Are LaMDA or GPT-3 sentient? No, but...

Key Takeaways

hey everyone david shapiro here for a quick video um i know i said i was taking a break but this is too good to um pass up um so what has happened is that a engineer from google um had a conversation with lambda and then posted this um medium article about like is this thing sentient and um this kind of conversation was inevitable um some of us have been thinking about this and talking about this for a while the short answer is no it is not sentient but it begs the question what does sentience mean um so for a little bit of background just in case you're not familiar lambda is a large language model similar to gpthree which is what i specialize in and there's you know discussions on the open ai subreddit or not subreddit the community forum and then here's here's the original blog post from google about lambda and you see that it has this like branching thing i'm not sure if that's how it actually works um because that makes it look like a hidden markov chain uh or something like that where it's just kind of statistically following a network um so like then there's you know the concept of bayes network i doubt that's how it works um but anyways the point being is that it is fundamentally a mathematical model that represents language just like gpt3 so what does sentience require though in my books natural language cognitive architecture and benevolent by design i differentiate between functional sentience and philosophical sentience so philosophical sentience is what i have and what i assume all other humans and actually many animals have which is a subjective experience of existence on the simplest definition of if you're conscious or sentient it is like it is it is like being something it is like being dave it is like being a human it is like being a cat we have a subjective experience of the world um and as best we can tell it's rooted in our brains because if our brain gets injured we might lose consciousness or we might lose sentience or it might be modified so for instance if you get like black out drunk you might still be up and wandering around but you'll have no memory of it if you have brain lesions or strokes or seizures your consciousness or sentience might be modified so for instance prosopagnosia i think i'm saying that right is face blindness so you can see a face but you don't recognize it you can't recognize any faces um and what that what that means is that there's information that is not getting into your consciousness um there's different kinds of blindness as well so for instance um there's there's a type of blindness i can't remember what it's called where your eyes are still functioning but your brain is just not processing the information and so there are people that have this kind of blindness that they can still like tell depth and like locate things but they can't consciously see it's really strange in the book um phantoms in the brain vs ramachandran talks extensively about these different kinds of brain injuries um there is a type of virus that causes you to lose proprioception which means you have no idea what your body is doing and watching people that have this this infection um it's really weird because their limbs are just kind of drift around um and then they have to like kind of consciously bring it back and it's it's really unsettling and it must be one just really awful to have that um so anyways point being is like okay consciousness and sentience what does it mean the definition that i use so just getting it getting down to basics the definition that i use for sentience for functional sentience is that you have to have an intelligent system that has self-referential information and so what do i mean by that my brain works because i am aware i'm consciously aware of my own thoughts and my own existence and my own body and i have i'm aware of my own memories right i can say like oh i remember when i was 10 and this is what happened i'm aware of the processes going on inside my own head i feel a certain way and this is why so self-explication for instance is one of the features or abilities that comes from being sentient so self-explanation is something like why did you do that and you can explain to me why you did something or why do you believe that why do you think that so self-explication is a really important test for functional sentience and this is something that i've worked on in both in in my natural language cognitive architecture it is very difficult to do so when um when this google engineer uh you know had this chat with lambda it was basically just kind of following you know following his lead and one thing that is very eerie though is that it appears to be able to self-explicate um because it can it can generate reasons um or explanations as to why it thinks a certain thing um but that ability in and of itself does not necessarily mean that uh it is sentient because again like self-explication that's just something that we can do because we're sentient but that is not the definition when you look at the definition does it have information about its own internal state the answer is a resounding no lambda and gpt3 have no idea what they are they have no idea what's going on inside of them um you know i i did a test once where i asked gpt3 like where are you and it says i'm in a room with a man and a computer and it's like no you're actually in a data center somewhere in a in a virtualized container probably um so it has no idea who or what it is or how it works so by that definition of sentience and again this might change my definition of sentience might change um but yeah so i um that was just something i wanted to address and uh i think that's it yeah i'll just cut this short but lambda gpt3 not sentient there's a big difference between philosophical sentience and functional sentience there are several of us at least around the world that i know of that are working on functionally sentient machines which has to do with self-referential information and loops nested within loops that allow the machine to kind of evaluate what am i thinking what do i believe why do i believe that what am i doing and why but uh we'll i'll get back to that in a month or two when i start my research cycle again um so anyways thanks for watching i just wanted to uh to jump in the conversation