Is it possible to fall over on purpose? What does the construction of words in a sentence tell us about the way we use natural language to convey thoughts and ideas? We often think about language as being an essential part of what it means to be human, but does the rise of new artificial intelligence technologies force us to reconsider this?

In this week’s podcast, we discuss language. As the popularity of ChatGPT and other large language models continues to grow, we ask what we are giving away to these models and speculate on how they might impact the way we use language. We discuss Ordinary language philosophy, Steven Pinker’s theory of the association between words and thoughts and linguistic codes in music from the American South.

A few things we mentioned in this podcast:

• Del Boy Falls Through the Bar - Only Fools and Horses
• Steven Pinker
• Noam Chomsky: The False Promise of ChatGPT
• The Language Revolution: How LLMs Could Transform the World
• Ordinary Language Philosophy
• Songs of Slave Resistance

For more information on Aleph Insights visit our website https://alephinsights.com or to get in touch about our podcast email [email protected]

Hello and welcome to the Cognitive Engineering Podcast produced by me, Fraser McGruer for Aleph Insights. In this series of podcasts we take a look at interesting topics and discuss what we think they tell us about analysis and decision making. I'm here with Chris Wragg, Jordan Fermanis and Nick Hare of Aleph Insights and this week we're discussing whether you can fall on purpose. Chris, ever fallen on purpose?

Well no, but that's because I think you can't. The reason this came up basically this week, could have been last week, but anyway a colleague of ours had to cancel a meeting because they weren't able to be in their house because they were having building work done and so I sort of kind of flippantly wrote that the meeting had been cancelled because their house had fallen down, they were having a wall knocked down basically and then this person replied, somebody then said oh my god that sounds disastrous and then they replied say don't know don't worry it's just one wall and it's fallen down on purpose, you know, i.e. the builders have knocked it down and we then got into a little discussion about whether actually you could fall on purpose or falling is an accident and I just had this sort of little moment where I stopped and I thought the word fall, you know, it's a simple set of four letters and yet it carries so much in, you know, when you hear the word fall you think of all these different meanings, connotations, images in your head and I thought it's just such a simple thing and it was a sort of revelatory moment for me about the sort of power of natural language and what it unlocks in people and so on and in the background I'd had all this hype that's going on at the moment about chat GPT and, you know, its seemingly uncanny ability to generate, you know, natural language that feels very human and all those kinds of things and I thought oh what are we giving away here, what is this thing that we are potentially sort of gifting to machines in some way and what does chat GPT and all of these other large language models, what does their performance in natural

language tell us about what it is? About what natural language is? Yes. Okay, nice. So look, couple of things as you were speaking. First of all, as we were just discussing before recording, I don't really know what natural language is but a bit like... It's what you might call language. Yeah, but a bit like Neo in The Matrix. I just looked at it for 60 seconds and I've got it all figured out, it's all downloaded so don't worry about that, that's fine. I don't think that's the plot of The Matrix. I don't think that's a thing that... you're thinking of Johnny Five

in Short Circuit. No, no, he's thinking about the fact that he... Does he know how to fly a helicopter or do Kung Fu? He learns Kung Fu in... Oh yeah, it turns out I've forgotten the plot of

The Matrix. There wasn't really much. You've lost the plot. Nick. Often voted the funniest scene in British sitcom history. Famously in Only Fools and Horses there's a bit where Del Boy and Trigger are at the bar and Del Boy's acting extremely cool because he's about to go and chat up a something like that and then Del Boy leans over to rest his elbow on the bar but the barman has raised the top there so he just falls over completely plank like. Yes. It's absolutely superb. I would say that you would say that the actor David Jason, the character Del Boy isn't falling on purpose but the actor David Jason is falling on purpose and I really can't see how you I think it's meaningful to talk of falling on purpose but I will accept that it's an unusual circumstance. Falling with purpose. There is a... Just hold on. Which is of course how you fly isn't it in the... yeah what? Just one thing you said you don't think one is possible without the other. It's not possible for the character to fall accidentally without the actor falling

deliberately. Correct. Well it's also the case that walking biomechanically is controlled falling. You put your centre of gravity beyond yourself and then you catch yourself with your

foot. Although I recently learned a new way to walk where one doesn't do that. Okay, hit us with it. Is it skipping? No it's meditative walking. Okay. Okay. Where what you do is very phrasal. This is what I do on Monday mornings but this is the first time. It's like yogic flying. Exactly the same. Right. But the thing is it's surprisingly difficult because what one does is before you try before you put your foot out for your next step you centre yourself and you try and centre your balance and shift your yeah. Is it like a slow march? Well I don't know but I mean but what it means is by the time you come to there's that moment when you raise and because you do it slowly right at first when you raise the leg you're shifting your balance which means that you're not falling and it's a very unusual experience. I think it sounds similar whenever

I feel slightly sombre I sometimes do a slow march. Do you hum the Darth Vader march as you're doing it? That's right. And yeah it's you have because you have to hover your foot above the ground

before you step. Right. You have to be completely in control. It's exactly the same. Yeah. But one thing so I know we're getting divert and lost but one thing interestingly is in real life one would never fall like Dale Boy fell in that and he was only able to fall like that because as you say well he's an actor right. Yeah. Because he maintains his kind of plank like position as he goes now which one wouldn't you would never do in real life. No. But the so but the point is that words have rules which don't seem to be straightforwardly grammatical. Let's put it that way. Yeah. So we're having a philosophical debate over whether you can fall over deliberately rather than linguistic or grammatical debate. And that's you know that's a particular view of language that we're taking which is that language reflects thoughts but is not itself like the driver or the substance of thought. Let's put it that way. Now in the you might have heard of the Oxford's philosophers. I think I'm going to say it was about the 50s or 60s when it was very prevalent the idea that you could solve philosophical debates simply by looking at the rules of language. So we could say here we would sit around and say and say you would say that David Jason has fallen over and you know therefore it is possible to fall on purpose rather than saying well but falling on purpose is actually a debate about some real world phenomenon. You treat it as though it was a debate about language. Anyway I think these days it's widely I think the prevailing view is that yes language expresses thoughts and it does it quite closely. We'll talk about how closely in a minute. But that it's not the same thing as thoughts right. So Stephen Pinker has written a lot of stuff about this about you know the relationship between thoughts and words and he's got an example which I think is quite good which he says that you can give me a biscuit. Correct? Correct. Go on then. I've got no biscuits to give you. But and you can give a biscuit to me right. If I had one. You think of them as the same. Now you can also give give me a headache right but you can't give a headache to me. So now that's right so words wise well the words you'd say they're pretty similar right. It's like they're both kind of just nouns and if you can give me a headache and you can give me a biscuit you think that the opposite that the other thing was true. You can give a biscuit to me and you give a headache to me. But he says the point is that that particular example reflects a difference in a sort of metaphysical claim about those things. Yeah. Which is that if you give me a biscuit then you cause me to have a biscuit. Yeah. But if you give a biscuit to me then you physically move a biscuit from you to me. And he says they're different they're just different claims which is why you can have. Give me the two sentences again. Give me a biscuit and give a biscuit to me. Right. So he just says that they're different claims. They might be reflected in you know we might think of them as the same situation but that's not a fact about language. That's a fact about how biscuits work and headaches don't work that way. You can't take your headache and give it to me. So that's anyway I think I think that's pretty persuasive. It's nicely that seems pretty yeah wrapped up nicely. So what we're saying is that actually when we're thinking about the way that words work it feels like we're having a group debate about sort of grammar or something but really we're actually having a debate about something to do with the real world and causation and the laws of nature. And yeah I mean my so but just while we're on the topic of falling over my daughter when she was little she first she she came up with this phrase which was you're falling me over which I is was really adorable but it's it I sort of that was interesting for me because that causes you to say well why can't you fall somebody over because I know perfectly well what she means and kept tripping her up. Exactly. I kept I kept kept falling her over to teach her a lesson. No. And it's because you know you've got a drill into how falling works and the answer is falling has to be something that gravity does to you. Well I mean I'm putting this forward as a theory but I think falling has to be gravity related. Yeah. If it's not gravity that's doing the work that's causing the downward motion if it's me then it's no longer falling right. Well there's a moment of potential decision or not because you can make a decision to let gravity take over or you can not make a decision and and something something else happens which means you're not involved in the decision-making process. But let's let's say gravity does the falling. But the yes but the correct way to put that would be I pushed her and she fell over. Yes. Yes. So so the thing is but but I can't fall her over because for the falling bit that has to be done by gravity right so I so you can push something now but the odd thing is like we couldn't in the case of David Jason say he pushed himself over even though actually that's probably the best way to describe what happened. But because pushing is I think our model of pushing is one entity acting on another entity. Anyway so there we are that's another example just on the topic of falling of how actually what looks like a debate about language oh no you're using that word wrongly is actually you're using the concept wrongly. Gotcha. Right well there we go. That's a nice foundation for our discussion. So let's move on to where we want to talk about how this relates to was it machine language or what was it to do GMT? Large language. Are you the last person in the world who doesn't know what chat GPT is? Pretty much probably yeah yeah I kind of know what it is but I kind of don't. Well you you're in trouble because I think we are going to pretty soon be able to replace you with chat GPT. I'm not surprised to be honest yeah surprise it's taking this long we didn't have to get that technologically advanced. Come in Jordan what

have you got? I was it's sort of related to what we were talking about but I was I was reading this paper on how on plantations in the American south slaves used to sing songs that had encoded within them they were like linguistic codes basically so like they provided directions along the underground railroad or they were a cause to rebellion or they used common metaphors so like the River Jordan was actually like a substitute for the Ohio Mississippi River things like that but obviously like the slave masters or other people weren't aware of these these codes and so I was also kind of thinking about how within language we there's so much like context that if you don't know about that something that just seems like a really facile or sort of easy to understand state but you miss the the true meaning of it and I think it's kind of it's sort of one of the things that interests me about chat GPT is that or these large language models is I feel like that sort of essence of language it kind of can't grasp like those sort of those ways that humans use language that might not be they're very sophisticated and they require a lot of like you need to be in sort of an in-group perhaps to understand what these messages are it seems to me that the way these models are trained it's difficult to impart them with that kind of

knowledge so that being the case and help forgive me but help me out guys so could that so so what sort of the enslaved people doing there would that still be classified as natural language because it's a construction within it and to become a constructed language it ceases to be just natural is that about it um but also so you're nodding jordan so that's good what's a human length like they're speaking in a human language english yeah well yeah but i mean and apologies because yeah i don't really understand this whole thing about natural language and stuff but to me it suggests something just occurs naturally rather than rather than designing a secret language or something like that which sounds more constructed anyway maybe may or may not be relevant but also help me out uh what's chat thing called what's it called chat chat gpt so jordan can you define that for me and also bring that in a little bit of of how that's relevant for what we're talking about

here okay it's a large language model that was developed by open ai what does large language model mean it's a it's a it's it's a it's a model a statistical model that generates

language and it's large because it uses it uses a lot of data yeah well yeah well actually i mean specifically it's got a lot of parameters yeah which which it uses to make its predictions so what it what it is is a predict it predicts what word is well actually technically token um so what chunk of language which you can think of as a word yeah uh is going to appear next given some foregoing piece of text yeah um so you can so i'll give you an example because i did a bit of uh using it to try and to try and sort of work out um you know what it was good or bad at has anyone else got an example of a kind of good interaction because i've got some

uh well i so i um i recently for uh some uh we were working on a project looking at um uh the future cyber threats related to space technologies right and for a bit of a laugh i um i typed in you know right write me some analysis related to um cyber threats related to cubesats which are small satellite you know tiny satellites that go out into space and it wrote this highly it was a short piece but it was it was highly um credible uh peter described it as a bit of an a-level essay you know that it was sort of that but but a bit of a synthesis and yeah yeah and but it was it was definitely uh credible what it's the things it came up with you know this is quite a niche topic it's not like a a well uh well researched topic and um the things it came up with were you know the right kinds of things it was written entirely um you could not discern that it had not been written by a person there was nothing in there that that wasn't and it was really it was pretty well crafted to be honest so um so i i don't have the specifics of of what that statement was but it was quite you know it's i don't know

three paragraphs or something but it's the thing is it okay so i don't want to go on and on about chat gpt everyone else in the world has written tons about chat gpt it's out there and you can give it a go but it is worth just getting over what it is done so that people understand what it doesn't do because that's what we want to get onto i think so but he's only trained on text right and the key thing to know is that it um so so when it's generating language all it's doing is well all it's doing i mean what it's doing is is is out of all the words that might come next it's assigning a probability to them based on it having effectively ingested all of the text that's been written on the internet ever more or less right um so it's looking at millions we think that they won't tell us despite them being called open ai they don't actually tell you how it works but it's probably looking at millions of features of text which might be something like well is the previous word the but it's also potentially the percentage of x's in the last 200 characters or you know some any feature anything you you might look at in in some text it will have learned what helps it make the most accurate predictions about what comes next and that's what it uses to right so so the point is it is able to simulate the real world a creation of text right but what it doesn't have almost certainly doesn't have is anything resembling a model of the world right now there's this um this is controversial and we probably want to unpick it a bit but the the point is that it it doesn't um it it isn't thinking about the things it's talking about right but it is still able to represent very realistic and complex sentences about the real world and they're completely can check contextually justified so it so it really does resemble the text that is being produced by someone who is thinking about the world but it but it is still a simulator it's a simulator for text creation we're getting very close to a lot of stuff here right um so two things so effectively what we're talking about is um uh what's that test of ai during test there you go oh it's sailing through that yeah exactly yeah well we're getting to that we're in that kind of area but also i feel like we've we've defined our terms here right we've we've talked about natural language we've talked about chat gbt um what is the question we want to ask and also you're getting close to it there um chris well i don't know if i've got

a an answer to the question about what our question is uh but i think i've got something that might sort of shift us in in in that direction um which picks up a little bit on jordan's point um about you know that and and and the discussion we've had so far that you know um about what language uh the context around particular words um what occurs in the human brain when that when you receive a word what it unlocks within your brain and whether that same process can occur within a language model when it receives a word does you know is the same kind of thing happening um and um i just want to posit uh i'm gonna ask you a all a question the same question um if you definitely know the answer to it just just um just park uh park your answer but um uh does anybody know what the word crungus means no fungus no right okay i want you to think about the word crungus and and tell me what you think it means just take a little time to think about

well it's definitely it's like hang on everybody have a little little think get your own get your

own fixed idea you need to think it's very clear what it means okay um go on then john john you

go you you uh like a uh what do you like almost like a uh tumor or something like uh yeah yeah

like a growth a growth or something yeah yeah i mean i feel like it's a noun um and that it is something is something sort of slightly unusual bit a bit um distasteful something about it there's a bit um and i can imagine living in some jungle somewhere or it's some kind of animal or sort of monstrous type animal okay yeah i think it's actually a sort of flaky residue on on top of something unpleasant yeah so you might talk about the crungus that had appeared on uh you know on

on a rotten uh banana or something right well here's here's the thing this is very very interesting by the way um because um uh crungus the word was effectively invented by a twitch streamer and voice actor uh guy kelly and the reason it became famous so he he made the word up out of nowhere although there are a few little traces of it in a totally different context but but it's effectively an invented word and he typed it into dali the um the image generator um and it came up with a um uh sort of hominid uh an orc type wow creature um and you know this this twisted ugly thing so it's obviously an you know you all had negative connotations um and he was like where where did that does anybody know whether and it created this sort of discovery people lots of people putting crungus into this this image generation thing and he was like have i have i discovered a you know a secret cryptid like the the yeti so effectively of all of us the one who is most like uh artificial intelligence is is uh is is fraser really from from that point of view but the point is what it sort of illustrates is this idea a weird idea of imagination on the part of a a model receiving language and creating something based on on what and it's because these things are so opaque even even if people published exactly how they worked they would still be opaque because of their massive complexity um it's been very difficult you know people have looked for traces where might this have come from and it's it's essentially a mystery

as to how well i don't think it's that mysterious if you think about um certain sounds within uh english let's say this cr sound okay if we think this crust uncrunchy and fungus um yeah it all starts to feel you know and also those sort of like i said i think i think we all described it as a noun as well yeah and nouns tend to behave in that kind of way as well and um yeah yeah but you're i

mean as a human it's not difficult to work that's the point right so we can we we can explain it right yeah but but we're explaining in in very human creative those are features of the word

right so the point is that i don't know how dally works but if it's based on something like phonemes so if it's looking at components of a word which it might well be then um you know because whatever features it's learned when it it's it's predicting what image will be associated with some text uh what what however whatever features of the text it looks at it might well have well actually fungus is quite close to some other stuff you know crust is close to some other things you've got things like krampus you've got similar looking words it might well have just picked up on that stuff yeah plus plus the fact of the similarity suggests the the explanation and the

truth in it so do you see what i mean in as much yes but what but what i think it illustrates is that when we go oh you know imagination that's an entirely human endeavor right we're just doing those same things you know you you talked about something that was like fungus you know you all had negative connotations why did you come up with those things traditionally we would have said well that's our imagination right it's it's sort of intangible and we can't really look at it but

actually it was just what that well i don't i don't i think it's misleading to describe it as imagination it well you're saying the same thing that's i mean that's what really but but what i'm what i'm saying is to say that dally has imagination is not completely correct i don't or at least i think it's contentious i think we're falling down the issues of what words mean again here right but no this is totally related so the point yeah it is but the thing is that what because we are able to um or we're able to sort of encode and capture the relationship of a word in this case to a picture but in the case of gpt3 uh of a word to other words that the fact that those words are highly connected right therefore when you have concepts that are connected they're going to be reflected in similar words let's go that far uh a statistical word generator like gpt3 will come up with those sorts of connections right but but but the difference is that you have in your mind the sense of a thing that it represents and what i'm saying is that that sense of a thing and all your other thoughts are expressible in language but merely being able to construct convincing language does not tell you that the existence of those thoughts right now it turns out that the structure of language for all the reasons we've been discussing is turns out to be much more closely tied to the structure of thought than i think a lot of people realize so that you can make a very convincing thought simulator um with a language simulator right but but let's not don't we shouldn't make the leap to say that just because you get the convincing language there must therefore be a thought behind it even though that's how we do it let me give you some examples from gpt3 right so i'm i'm going to ask you the answer i made this up totally i didn't want to rely on one of the stock examples because i would i because for example there's um uh this the two shapes which are called boob boob boobar and uh and kiki and you show people like a wobbly shape and a pointy shape and almost everyone in the world no matter what language they speak says that kiki is the spike spiky one okay right so i so i had to invent my own example and it goes like this you anyone can answer so roger floofed a zambo last week but because his zingle didn't bong he couldn't floof a zambo this week so can you explain what effect the zingle has on roger's ability to floof zambos i can't remember which way around but the point is you you can say that his zingle needs to bong in order for him to floof a zambo right based on the information you've got i can tell you now that chat gpt has not got a clue what i'm talking about and it said i there are no it doesn't know what flew for zambo zingle or bong means and cannot provide a meaningful response i see where you're going with this right we probably can yeah we we right so we we can answer that question chat because it doesn't know what those words mean yeah right and then i also just to sort of i'm surprised actually but demonstrate that it doesn't have a model of the world which a lot of people mistakenly think it must have but it doesn't i i've got so i said to it i've got a pile of books about the colours of the rainbow i've put them in order according to their position in the spectrum red orange yellow green blue etc then i rearranged them so that they're in alphabetical order can you tell me which books change position now um the point is you could work that out because you're thinking of a pile of books but chat gpt is flummoxed and says the books that change position when you rearrange them uh are indigo and violet in the order of the spectrum indigo comes before violet but in alphabetical order violet comes before indigo total bollocks and and and it thinks the other colours don't move and that's complete uh balls in fact the only one that doesn't move i think is um actually they all move so so um so yeah and it and i tried to play nim with it which requires remembering how many matchsticks are in each pile and the point is that you know it's really really good at generating language but it's very clear there is not some idea about what that language is about now a lot of um people will say okay the turing test the point of the turing test is to get rid of debates about what does intelligence really mean and and to just say well it's if it passes this let's just agree to call that intelligent but i i that it and that obviously is really useful it gets the method kind of philosophical debate out of this debate about where the machines can think um but i i don't accept that they're the same thing right and i'm not being a reactionary because i'm massively pro ai as you know um but but the point is that i think people who say look if it behaves exactly like a person producing text then it must be we we should say that it's the same as a person

that is simply not valid uh yeah go on jordan yeah massively agree with what nick's saying um i think for the reason that i just i don't the way that we learn how to speak is just totally different from the way that these models learn how to generate language it's just like it's just not the same thing so like the outputs are as well not gonna be the same like i was reading an article that um noam chomsky wrote about chat jpt and he compared like a child learning language to the way that this these models learn language and he was just saying you know like a child learns how to speak with minimal exposure to words like it might learn a couple of words a day or a week or whatever when it's very young but it's still it's it's it's using them in a in a in the way that we use them as humans and it's communicating with its parents and it's kind of formulating an idea about the way the world works whereas these models have all the data they they have a huge amount of words but they're very explicit in the way they use them it's very they they order things very grammatically it's very it's got no kind of concept of of what's going on around them or the relationships that the how to communicate with people or anything like that it's just it's just purely like a statistical model yeah i mean before

you come in it just because there's yeah there's a i think this is fascinating by the way um and it goes to the very heart of what it means to be human right but um there's a reason why language exists maybe you could say which is to in to help interpret and communicate the world experience that we have and so it's really difficult i think we're saying to replicate that in something

that doesn't move around in the world um chris yeah so just a couple of things really what one is um while i think it's not thinking in the same way as a human i think some of the things that we do which we um ascribe as being um uniquely human in some way around the use of language you know after all language is one of the things that people talk about discerning humans from other other animals um but it does some of those processes so like for example an author writing they're they're writing and they're thinking about the next word coming and when they get the right word right that that process of of being an author and being a good author when they get the right word and it feels right and they imagine that that is somehow part of a creative a creative uniquely human process it's like the writing process is is you know um it is sort of viewed as being an organic process that you know and actually that sense of what the right word is coming next is exactly what these these models are doing and are um we just we just don't know how we're doing it we do it sort of intuitively but we're obviously latching on to the same kinds of things so i

think there are yeah but we're not just looking at the features of the foregoing text that's the difference we're thinking about the thing in our head writing isn't just writing is meant to be

well yes so we've got an image of of the thing but in terms of that search for what is the right feeling word here for the context of what is going on that i think is is what they're what they're doing so so i just wonder whether in in in something effectively bit by bit this is this is kind of creep into genuine intelligence as far as i'm concerned because yes it's not doing all this stuff over here but it started doing this stuff and i you know these things that we we see as bastions that are unassailable you know i think it just it just illustrates that actually bit by bit creep by creep you know at some point um perhaps it is all uh achievable yeah yeah i

mean i think i think if we have the think ahead to having a a thing like chat gpt but instead of being trained on text it's trained on the world um which obviously includes text then i think it's much easier to imagine you know the its internal model being something much more like what we have when we're generating words you know of being actually representing some actual real world things um and the features of those things something like that would be us that would be

very very similar to i suppose my position is you've got you've got you've got linguistic determinism at one end which which you know this Orwellian idea of of language is thought right which we've decided not not the case then you've got the other extreme which is that language is inessential to thought you know and and sort of pinkers mentalese and and that sort of stuff and then something in between which is that well you know actually i suppose linguistic relativity you know stronger or weaker to some extent it language is determinant of of thought and so while this thing might not be thinking giving away language is one really important component

of thinking so but uh jordan and yeah sorry go for it yeah but the other thing is this is all just text and another thing about language is it's spoken as well it's not just so i think for these models to be really good i mean they probably need some audio video like they'd need more tone yeah yeah like how this sentence is spoken not just the construction of the sentence in terms of what word goes next but like sarcasm irony you know these like these language techniques that we use all the time like that would that would probably impress people a lot more if it if seemed it was trained on that kind of data and could understand those things because when we read

we know it's reflective of spoken language and we carry all of that yeah i think that's coming

um right so we're gonna we're gonna stop there um i've got a question in a moment um yeah i don't know if we've come to a full conclusion there but i think we've given it a rigorous going i don't want actually i did have a couple of things if we've got a minute it's just about this you know uh how to think about chat gpt and not be sort of i don't want to say deceive but not draw the wrong conclusions is i would liken it like to a map simulator and you can imagine feeding a simulator a load of uh os maps and it would be able to draw really convincing maps it would go well when you've got this many contour lines together often you'll end up with a little circle near there because that's a you know a top of a mountain but it might not know that or you know you've got roads don't just stop randomly and you know when roads meet there's often a roundabout and you can imagine it being able to generate really convincing maps um but would you say that it understands geography like does it know what a road is i mean i mean the point is it it doesn't it and it just doesn't and i'm not going to accept anyone claiming that because it's really good at generating maps it knows about the world um impressive though it might be right and that's all that but the other thing is this idea that you know oh we'll look at chat gbt wow we're very close to a general artificial intelligence to me that it might be imagine you've got someone who's a really good sculptor and they're getting more and more realistic and the last sculpture was indistinguishable from a real human being right someone like ron murk or something you know one of his one of his sculptures which are really hyper realistic is is he only you know uh is he only is he very close to creating a living human being as a result of being able to produce these increasingly convincing sculptures i i would say no i would say he's closer though but we but we but he's he's still a long way away we can't look at the sculpture and and decide how how near or far we are from creating the living human being that's what i'm saying um so there we are anyway i will stop there because i also acknowledge that there's a lot of people who don't think that who think that i'm talking rubbish and i'm being overly metaphysical and i think i'm being very hard-nosed about this and i'm so i'll be the last person to join in the whole oh ais aren't really intelligent you know but i think you we shouldn't just do away with these sorts of distinctions there is a difference but also this feels so close to so many things so for example i'm thinking about what you were saying but it's closer and always getting closer which you know you know that thing of approaching but always eternally approaching but never getting there which sort of ties into sophistry and um also with the the big bang and how we're always finding out more it just it is very alephy right literally in what i'm saying yeah chris well i

was just going to say i i think it's a significant step though that's that's that's the point i think like you know creating a a sort of perfect depiction of something from the outside so that it would represent so that you could physically represent that thing when it could you know that would be a significant step and i think i think cracking the ability to generate language is a

significant step but it also i wonder if we're always just still given technological developments are we still asking the same question essentially which is you know do um androids dream of electric sheep it's still that that we're talking about i think and i don't feel that we're any further along now we were 40 years ago let's say anyway anyway got a question um it's pretty deep

this best or worst fall you've ever had or seen uh yeah okay so i i think my my best fall i was probably about nine i think uh and uh me and my mate phil were on his bike he was giving me a backy on his on his bike and uh we were riding through these trees and he just sort of completely lost control of the bike uh and we were it was a bit like um in star wars the speeder bike scene on on end or you know in return of the jedi it was it was it was we were kind of heading towards our doom but also giggling at the same time because it was uh so it's so it's just a a quintessential childhood moment in my in my memory falling off this this bike a wonderful moment but more for

you than him of course he ended up in hospital that's right broke his legs and you know but

you were fine so yeah he cushioned your fall yeah he did yeah yeah before the days of airbags we

had uh uh children yeah uh jordan yeah i did i remember when i was in india and we were going out for dinner one night and we had i was on a motorbike pulled up into the with the restaurant and i was walking towards it and there was like a few planks over some ground and i was like okay i can just walk over those and i stepped onto one and i just went straight down waist deep into a pile of cow dung uh and i was like what do and sort of climbed up and i was just yeah just had to go and have a shower or swim in the ocean or something just to get it off but yeah it was a bad fall quite good from our point of view you know comedic fall yeah that was brilliant oh yeah it caused a

lot of lust for whoever was around yeah yeah um okay mine uh well look um i'll start with the bad ones uh yeah loads um look at my arms right and i'm covered in scars and i've got all sorts of metal inside when i was in an accident got knocked off a bike that was definitely a falling moment and a great a good fall was i remember being about eight nine years old and jumping across a ditch which classically these days is now Sainsbury's that area jumping across a ditch and like whoa falling back and my brother my big brother reached out and grabbed me and pulled me back and i still remember that moment so yeah nick yeah well my my worst one involves someone else specifically my son um when up until he was about five uh at his bedtime i used to fly him upstairs which meant that i would pick him up and uh hold him horizontally and run upstairs with him and throw him onto the bed which he obviously loved and then one day he when he was getting a bit big i tripped on the one of the stairs and smashed his face right into the floor and the poor little guy was unable to he didn't even have control of his hands so he couldn't do anything about it did you fall him over i i literally fell him over i uh so yes no i i still feel really bad about that so sorry sorry tom but did you continue no that was pretty much close to the end of flying him upstairs very good all right well we've covered a lot of ground there um thank you as always for listening to the Cognitive Engineering Podcast i'm Fraser McGruer we've been here with Jordan Fermanis, Nick Hare and Chris Wragg of Aleph Insights. Until next time, goodbye.