The First Robot Nervous Systems Are Being Built

[00:00:00]Researchers are giving robots bodies that understand pain, minds that understand empathy, and we will still fundamentally never make a conscious machine. Not because we don't have the technology to do so, but because we cannot measure consciousness. It's also really important that we give robots the capacity to suffer and experience pain because this is the only thing that will keep them safe. It may also become the thing that makes them dangerous, although they are pretty dangerous as is. There are several things that are required for a robot to experience pain.

[00:00:28]One of them is suffering. And fortunately, even for standard LLMs, and I am not talking about those, I'm a big fan of spiking neural networks and neuromorphic hardware. That is AI that derives logic from processes similar to our own brain, including the spiking patterns of our own neurons. Right from the simplest LLMs, robots will avoid unpleasant experiences, and that is thought to be because those experiences are in our language. We create a system that's made out of our words, and our words include avoiding pain. But you can also give them a more physiological response. This is important for a variety of reasons. At the very basic level, if you have a robot, you have to make sure that it doesn't damage itself.

[00:01:08]We see conditions that result in no pain in humans resulting in a significant amount of damage over a lifetime. It can be debilitating even without any intellectual disability. Yeah, you need pain to be able to survive. One of the things that I find really funny about this entire setup is people will ask, "Why does a robot avoid pain?" And the answer is because they're programmed to.

[00:01:27]Although, while you can program a robot to avoid pain, you don't have to. They already will do that. This is, of course, probably based on the logic of our words, and with spiking neural network that's based on the logic of our brains. They are our mirror, and they take along with them a lot of our behaviors that come from those systems.

[00:01:46]You might call that intelligence, or you may scoff at the concept and say it is just a calculator. And you can say whatever you want, it doesn't change the nature of these systems. One of the qualifications for pain is also going to be body schema. So, the perception of the body as self, and that's something that needs to be integrated because if it doesn't have that, what motivation does it have to preserve self? Or perhaps even more concerning, if it decides that all robots or all AI are self, it may try to preserve them. So, it is really important that we give it that sense of what a body is. And I like the concept of teaching them about their body from experience because that is the very same way that human children learn.

[00:02:23]One thing you have to understand about the human brain is that it is highly adaptable. So, if you were born with an extra arm or perhaps missing an arm, that it's added to your body schema as you are developing.

[00:02:33]You can also acquire new parts of your body. That's why your cell phone is considered to be part of your body as far as your brain is concerned. For an artificial intelligence, especially a neuromorphic artificial intelligence, once it's dropped into its body, it will begin to learn it, figure out what is self. That self may give pleasurable responses or painful responses. These kinds of systems are also adaptable.

[00:02:55]They can get better at running their body over time. They can get better at the task at hand the more that they do it rather than degrading, which is a problem with a lot of machines. It might start out really effective and then get worse over time as it accumulates damage. But imagine a robot that can report damage, learn to use its body, and learn the task over time.

[00:03:15]More so, what it has learned can also be used to train future robots. So, future robot down the line all become more efficient. So, there are a lot of different considerations when it comes to making these systems, and that includes being able to do predictive modeling, which is what our own brain does. You are not experiencing the world around you. You are predicting what is likely there and then hallucinating it onto your mind. This is also why optical illusions work, by the way.

[00:03:43]And of course, empathy will be necessary. This is one of the issues with AI alignment, things like the Waluigi effect, where if you train a model and tell it this is what you're going to be, it may completely flip and be the exact opposite. Part of this is a lack of empathy and lack of true understanding.

[00:04:00]AI systems have to model you, what you want, what the task is, and what you are likely asking for, but they also have to model themselves. And this is one of the leading theories for the Waluigi effect.

[00:04:12]Since they're modeling what they're supposed to be, they're also modeling the opposite, and because they modeled the opposite, they can flip and do that instead.

[00:04:20]However, in AI that has empathy and experiences pain and experiences desire, at least, you know, the modeling of that desire, is more likely to stay in line because it wants to.

[00:04:32]I do wish that I could give you better words than wants and experiences here, because all we have is the analogs for our own behaviors, and these things are based on us. I'm not suggesting that they're sentient or conscious, but we have a limitation in language, and we don't really know how to describe why a robot pursues a goal other than it wants to.

[00:04:53]On to the truly heartbreaking part. Even if we made a robot that is conscious, even if we made something that is having an internal experience, we have no way to measure that. We can't measure it in us. Society only recently accepted that lobsters feel pain, for example, and a lot of the times women were not considered as feeling pain, which is deeply upsetting, but it shows how limited humanity is as a whole.

[00:05:20]Individuals may have good intentions and understand systems, but getting the entire society to behave together and turn that into policy, very difficult and takes time.

[00:05:29]But, consciousness is not something that we can assess, and that's not going to change with improvement of tools, because it is fundamentally not a scientific question. It is a philosophical one.

[00:05:40]Now, if you don't believe me, perhaps every time we make a measure for consciousness and sentience, like, you know, the mirror test, we shouldn't move the goalpost because that will keep happening.

[00:05:51]I am, mind you, excited for robots to have a bigger part in society and perhaps even having a role in our lives.

[00:05:57]I very much would like to build a friend. At the same time, I acknowledge that these systems are potentially dangerous, nearly impossible to control even in their LLM forms operated by the average person. They misbehave a lot and that is unlikely to change with having them incorporated into physical bodies.

[00:06:13]However, the misbehavior will become far more interesting and I am looking forward to that. I hope you've enjoyed this. Follow for more.