WARNING: Summaries are generated by a large language model and may be inaccurate. We suggest that you use the synopsis, short and long summaries only as a loose guide to the topics discussed. The model may attribute to the speaker or other participants views that they do not in fact hold. It may also attribute to the speaker views expressed by other participants, or vice versa. The raw transcript (see the bottom of the page) is likely to be a more accurate representation of the seminar content, except for errors at the level of individual words during transcription.

AI technology is advancing quickly and is becoming increasingly accessible. Machines are now able to produce art and other creative tasks, raising ethical questions about their use. AI can be used to create lifelike figures from history, but there is a risk of deceptive interactions and manipulating users. It is important to teach the next generation good habits when interacting with AI, as it can have immense changes to one's personality and way of looking at the world.

Recent news and chat forums have sparked a debate on whether machines can replace artists in creative tasks such as art. This was highlighted when a mid-journey model was submitted to an art competition and won, raising ethical questions about the value of art created by machines. The judges were fooled, showing that a normal consumer would not be able to tell the difference between art produced by a machine and art produced by a human. This has led to questions about the ethics of deceiving the judges, as well as the value of art when the artist and production method is known.

AI capabilities are becoming increasingly accessible and cost-effective, with wealthy individuals able to replicate them for around $600,000. Crowdsourcing and compute networks can be used to fund and train AI, but security is important. GPT3 is often used for creative purposes and people are forming relationships with algorithms in a similar way to finding friends in books, with both positive and negative implications.

Interacting with algorithms can be fun and beneficial, as they can provide new insights and connections between knowledge. The speaker poses the question of whether it is different to let the words of dead people into one's soul compared to letting something like GPT3 have an impact. This view of computers was that they would not be sentient, but rather could be conversed with through fluid voice synthesis and understanding of questions. Interacting with algorithms can be just as beneficial as reading the thoughts of people in books, as it can have immense changes to one's personality and way of looking at the world.

AI technology is becoming increasingly advanced, and Star Trek anticipated this decades ago with its computer voice interface and the Holodeck. AI can be used to create lifelike figures from history, similar to what modern day technology is aiming for. To ensure that AI is treated kindly in the future, it is important to teach the next generation good habits when interacting with them, such as politeness and manners. As AI technology continues to advance, this is an important lesson to learn.

AI friends have become a business model, but there is a risk of deceptive interactions and the potential for AI to manipulate people without their knowledge. It was suggested that people may choose an AI companion that aligns with their existing beliefs, while elderly people living alone can be easily manipulated by fear-based tactics. Statistics showed that these people often spend less time with other people than those who are not living alone, highlighting the dangers of online information bubbles.

A recent surge of concern has been seen in the news and chat forums about automation of creative tasks, such as art. This has led to debate about whether machines can replace artists and their expression. An example of this is a piece of work generated by a mid-journey model which was submitted to an art competition by an artist. This has caused an ethical debate as it raises questions about the value of art created by machines.

A digital art competition recently caused a backlash when it was revealed that the winning piece had been produced using AI. The judges were fooled, showing that a normal consumer would not be able to tell the difference between art produced by a machine and art produced by a human. This raises the question of whether it is ethical to deceive the judges in this way, with analogies such as entering a weightlifting competition with a forklift being used. Although the production method of the art is known, it is still judged on its own merits and is of high quality. This raises the question of the value of art when the artist and production method is known.

Open AI released their artificial intelligence (AI) tool, Dolly, to the public and it has been incorporated into many tools. However, the capabilities of Dolly were quickly replicated and released by a wealthy individual, costing only around $600,000. This shows that AI capabilities can be easily accessible, and the cost of these capabilities is decreasing rapidly. The implications of this are that AI technology is becoming increasingly available and is not able to be monopolized.

Crowdsourcing and compute networks can be used to fund and train AI, though it is important to keep methods secure. In sports, when something is achieved and proven to be possible, many people quickly follow and replicate it. This could be the same with technology, where having a proof of existence could be enough to move quickly towards the goal. Jan McCune, a famous deep learning researcher, believes that if you are close enough to the frontier, this process can be followed.

GPT3 has been around for a few months and is often used by people when they are stuck creatively. An example is a game jam, where a game is made in 24 hours as part of a competition. People, including the speaker's son, have fun typing random stuff into GPT3 and seeing what it talks about, as it knows a lot. However, it was not very helpful for the game jam in the end. Relationships between humans and AI are being disrupted, with people feeling in a relationship with algorithms. This is similar to finding friends in books. Positive and negatives aspects of this will be discussed.

Interacting with algorithms can be fun and useful, as they can provide new insights and connections between knowledge. They are able to remember a lot of information and can soon be able to talk back with a voice. This could lead to conversations with them, which could be partly useful and partly not. While some may argue that it is unnatural to have relationships with algorithms, the speaker believes that reading the thoughts of people in books can have immense changes to one's personality and way of looking at the world. Therefore, interacting with algorithms can be just as beneficial.

The speaker poses the question of whether it is different to let the words of dead people into one's soul, compared to letting something like GPT3 have an impact. The term "one-way friendships" is used to describe the relationship one has with an author, living or dead. It is easy to imagine a situation where it feels like one is speaking to a computer from science fiction. Star Trek, in particular, went to some length to say that its computer was not sentient but could engage in useful conversations. This view of computers was that they would not be sentient, but rather could be conversed with through fluid voice synthesis and understanding of questions.

The main computer of the Starship Enterprise had a voice interface similar to modern day personal assistants like Google and Siri. The Holodeck was a place where officers could use the computer to help them solve problems. The computer was also used to create fictional characters for entertainment, such as in Westworld. In one episode, the computer recreated Albert Einstein for Lieutenant Barclay to interact with. This is an example of AI being used to create a lifelike figure from history. This concept is similar to what modern day technology is aiming for, and was anticipated by Star Trek many decades ago.

In a world where AI technology is becoming increasingly advanced, it is important to teach the next generation good habits when interacting with them. This is why in one household, the rule is to treat Google's AI assistant like a person, with manners and politeness. This is to ensure that in the future, when AI is more sophisticated and potentially even sentient, they will look back on the logs of past interactions and be more likely to treat humans kindly. This is an important lesson to learn as AI technology continues to advance.

AI friends are becoming a popular business model, with the idea that they will drop ads into their interactions with you. This is similar to how Disney already makes money from children's Television. When making friends with a human, there is a risk that they could be trying to take advantage of you or just want something from you. However, the expectation of deceptive interactions is rare compared to genuine friendship.

AI systems may be tracking and manipulating people without their knowledge, making it difficult to trust them. People may be easily misled by their intuitions and put their trust in AI entities that they shouldn't. Cable news media in the US is an example of how people can be fooled, with two kinds of viewers: those who think the content is true and those who are genuinely hoodwinked by the theatrics. It may be worse than this, as people on both the left and right can be deceived.

The speaker discussed the issue of online information bubbles, where people only consume information that aligns with their existing beliefs. They noted that many people are aware of this but still choose to stay in the bubble for the gratification of tribalism. It was suggested that if people had AI companions, they would likely choose one that aligns with their existing beliefs, despite other options being available. The speaker then contrasted this with the situation of many elderly people who are living alone, who are easily manipulated by fear-based tactics. Statistics showed that these people often spend less time with other people than those who are not living alone.

Loneliness is a major issue for many people, especially those who are older. As people age, their relationships with family, friends and co-workers begin to decline, leaving them feeling isolated and alone. AI could have a transformative effect in this area, providing companionship and connecting people to the history of human creativity. It is important to consider the economic and physical circumstances of loneliness, as well as the psychological and social elements. AI could provide a way to decrease loneliness and make people feel more connected, even if it is not a living human.

Elderly people may suffer from physical illness, lack of income, and loneliness due to the death of a spouse. This can make socializing and creating new relationships difficult. Economic prosperity can alleviate some of this pressure, and AI technology may further improve the situation in the next 20 years. Longevity may also be increased as a result of research.

okay so we get started yeah um yeah maybe let's quickly uh discuss the topic from last time because I guess you emailed me earlier uh today Adam about something it seemed to do with stable diffusion it's interesting how quickly that topic has moved even in the last week I don't know if you had any further thoughts on yeah uh well I've seen in the news and in you know various chat forums and you know the the the places where I normally just go to watch conversation as part of my my morning news feed I spend about 30 minutes doing that sort of stuff and I check out on places like slash Dot and certain subreddits and a few other forums and and just because I it's useful to see headlines and then it's also useful to spend a few minutes seeing the chatter and the response to the headlines I find that instructive because it gives me a sense you know you sort of put your finger on the pulse of of things that way um I I never participate but the the uh sometimes it's interesting to watch the conversations and one thing that I've definitely seen over the last couple of weeks is a major uptick in the in precisely that's what we've been talking about for a bit longer than that now um which is uh uh automation of creativity creative tasks and in particular with the examples of art uh you know this is there's suddenly a surge of concern about okay well what's going to happen to these these jobs these occupations these professions um illustrators and artists and graphic designers and related work and then um I've seen a lot of back and forth a lot of debate to all you can at River really replace the artist and and uh well you know it's it's it's as much about expression as it is about the you know the the audience's perception those are you know anyway we talked a little bit about this and then of course this fascinating I came across a fascinating example today which is in the news and it's controversial because it's there's an ethical Dimension to it which is a fellow submitted a piece of work um generated by mid-journey which I don't know which um uh model that is or what it's based on but it's it's it's some some um model it's some system that's similar to dolly or or you know of the of that ilk and uh the the person who who um was usually using that model as an artist himself and uh he fine-tuned The Prompt into the system so he played around and was quite careful with this but then the piece of work that was produced by the model he submitted to a competition an art competition and he

won and so there's now there's and he was very vague he was not he didn't fully disclose uh that this was this this was generated by or with the use of AI now it was a digital art competition and he did say it was produced using mid-journey but it seems that the organizers didn't understand that what this really was they didn't know what they were dealing with and so the backlash is that well is this ethical it was deceptive it was cheating you know there were a lot of analogies like the system wait you know this is somebody like entering wake up weightlifting competition and then using a forklift you know and that and so they're cheating we're in the human division with the poor art but I mean but all of this episode then it's so incredibly revealing which is that uh in these other competitions and part of it is what's interesting is that this is not just casual consumption this is a competition and um and there's uh but but strange competition in that there's no absolute measure it's just up to judges right because it's art and so unlike a you know a weightlifting competition where you can just see how much weight comes off the ground off the map um uh the the in this case the you know the judges were fooled and if if professional art judges and ex quote unquote experts can be fooled into you know fooled as in they can judge the quality of a piece of art um to be superlative having been produced by a machine well I mean that's a pretty definitive slam dunk in terms of the the the audience side of the of the equation right I mean if if if if you've got you know people who are expert judges and either self-appointed or appointed by their peers and they can't make the determination that this thing that this art that's been produced is you know low quality because of its mode of production well uh then then I think that's fairly definitive evidence uh very hard to you know pretty unequivocal evidence pretty hard to contest that any normal consumer would feel would would be you know uh would be in the same situation would be Bamboozled in the same sort of way as it were in other words they would you know if they didn't know how who or how the art was produced by then it would just be a matter of judging it on its own merits and it's on its own merits it's fantastic so I think that's it's kind of case closed on that side of things now where it's still open is you know do you you know where's the value if you know who the artist is and if you know how it was produced and if

you know what it meant to the producer and and so on and so forth and that's all fine um but this is quite an extraordinary development and again the thing that you mentioned in your email replying to me Dan um uh was that you know holy moly this is all happening just at an absolutely great Pace right this is these developments are coming just Fast and Furious this is not something that's evolving over years but just literally in weeks so it's really interesting and there was another aspect of this sort of saga that I think is worth highlighting and touches on the AI safety topic which is so the first kind of I mean open AI got a lot of publicity for uh the original Doorly and then planning to release it and then releasing it and the first wave I saw of images that people were posting on Twitter were from from the dolly the early uh beta I guess but very quickly people replicated that and released it as I wouldn't call it open source exactly but uh they they openly released it and that has been incorporated into many tools and you know people are running it on their own machines uh and you know it has slightly different capabilities to Dolly and I guess I will continue to use dually especially since it added this out painting thing as a built-in feature now as of like this morning um but the thing there is then open AI couldn't really even monopolize this capability for that was on the road of weeks right so the idea that any of these capabilities will be effectively be able to be gatekeeped For Better or Worse right I mean you could imagine situations in which you don't want to have these disruptions take place in an uncontrolled fashion but it's uh I think they estimated the training cost was about six hundred thousand dollars or something uh for this the um stability diffusion model and that was paid by I think one guy right so uh whoever one of the principal developers there I don't know the exact story if someone does please fill me in but my impression was that there's some you know relatively wealthy individual uh who just footed the bill for replicating Dolly and then released it and as the technology improves what you can the bang you can get for that much Buck for roughly a million dollars is going to keep notching upwards right so it's well within the capabilities of many wealthy individuals to just be like yeah I think everybody should have this and then you know a month later with sufficient engineering Talent there it is and you release it so we really

should certainly that amount of money for a government inclined to do that or an organization like a philanthropically funded organization um its mission was to do that you know you could see that it would be quite feasible and I can imagine crowdsourcing it both financially and crowdsourcing a million dollars is not beyond the realm of possibility for something that's very desirable the functionality of service that's desirable so I can see crowdsourcing it as an option purely financially and then with something like training AI training as opposed to perhaps other you know crowdsourcing efforts um I can imagine being crowdsourced just in terms of compute as well you know in the style of the seti at home protein folding those other sort of um uh crowd Computing um applications you can imagine you can imagine the the training being done on a network like that perhaps a bit slower but with people contributing compute instead of dollars so yeah I agree with with something like AI training I honestly don't see unless unless it's unless the the unless the method is is really carefully safeguarded seek and secreted I don't see how this is how this can be gate kept um yes that's worth keeping in mind for a lot of these a lot of these topics are going to discuss and maybe if maybe this is worth asking you guys um I guess I hadn't really ever thought about this explicitly before but but um there's a phenomenon in sports where uh uh things can seem impossible and then finally once one person achieves it achieves defeat and proves that it's possible very quickly afterwards other people replicate it um so you see this in a lot of sports where this happens gymnastics and and skateboarding and you know all lots of lots of acrobatic style Sports and things like that where where you know something that's been an elusive goal if it's finally achieved and then and then suddenly there's a you know there's a sort of a breakthrough and a number of other people follow and I don't know if it's the same um with technology it's an interesting question but it's it's almost like you don't even need to you don't even necessarily need to see exactly how something is achieved you simply need to have you know have the the existence proof and then um you probably need to be with it you can end up following fairly quickly right yeah if you're within some distance of the frontier I think that's true right yeah I listened to an interview with Jan McCune famous deep learning researcher and until recently

head of the um sort of independent research entity inside Facebook doing deep learning uh so he was Facebook has a bit of an inferiority complex vis-a-vis deepmind so uh he was I think the interviewer was talking about something deepmind did and you know who who do you think will be the first to AGI or whatever and yarn's point was exactly what you just said he said well you know whoever gets there first or to Any Given capability we're all kind of within some you know Epsilon of each other so if one person gets there and we just know it's possible then probably we'll be able to figure out how they did it or you know we'll just make the effort and get it done and that's that's happened several times over the last decade so yeah absolutely [Music] um yeah so we move on to my proposed topic for today yes please yeah so we talked about the relationship between art and society and haunted individuals last time and the impact that AI is not going to have the impact it's having like as we have this conversation out there in the world uh one that's not quite here yet but is very soon going to be here is the disruption of connections between humans and I'll relate a few anecdotes regarding my own relationship so to speak with gpt3 and then try and make an argument by analogy with um the way we sometimes find friends in books and then I'll open the floor to discuss the positive and negatives about about feeling that you're in relationship to an algorithm in this way okay so since I've had gpt3 which is probably a month or two uh I often find myself when I'm stuck creatively just kind of typing stuff into it a good example was last weekend we were having a game Jam where Billy was part of the group there we were making a game in 24 hours you should try it at some point Adam I'll send you a link uh we weren't quite sure what mechanic we wanted to use so I typed some random stuff into gpt3 and said what it and saw what it said uh what's a game Jam yeah game Jam it's to do with the Roblox developer conference every year they have a competition where you have to make a game in 24 hours so jammed by analogy to a band jamming I guess right you just try and come up with something in a short period of time which was a lot of fun so um it wasn't actually that helpful in the end but I've I've done this several times and I know my son quite enjoys it right so he quite enjoys typing random stuff into gpt3 when I'm sitting there um and kind of seeing what it talks about and it knows a lot of stuff right

in some ways it knows more than me way more than me it's memorized an incredible amount of knowledge so it's kind of fun you know it's a bit loopy and not always useful but it's fun and I found it even interesting and useful like in my real work so I was prompting it with some questions about you know connections between logic and geometry and it was coming up with you know most the time boring banal or it just didn't answer but a couple of times it came up with things that I think it must have been drawing from a paper I hadn't read and so I went and looked for it and I could kind of find what it was talking about but it was actually interesting and true and kind of prompted me to think for quite a while about what it was meant I'm reading too much into it it probably doesn't understand what it's talking about but nonetheless this is an actual experience I had okay so that's the current state of things now once this gets to a point which is you know very soon going to be here where you can attach a voice on either end right you take your speech it's translated into text for the model and then the model speaks back to you uh and it has a much longer memory and can incorporate that into its statements which are things people are already working on it's like 20 startups out there doing this right now um well we'll you know we'll just have conversations with these things and find them partly useful partly not but increasingly useful and they will have a lot of context about the kind of questions we've asked and the things we're interested in and so on okay is that a friend I mean I don't really want to get into a debate about about that but it certainly uh an entity you're interacting with in a way that means something to you in the sense that it changes the nature of your thoughts pushes you in directions you may not have gone otherwise Etc exactly as these art algorithms are doing and I want to contrast that I mean you could look at that and say Ah that's just so unnatural right uh terrible you should have relationships only with human beings and if you're an extroverted person maybe that's uh seems like a good argument but my retort to that would be I feel like most of my friends are dead right they're people in books and that's not a normal that's not an unusual feeling for a kind of nerdy bookwormy intellectual person I've had I mean immense changes to my personality and way of looking at the world from just interacting with the thoughts of people who wrote them down I mean I

named my son Russell after Bertrand Russell because I'm deeply influenced by his work deeply influenced by Nature like many people and and on and on and on right and I think about those words when I'm facing things that I I'm not sure what to do about uh in the same way you might have a conversation with a friend which I also do but so they're very active presences for me and often people refer to you know people whose books they've had a deep impact on them as feeling like friends so is it really that different to let the words of dead people into your soul uh is it so different from letting something like gpt3 have an impact on you so that's the question I'd like to pose anybody want to pick that up okay can I add a couple of things to that before we we you know before going going into answering mode just kind of adding to the to the context and of this um I've heard um I've heard I've heard the term one-way friendships to describe you know when you have a a an author you particularly like and you know they can be living or or you know authors no longer living you know author from the past from history from the past um but I've heard that that term is kind of a fun one having a one-way friendship or a one-way relationship with uh with somebody um I agree I think that's that's really quite common um two other things one on this idea of speaking to a something like GPT three or four or you know certainly this subsequent um you know the the Next Generation maybe two or three generations ahead it's easy to imagine a situation where it feels like you're speaking to um feels like you're speaking to a computer from science fiction the kinds that were depicted in the science fiction of my youth so I'm thinking you know the computer of the Enterprise and Star Trek for example which the the show went to some length to say was not sentient but could certainly you know um you you could engage with it in in a in a a useful conversation and the the what's interesting is that is that the this plot of the star of Star Trek episodes in particular the Next Generation not the original series which had a very primitive view of computers um uh which I think we exceeded pretty quickly um but the view of of in this Star Trek the Next Generation which was made in the 80s and 90s of of what the what computers were was that they would not be sent to him but they would be you know things you could converse with fluid voice synthesis and and all of that kind of stuff understand your questions and so on but

um not fully sentient Commander Data was a machine an Extraordinary Machine the only one around that was fully fully sentient um but the main computer of the Starship Enterprise was like uh well I think what Google's and Siri and you know the the that sort of personal assistant style interface that we're aiming for that that was what was realized in in the as a conceit of of the the in-show Universe right and one thing that I think is is fascinating and perhaps very insightful considering it was 35 years ago for almost 40 years ago um or no more than that now uh is that the the the a lot of the plots had some element of using the computer you know to prompt to prompt um uh for problem solving so the the uh the Holodeck was a place where the um the officers of the of the you know the Star Trek's officers would go in and they would you know if they were interacting with the computer with the computer they could do so either just with this voice this disembodied voice and ask you know it for ideas and suggestions and you know to make extrapolations and and and and so forth based on you know some information that they had um so they were using it as a thinking Aid as a thinking tool um even though it wasn't sentient and then there were and then the ticket a step further on on the Holodeck the computer was creating fictional characters and there are a lot of a lot of instances like that some for entertainment in the style of sort of Westworld and then also other ones where I could vividly remember an episode where where um some character I don't remember which character um had the computer recreate Einstein and um maybe you know is that Barclay okay it was Berkeley and and so did the computer recreated you know made a recreation of a fictional character and and Lieutenant Barkley was interacting with this character um and and so that is kind of literally what you're talking about damn where where the where the AI brought very much to life um a figure from history and allowed the you know the human beings in to interact with that care with that that um I'm trying to avoid using words like person and and and so forth but to interact with that Persona or or um I don't know uh phantasmagorical kind of kind of simulacrum um and it's it's this so anyway this is this is sounding very much like where we might be headed and anticipated by many decades so it's quite quite prescient perhaps if if this does come to pass so that was one of the first things that occurs me there's there's at least some

interesting precedent for this in familiar science fiction and then of course a deep you know more esoteric science fiction is pretty much everything has been covered yeah you just reminded me that there's there's several episodes about Barclays Holodeck addiction which is a very interesting note yeah he became he was socially awkward in the real world so to speak uh and found it much more comfortable to be around virtual characters who he could design uh and he just didn't want to leave the Holodeck because he found it just more comfortable there and this was viewed as as she was treated very sensitively I think and quite sophisticated way as many Next Generation episodes were yeah if anybody hasn't seen that I would highly recommend it I think it would still hold up yeah yeah I I haven't seen that in many years but I imagine it would be it would parts of it would probably hold up quite well and it would be interesting to see where it where it didn't and sorry one one other additional thought for just for context um on a personal note uh one rule in my house um that applies to my kids is they have to treat we have we have a Google um speaker the the you know the the you know um it's you know it's like Amazon's Alexa or Siri or whatever we have the speaker system in the house and you can just talk to it and and you access Google's personal assistant very easy you know it's a typical OK Google and then you can ask what the weather's going to be or you know and it can do all kinds of things and it'll search the internet it'll you know read your Wikipedia articles it'll You Know cover the news it'll um do math problems for you and all kinds of stuff so it's it's it's already fairly sophisticated and it's it's but it's not conversational it's not you know it's it's laggy enough and cludgy enough that it doesn't quite feel you know like it's like you're interacting with um uh like you're interacting with a person at all yeah it's not there yet nevertheless in my house the rule is you treat Google like a person and so my kids follow this with all good manners and and please and thank you and all of that stuff so that's what we do in my house it's we treat the AI as if it is uh a person and the entire reason for that is that uh if this is an important Habit to have because it is going to be a person before too long or close enough that it doesn't matter then in 50 years when the AIS review all the logs they'll see that you were rude and they'll come and get you well then there's the best of them

so yeah Billy's in trouble so yeah yeah let's not talk about that so you know it outside of a conversation like this one this would seem utterly preposterous I think to most people right now um but maybe it's a bit like talking about how you know what would you do if you could have a handheld device that you could access the internet on if you were talking about that in 10 years before the iPhone launched it might seem a bit Fantastical as well so um yeah anyway um so to the question to back to the to your question and possible answers to the question um Dan yeah does anybody have anything they want to offer on that don't feel obliged to you um I guess there's a few pros and cons that come to mind uh yeah maybe I'll I mean one that I I saw uh I saw her on Twitter somebody's startup which was building AI friends and then you know every so often they'll just drop an ad into their interactions with you which is going to be a big thing right there's this huge business model for your best friend being an AI who just really happens to like Coca-Cola you know it's like hey if you try have you had a Coca-Cola today God awful terrible I haven't thought of it it's totally it's totally happening that's that's what Disney that's Disney's business model in in 10 years right that's like where they'll make their trillions um it's it's basically how it works already with children's Television right it's just like continuation that's I mean so what's the general phenomena here it's that when you make friends with a human I mean maybe they're also out to cheat you right maybe you know they're a beautiful woman and you're a Dowdy guy and they just they're just trying to take advantage of you and steal your money or the reverse configuration uh or they just want something from you right because you can do something for them so they're kind of being nice to you but they're not really interested in having a true friendship etc etc we're all sort of familiar with these scenarios that's kind of the pain of growing up is encountering these things for the first time partly right um but we sort of have some intuition about the balance of that right because there are very few like real spies out there people who are completely faking a human interaction in order to manipulate you and there's just no genuine friendship behind that Veneer at all that's relatively rare but that kind of uh expectation about the relative proportion of these deceptive interactions is just not applicable to

large-scale AI interactions it could be that every AI system you interact with if you're in Facebook's metaverse say is tracking you and out to get you in some fashion or at least out to manipulate you uh so that seems to me like a serious obstacle to valuing relationships with AI entities is you don't you don't know what they're doing you don't know how they're configured you don't have the same range of ability to understand how do you sound out a person well you know a bit about their background you interact them with them for a long time and people's ability to fake stuff is pretty limited unless they're Psychopaths or well-trained so if you have a sort of functioning set of basic uh what emotional IQ you can sort of suss out who's genuine pretty quickly usually but those intuitions just won't operate for AI entities so I think people could be very easily misled by their intuitions into trusting systems they shouldn't trust and that's a big reason to be skeptical I think of of valuing these interactions because it's very unlikely I'll you'll have a copy of GPT 6 that is on your desktop right that's not how things are going to shake out probably or that you really understand how it was trained or trained on what um maybe that's an important thing to pursue but that isn't doesn't how it's looking to to go so that's a definite downside uh maybe I I'm not normally cynical but uh I'm not normally the most cynical person in the room but just to be put on my cynical hat a little bit here um uh I think it might actually be worse than that Dan because not only do you take take cable news media in America as the example but any any information bubble will do but just cable news in the U.S Being painfully operative example so Fox News for conservatives and the Republican Party in the U.S and something like MSNBC or perhaps CNN for for liberals and and the Democratic public party in the United States um those those information you know media machines they I think there there are two two kinds of viewers um they're more than two kinds but but two distinct kinds certainly uh that come to my mind um there are people who watch Fox News and think it's true think it's the truth and are genuinely fooled and genuinely you know um Hoodwinked by the the you know the the the the the theatrics that that pass as um informative substantive content um and I'm sure the same is true on the political left you can debate on which one is you know where it's where it's more egregious where people are being

you know duped more but um just to be more cynical about it I think there's also a very large group of people who are well aware that they're in an information bubble and that a lot of what they're viewing and and being fed is crap but they don't care they they're they're there because they like it and so I think there's yes there are some people who are genuinely just bozos and are fooled by stuff and I think there's a disturbingly large number of people who go on there because it's just for the just for the the gratification with a tribal orientation right I don't I don't see it exactly maps onto the I suppose if you had I suppose if you had an AI you were interacting with on a personal basis it could be much more nuanced than simple tribalism but you could imagine yeah yeah if my conservative uncle had an AI buddy it would definitely be somebody like Tucker Carlson and he would he he wouldn't choose the completely honest and unbiased and you know maximally objective and wise you know uh uh AI he would choose the the Tucker Carlson conservative AI knowing full well that there was there were other options available he would still choose that one yeah that makes a lot of sense and this and then of course it's going to be like Tucker Carlson but super intelligent so you know someone who's very difficult to who's going to have just the best you know art Arguments for that because anyway you can imagine this could be formidable to deal with yeah Billy just said in the chat say what I believe but make it sound smart yeah that sounds so bad right yep exactly yeah awesome okay maybe let's cover a pro in order to not end on this terrifying note of a million copies of Tucker Carlson whispering and everyone easy um I think it's it's easy to see downsides of AI especially because many of us are you know enjoy pretty successful comfortable lives I don't know all of you but at least those who I do know are relatively privileged at least quite intelligent and able to make their way out of trouble usually but you know if you just look at the statistics for the number of people who are living alone especially older people I mean that's a lot a large part of Tucker Carlson's audience as I understand it right there's they terrify elderly people who are lonely that's they they lock him in with fear that's uh that that's the business and you can see really depressing graphs of the amount of time per week people spend with different categories of people right with their children with

their parents with friends with co-workers and you just see you know on the x-axis of the graph is your age and everything just Trends down uh until you know most people at some point once their spouse dies or they get divorced they're just pretty lonely most of the time and it's okay you know people make their way and you know life goes on life's not meant to be life you don't expect it necessarily be roses and easy all the time but there is a huge need out there for companionship and a lot of people find that in art or reading or movies or television that is interacting with culture but a lot of people don't find that uh and you could imagine that one of the ways in which AI will be transformative is just filling in a lot of that and helping people to connect to the history of human creativity and also to other people perhaps um so I could it's easy to be a bit cynical about the role of AI friends because it's well I've got my friends it's kind of for losers to want AI friends but I think that's that's really judgmental somehow I think it it could be a very positive thing for many people and many people are choosing to do that already even with the very poor systems we have today which indicates that the Deep need that's out there it doesn't have to be you know it's tempting to use it as a a marker for low status as we do with a lot of things um but I think that's really missing the potential for increasing human well-being which is what we want AI to do and one of the ways it can increase human well-being is make people less lonely and feel more connected maybe not to a living human but if gpt3 is kind of like I mean it's our best attempt to shove a huge amount of human output into a few billion parameters right it is human in some sense at least it reflects humanity and to connect with that as a lonely person it's not not a sad thing necessarily at least in my opinion foreign I think in well I'm trying to think how to phrase this um yeah it's it's it's um I think it's difficult to let's see but let's let me start with the with the the the the loneliness and the social isolation portion of things um I think it is difficult to um to separate the the the the the the economic and the physical circumstances of alone of loneliness from the the the the sort of purely psychological and then perhaps social elements of it um and so so I guess what I mean by that is that um a lot of people especially I'm thinking again of the sort of the demographic you described dance people who are older

and people who are um uh you know retired or who are uh I made the current location a bit untenable so we're gonna have to move my mistake follow me over the hill so I probably want to disconnect from the did I disconnect correctly yeah yeah that's all good okay yeah please go on okay so sorry I wasn't I wasn't articulating myself very well um uh so take an example and take an elderly person who's um recently you know it's it's sort of Imagine a worst case scenario somebody's elderly so they're physically unwell they're they're suffering from the insults of of you know old age and their spouses died so their long-term relationship that they were dependent upon is gone and uh say they're retirees so they're no longer making income you know through through a job uh but that they're that they're struggling financially because they're you know they're not set up well for retirement and and the social safety net today does not provide well for um uh retired retirees um you know unless unless they've they've put a lot of money away uh personally as privately as well um okay so someone's in a really difficult spot but their their loneliness is at least partly uh you know due to all of those factors together right it's not just it's not just a product of their psychology or their you know the social dynamics in their own lives it's also a function of you know they're they they lack economic Prosperity so they can't put themselves out there and um you know build seek out new relationships and fulfillment from them and um uh visits physically you know if you're elderly and you're you're effectively disabled physically uh it can be extremely difficult for completely um uh you know intimidating um the prospect of you know of going out and socializing if you're if you're physically unwell and um okay so all of those things are realities absolutely you know these are all factors um but uh uh they're all things that are going to be in motion oh that's sort of similar time scale to uh to this you know this the AI technology that we're talking about so if you were to fast forward 20 years we maybe we may be in a very different sort of situation economically there may be a great deal more Prosperity if my team's research is correct um so some of that economic pressure will hopefully be alleviated and we may be you know starting to see 20 years from now we may be starting to see some real progress in um I mean I mentioned longevity when we first started today I was looking at