One of the dividing lines I often hear in discussions about whether we should regard an artificially intelligent machine as a fellow being is, does it have the capacity to suffer? It’s an interesting criteria, since it implies that what’s important is that there be something there for us to empathize with. But it raises an interesting question. What would it mean for a machine to suffer?
First, what exactly is suffering? The Buddhists and Stoics thought this through millennia ago. Suffering is about desire, or more accurately, unsatisfiable desire. We desire scenario X, but are faced with scenario Y. When we can’t fulfill our desire to bring about X, the continued discrepancy leads to suffering.
For example, say I’m experiencing pain, electro-chemical signals from some part of my peripheral nervous system that my brain’s body model indicates could mean injury. I desire (intensely desire) not to be injured. In the short term, the discrepancy between the signals indicating injury and my desire for there to be no injury, motivates me to move in such a way as to avoid it. For instance, if I receive signals from my hand that my brain’s model interprets it as a burning sensation, I move my hand.
We might say that I’m suffering during the period between noticing the discrepancy and rectifying it, but that’s not what we generally mean by the word “suffer.” If after I move my hand, I continue to experience a burning sensation, and it lasts long enough to be considered chronic pain, then we’re getting to a situation that we might commonly consider to be suffering.
I’m receiving signals of damage, but I desire not to be damaged. Consequently, portions of my brain induce a state of wanting to correct the situation. But it can’t be corrected. Intellectually I might realize this, but the more primal levels of my brain do not. Despite the intellectual knowledge, I can’t just stop the urges and desires from the lower level aspects of my mind. Hence, I suffer.
The above example is for physical pain, but it applies to any scenario that is at variance with my primal desires. A friend gets hurt. I desire that my friend not be hurt. But it has happened and can’t be changed. Nonetheless I continue to intensely desire that my friend didn’t get hurt, leading to mental anguish, aka suffering.
Or perhaps I experience some severe loss of social status: a mate leaves, my reputation is damaged for some reason, something causes my friends and/or family to ostracize me, etc. All of these scenarios can be at variance with deeply held primal desires. Even if I know intellectually that they can’t be fixed, the emotional part of me won’t know, at least not for a period of time. During that period, we could call my experience of the discrepancy, “suffering.”
If you think about the stages of grief as they’re often described in the Kübler-Ross model: denial, followed by anger, bargaining, depression, and finally acceptance, it makes sense to see this as our emotional selves coming to accept something our intellectual selves might have understood very early on, perhaps our brainstem and limbic system coming to accept what our thalama-cortical system already understands.
I’m reminded of something a friend told me at the funeral of one of his children. When I asked if there was anything I could do for him, he replied, “No, there’s nothing that can be done. When something like this happens, you have no choice but to put in your time.” In other words, to put in the time suffering grief, of coming to terms with the new reality.
Of course, the Stoics and Buddhists claim that there is something you can do. You can change your desires. As I understand it, both philosophies include techniques for attempting to do this. I have no idea what the success rate is on these endeavors, but it seems like some desires are easier to change than others. For example, I can probably let go of a desire for, say, a job I didn’t get, or maybe even for a mate I couldn’t attract, much easier than I can change my desire not to experience chronic pain, hunger, or other threats to survival.
Still, the insight that suffering is intimately associated with desires is a powerful one. We won’t always succeed in banishing desires that cause suffering, but being aware of the relationship can help.
(It’s interesting to note that some opioids alleviate pain by disrupting this relationship, by working not so much to inhibit the painful nervous signal, but make you not care that you’re feeling it, by essentially removing or lessening the desire to not be in pain.)
So, what would it mean for a machine to have this experience? Does it actually make sense to ask this question? It implies that the machine would have layers of functionality similar to humans, that would be in differing, possibly contradictory states. But why exactly would we build a machine like that?
Wouldn’t it be more likely that a robot caregiver is highly motivated to take care of its patient right up until that patient is gone? Once the patient is gone, once the robot is sure they are beyond its ability to help them, is there any utility in the robot still receiving that motivation, that unsatisfiable desire? Along the same lines, once a robot realized that it was damaged, wouldn’t it make sense for it to simply log and then ignore the damage signals, at least until an opportunity for repair arose?
The discrepancy in humans results from the way we evolved, from adaptive emotions arising through random mutation and then being selected for. Evolution isn’t a precise engineer. Having our emotional desires be able to quickly adjust to logical realities probably never provided a strong survival advantage, so it was never selected for.
But for machines, it seems more likely that we’d design their intellectual and primary programming (their version of desires) to take into account their logical understanding of reality. Now I could see us being very conservative on when we want a robot nanny, automated nurse, or self driving car to come to this type of conclusion, but it doesn’t seem like it would be months or years afterward, as it often is in humans.
This raises the question of whether these machines would ever be entities we could really empathize with, if we’d ever see them as fellow beings. Even if we eventually come to have a perfect understanding of the neuroscience of suffering in humans and animals, would it make sense to impose that on artificial intelligences? It would seem like a cruel and unnecessary move. I don’t doubt someone will do it as an experiment, but on a mass scale, it seems like an unproductive strategy.
Unless of course there are aspects of this that I’m missing. Could suffering perhaps serve a purpose in automated systems that I’m not seeing?