin. Presumably posthumans and active mindfiles would have the same capabilities in terms of manipulating the intensity of their input sources.

Now, if the machine doesn't have this sort of access to its "mind," which seems likely at least in the earliest stages of development, it will be in roughly the same position with respect to suffering that modern humans are; i.e., its internal neurocognitive structure is effectively a 'black box' manipulating signals of its own accord, so the machine wouldn't have any opportunity to consciously interrupt the flow of pain-signals.

Finally, the question of why you would cause a machine intelligence to feel pain also seems worth investigating to me. After all, it would seem possible to design a machine intelligence without providing an equivalent of 'physical' pain -- so why would we go out of our way to provide this "benefit"? Note that among sentient machines are likely to be uploaded human minds, which presumably would prefer not to experience physical suffering again; the pain response would seem more like an unfortunate meatspace necessity we would be gratefully ridding ourselves of (as with aging and death.)

So yes, in general, I think a sentient machine could suffer -- but again, with the strange caveat that once a given stage of development has been reached, it could potentially have a choice in the presence, degree and quality of suffering experienced at any given moment. So could this really be called "suffering" any longer?

Answer 3

1. Well, in philosophy and in academia generally, the word 'sentient' refers to the low-level capacity to experience feelings like pain and pleasure. Thus (if I'm not mistaken) scientists recently confirmed that fish are sentient. And thus human fetuses acquire sentience at around 24 to 30 weeks into the pregnancy. But I take it you are using 'sentient' in the sci-fi sense of the word: i.e., the high-level capacity for human-like intelligence. Given the first sense of the word, the answer to "Could a sentient machine suffer?" is pretty simple: obviously yes, because the capacity to experience pain and pleasure is precisely what sentience is. But given the second sense of the word, the question is less simple, because it raises questions about the connection between intelligence and conscious experience, and the capacities of mere machines.

2. Even if X can think, it doesn't obviously follow that X can experience. And thus even if mere machines can think, it doesn't obviously follow that mere machines have conscious experiences. The dualist philosopher David Chalmers famously argues that conscious experience cannot be naturalized into mere physics: i.e., no mere arrangement of physical stuff is sufficient to explain conscious experience, not without additional phenomenal properties linked to physical stuff by additional psychophysical laws. But I'm pretty sure he's more optimistic about naturalizing cognition into mere physics: i.e., mere physical systems might be capable of thinking on their own. But even if he's right about this, it might nevertheless be that any sort of physical system capable of engaging in thought will also (thanks to the psychophysical laws that actually obtain) have conscious experience: i.e., even if it's logically possible to have a thinking machine that cannot suffer, it might be nomologically impossible (impossible given the laws that govern the actual world).

3. Even if suffering is of great moral significance, it doesn't follow that beings that can suffer have any rights. Bentham himself does not recognize animal rights, despite his view that their suffering matters as much as anybody's suffering. And even those friendlier to rights than Bentham commonly think that rights belong to a special category of morality that only rational agents like humans are eligible for. If this is right, then whether a sentient machine has rights would depend not just on its intelligence or its capacity for suffering, but on its capacity for rational decision-making and perhaps even on its capacity for free and autonomous decision-making of the sort Kantians champion.

Answer 4

Yes. And it's quite likely that pain and suffering would be a useful design decision in a sufficiently complicated (and resource-intensive) machine.

Issac Asimov's third law of robotics suggests that a sentient machine would need to know when it is being damaged or at risk of being damaged. Sensors that detect damage would be analogous to nerve-endings in humans and other animals (though likely based on different physical properties). Presumably signals from those sensors would be designed to interrupt the normal function of machine and demand to be addressed immediately. It's not clear that such signals would be experienced differently from what we call pain, if they are experienced at all. (What I mean is that what constitutes "experience" is difficult to pin down, so robots might not experience anything and therefore not experience pain either.)

Now good design would suggest that signals indicating damage of function would persist no longer than necessary to deal with the damage. While the damage exists, the signals should continue and when the damage is repaired, the signals should subside. Certain types of damage would be signaled with higher priority (or potentially greater intensity) than others as some types of damage would be more urgent and important than others. Some signals would be of very low priority so that they could be ignored while pursuing more important goals. A well-designed machine would even include senors that detect potential hazards such as high temperatures or high-speed collisions.

If a robot can be said to experience anything at all, it must also be able to experience a sensation we might call suffering. Failure to sense damage would be a serious problem for all but the most trivial sentient machines. I'm reminded of a book coauthored by Dr. Paul Wilson Brand, a pioneer in the treatment of leprosy or Hansen's disease. The book points out the sufferers of congenital analgesia are at great risk of physical damage from activities the pose little threat to normal people such as sleeping, walking or eating. Considering the difficulties faced by people who lack pain, we really aren't grateful enough for our occasional toothache or sore feet.

Answer 5

On the slavery issue:

We are all slaves to some extent. We are slaves to our bodies needs for food and water, as well as its functions that are less desirable including excretory and aging. The fact that these are not truly volentary that does not mean that they must cause suffering. In fact a good meal and fine drink are among one of the ways we choose to reward ourselves for achievements. Toilet humor is among the most common, and often the action is pleasurable if only for the relief it provides. In addition while I am not condoning the slavery of the past many slaves had luxurious existances. Indeed there is also a subset of humanity that chooses slavery as a lifestyle and the gratification that they derive from it.

On Suffering:

Suffering likewise is a state of mind. If you do not have something to contrast then suffering is just normal life. What is suffering is also different among people. Masochists enjoy pain and abuse. Many people choose to deprive them selves of breathable air for pleasure. In order for a machine to suffer then it must also know joy or at least a condition that is the lack of suffering. It must be able to make a decision what is pleasureable and what is not. In order to suffer it must be able to choose to suffer.

On what would be unethical:

Humanity is likely to determine what the machine finds unpleasureable. We will likely have a subset among us who would exploit that. Some that would derive some perverse pleasure in causing the machine pain and suffering. This would be as unethical as torturing an animal or another human. And causing the termination of such a being for no reason would equally be unethical. Such an action would be equivilent to murder.

It would also be unethical to deny an AI with the ability to distingish pain and sorry from joy and pleasure from experiencing both. This is not to say that it is your duty to inflict either. For Example, hiding a painful truth that the AI should know, even with the best of intentions, denies the AI the experience.

Answer 6

To turn the question on its head - if physical matter in the form of a computer can be sentient and suffer, then should we say any configuration of matter that is minimally Turing complete is potentially capable of suffering?

In which case, physical laws being Turing complete (because we can make computers) means the entire universe may be capable of suffering.

Answer 7

There is no doubt that a sufficiently sentient machine could suffer. By definition, being sentient is being able to experience sensations or feelings. Suffering is an unpleasant feeling, hence a sentient machine must in principle be able to experience suffering.