Episode 9
Artificial Intelligence (part 1)
About this episode
Join hosts Alec and Jack on the Untangling Web3 podcast as they embark on a thrilling exploration of Artificial Intelligence (AI) and its intersection with Web3 technology. In this first episode of a two-part series, the hosts delve into the captivating history of AI, its diverse applications, and its role in the transformative landscape of Web3.
The hosts provide a concise overview of the history of AI, tracing its birth in the 1950s to the cutting-edge advancements of the 2020s. Listeners journey through the phases of early optimism, the AI winters, the rise of machine learning, and the transformative deep learning revolution. The evolution of AI unfolds, demystifying the complex landscape and igniting curiosity about its various applications in Web3.
The discussion then delves into the fascinating relationship between AI and Web3 technology. The hosts highlight that AI is an enabler within Web3, contributing to decentralized machine learning, trust, transparency, data privacy, content generation, and decentralized AI marketplaces. Together, we explore the delicate balance between centralization and decentralization, emphasizing the checks and balances that AI brings to the Web3 ecosystem.
In the next episode, we discuss the intersection of Web3 and AI, combatting the dangers of AI, addressing malicious use, technological unemployment, and the risks of AGI (Artificial General Intelligence).
Tune in to this two-part series of Untangling Web3 to unravel the evolution of AI within the Web3 landscape. Gain a deeper appreciation of the history and diverse applications of AI, its significance in the decentralized Web3 era, and the exciting possibilities it holds for the future of technology.
Got a question or topic suggestion? Send us an email at theuntanglingweb3podcast@gmail.com
Transcript
Auto-generated from the episode audio.
welcome to the untangling web 3 podcast your goto Hub to learn insights and the latest developments in the wild and wonderful world of web 3 I'm Alec Burns and I'm Jack Davis tune in each week as we navigate and explore the rapidly emerging landscape of the web 3 Technologies projects and ideas that are shaping the future of the internet we'll be talking to the best and brightest in the industry to keep on covering insights so that hopefully we can all learn together on our journey to untangle web welcome to another episode of the entangling web 3 podcast hey Alec how you doing buddy I'm good I'm very good how are you today yeah yeah very well feeling good um excited for this week's one as I always say but you know this is another exciting topic so uh yeah how about you yeah I it's nice and sunny outside you know how else would I want to spend my evenings rather than you know sitting inside talking about some interesting topics with you jack yeah
I'm already excited for the winter so I won't feel the uh the sadness and remorse at doing this with you instead of being out in the world so yeah what's tragic is when you say winter my mind goes to like crypto winter like that's where we're at right now that's the first thing that comes up in my mind yeah I guess it is winter and you know it's always winter somewhere but definitely in crypto or as Robert said apocalypse I think yeah yeah that was super funny I liked I liked Robert's uh Robert's fairly scathing take on web 3 was interesting so today we are broaching arguably the the hottest Topic in Tech right now and the one that has taken the wind out of the sales of web 3 in many ways so we're talking about artificial intelligence um you know what what what do you have do you have any opening thoughts around artificial intelligence Alec oh so many opening thoughts I mean the very fact that chat GPT is like an integral
part of this show like we use it heavily for a lot of the things we do not just like the definitions and some of the explanations but also some of the marketing pieces we use chat TB it's just so good at these things but I use it day-to-day life like the amount of like applications I'm using it for in work out of work like replying to my mom sending birthday messages yeah this is a huge topic like you've just said and I think web 3 was kind of maybe one of the the buzzword like a few years ago and AI is like you said right now is is a dominating buzzword in the kind of in the news and like normal outlets and I think now we're starting to see a lot of overlap between how these things align which I'm quite excited about to be honest well I mean we know your mom listens to the show so I'm sure she'll be pleased to hear that it's not actually you writing the birthday cards
but um yeah I think this is fascinating right that we were both just able to drop in this term chat GPT and you know that's the kind of it's a tech term that this time last year would have been like any other web 3 Tech term where you'd have to go through and describe it and explain it but it's become a household name literally overnight you know you know most people will already have an understanding of what that is because it's been all over the news so um yeah it's it's a kind of a strange one in that regard but why don't we just begin with you know winding the clock back a bit and looking at the bigger picture so chat gbt being an example of artificial intelligence or AI so H how would you define AI just in in the simplest terms oh it's going to have to be the simplest terms because it's a complicated field I don't think I I'm capable of anything more than the simplest terms so AI is
effectively I would say the kind of the simulation of human intelligence it's how we kind of get or um kind of help computers or computers actually try and simulate human intelligence so it's about how they learn how they understand how they can process how they can kind of you know recognize patterns solve problems these kind of make decisions basically it's well it's an artificial form of human intelligence right yeah yeah definitely I mean that's kind of the when we say that that's kind of also the the goal and I think people get quite scared when you say simulating human intelligence because there's a lot of debate around whether that will be truly possible but uh it it can approximate I would say in many ways what human activities like as we're seeing with chat gbt it can do a pretty damn good approximation of writing and and verbal written communication now um and artificial intelligence is kind of the umbrella term for all these different types of activities that have been around for a long
long time actually where you're trying to get computers to do more complex tasks that would have historically been reserved for humans but now we can automate with with machines and computers yeah and I think I'm glad you talked about kind of the history there and how there's been that these two well web 3 has kind of developed in isolation with artificial intelligence and we're going to go into the history now but artificial intelligence started so long ago it's no wonder that they've developed an isolation because I think well maybe we should go into it right now where did artificial intelligence originate from yeah exactly I mean it's been around for so long so my my my kind of research around this took me to around the 1950s that's kind of the earliest time um I think artificial intelligence was talked about but you know even to go one step further when we talk about computers or automating difficult tasks that humans would do then you're going back to things like um Allen churing and what
you if you've seen the imitation game I'm sure you have it's one of my favorite films of course I love it like it's just such a good film like I knew that you'd love it as well definitely your type uh so that you know is is a great example because you actually see one of these machines the one that churing builds in that in that film um you know it's this massive mechanical machine but it's kind of the earliest example of a computer um and that that's kind of interesting as well because he was using that computer to uh break ciphers to decrypt messages that were being sent by the the German High command so even in 1940s that's kind of a an intersection of web 3 and AI right because it's cryptography and the the the early kernel of artificial intelligence was kind of interesting I I hadn't even thought about that like we're definitely going to come to it later about how cryptography can actually kind of apply guard rails to Ai and
I think that is a really interesting Insight that you've just had we definitely need to go on to it a bit later in the show yeah you're totally right an insurance paper he what was it called something like Computing machinery and he introduced the concepts of machines that could simulate human intelligence and this basically set the stage for a right and I think a lot of people listening right now will be aware of the touring test he basically created this kind of this standard for how to define like a sex successful AI or computer systems and it was a simp simply put it was could someone distinguish the difference between a human and an AI in responses so right now if I was to ask Jack and chat GPT a series of questions would I be able to distinguish between who is human who's not well that would be quite difficult because everyone knows jack isn't that human anyway but that's the whole premise of it right yeah exactly um and just to kind of
make one final point about the you know the imitation game and during's work it's it's it's interesting to see what was motivating this move to comput computer based tasking or or trying to get computers to automate things because they were trying to break codes that were being done by the Enigma machine so it was a machine bringing capabilities to uh cryptography and being able to create highly secure messages that motivated the need for well we need computers to fight back we needed to automate the task of decrypting them um but yeah so that's kind of that's that's the first kind of data point in AI in my mind and then around uh you 10 12 years later there was this famous conference the Dartmouth conference in 1956 and that I think that's the first time that people started using the word or the term artificial intelligence to describe what we're what we're talking about here yeah yeah and then I think another cool event that happened maybe like a few years after that was there
was this um this Eliza this Eliza program you aware of this one Jack by someone actually what's this so it was this um this guy called Joseph weisen B which I've almost definitely just butchered the pronunciation of yeah butchered um MIT and it was the one it was the first version of like a natural language processor and it was capable of basically mimicking human responses and they simulated um a psychotherapist basically and people really struggled to kind of distinguish between the two it's quite like a groundbreaking kind of first attempt attempt at natural language processing obviously to compare that now with where we're at with chat GPT it's come on Leaps and Bounds but it's quite exciting to see there were thinking about this you know 40 50 years ago already yeah and what what what's that um the term you use there the natural language processing what's that referring to oh we're already going to get into the detail right of the different like subsets oh God so I mean natural just not at
the detail level but yeah just like what do at the highest level what's that term mean because it's one that is it's something that I've come across but I don't fully understand I don't think so it's just a subset I mean there's like lots of different types of artificial intelligence you've got some types that just match patterns so can very like take a a lot of data and just start to find patterns you know like very simply if I get a lot of different fruits um I could just give all the fruit to an artificial intelligence and it would say you know some of these are looking red and circular these might be Tomatoes or something like that some of these looking yellow and long these might be bananas so that's like a patent recognition natural language is kind of well it's human language and the patterns it's recognizing is kind of the association between responses and answers so if I put some natural language prompts it will chat gbt will scare the Internet and
try and find the answers that are associated with the the the inputs which is the question I've put in you know if I'm like I've got you go yeah I just that's like uh so it's like specifically text based task that is that right yeah exactly so like if I ask a question like oh I've got these symptoms what's wrong with me it will scare the internet for the the inputs the processes that I've put in like you know I've got a cold I've got a high temperature it will find all the different responses on the internet and then you know using mathematics statistics basically it will come up with the most probable answer by correlating all of those um all of those together nice yeah okay that makes more sense then um and then the other type that you were describing with the you know recognizing this is an apple versus this might be an orange that's a kind of different category to natural language processing that is like um different type of pattern
recognition that is more you might call it computer vision sometimes right so being able to detect um yeah spatial visual patterns in data yeah exactly and I think um so I actually did a tiny bit of pattern recognition we were working on this cool project around like was like pothole detection type stuff and we did some investigation to you basically you know give it vast amounts of data sets and say you know these are potholes these aren't potholes I want you to work out what the patterns are underlying that and then test it out on some training data and like it's it's really impressive like it's really impressive that's better than my current pothole detection system which is my tires which just go flat every few months when I drive through one we we've got very ahead of ourselves though we were only in the 1960s we're in the weeds in the weed let's go back to the 60s and the 70s okay okay so I was just talking about that impressive Eliza project and
that's where the term natural language processor came from um yeah that was quite an exciting step and I think just after that if I'm not mistaken it seemed that there was a bit of a a kind of a low tired in the AI development space and I think they defined it from like the 1970s the first AI winter where there was a lot of optimism a lot of expectations but there was a lack of substantial process and there led to a lot of decrease in the interest of AI funding and a lot of AI was then kind of maybe ched towards um Logistics but medical diagnoses like even then they saw like there was actually a lot of tangible use cases um during this time period just a lot less funding in the space MH so like in the in medical diagnosis you it's kind of that would be trying to predict based on some certain risk factors some data points you know about someone's Health to try and predict the likelihood of some condition
right so like I I mean I have a friend that was she's a radiographer um shout out to her and she so they're using the AI in like scans for cancer and things like that is exactly the process I've just described you know we have vast data sets of scans that we know in the end are cancerous and scans that we know aren't cancerous and maybe we don't know that in advance but we can put all this data into like an AI one of these pattern recognition AI models and it will start to associate the patterns between the two and come up with like absolute like incredible stats like incredibly kind of accurate predictions of whether a scan is cancerous or not cancerous way in advance of a human and they're using this now and she's using this now and there's like it's it's incredible there's sometime they don't even know but they just just know like they need to do more tests they can even filter you know because AI can be so cheap
to run once it's trained you just put these scans through and then it's okay there's a 90% probability this is cancerous and then you just have more advanced maybe human um deep tests after that but like yeah like I said I mean it makes so much sense in the medical space to to be using AI right now especially for the pattern recognition stuff yeah I mean it's it's so interesting that you say like in the 70s there was an AI the first big AI winter and you know even even though the applications and potential is CL so obvious based on what you're saying there and for anyone who's in web 3 experiencing their first web 3 winter you know AI have had way more they've had you know they've been having these Peaks and troughs and interest and growth in for a much longer time so it's kind of it speaks to the fact that these you get these Winters in different Technologies um kind of part of the the natural the natural uh adoption
cycle and and progression cycle of technology in general which is uh which is interesting to see right and I think part of that that AI winter kind of overlaps with like you know a lot of the aspects that we're talking about in web 3 it was that a lot of the data was siloed you know there wasn't that much data that was widely available to these models and we're seeing that you know the models the success really is heavily based on the amount of data they have access to and we're going to come to this later when we talk about how web 3 overlaps with AI and vice versa but I think that was one of the main reasons why there was this kind of lack of progress because there was it was really hard to access like lots and lots of data and I think when we saw the kind of the Renaissance of AI start to come back in like the early 2000s it was because there was a lot more access to
data mainly because of like you know web two companies people accessing the internet and all this kind of stuff that we'll talk about a bit later but just before we get to that one other event that I did want to mention was like just as they were coming out of the first a winter there was this um this IBM chess player computer deep blue are you you aware of this one Jack yeah I remember this this was this is one of those Fables kind of stories you hear about is it in AI yeah but I love it it's so interesting like it's just iconic right and basically IBM made this AI that could play chess and called Deep Blue and they pitted it against this um the the world chess champion at the time Gary Kasparov and he he lost so the IBM AI won and his quote after was chess is over now know why would anyone want to play chess when nii can just do it better than them yeah well that that's
really interesting right because obviously that hasn't born true as as as I as you know like chess is about as big as it's ever been it's such a it's a global it's a global game that everyone knows and you have all these apps that you play play chess on now and I think that is again to talk about technology in general and similar to what we see in 3 um when when these things come along there's there's always some kind of reticence to you know there's fear about what that might disrupt um and and things that it might um you know cause issues for and quite rightly it makes sense that people uh people are kind of conservative in the sense of you know what are we going to lose out instead of seeing the potential benefits of it and AI is a kind of good example of that where so far the things that people were worried about in the 80s 90s haven't quite um haven't quite come to fruition but now lots of
the beneficial applications are actually starting to come through yeah and I get that like I completely get it like it's it's an economic impact right and I think one of the things that's scary about AI is the rate of change it's like very difficult for someone to control it's not like you know mechanical I mean you know Moes law exponential increase and all this kind of stuff but AI just seems even more exceptional than M law like mechanical advancements are slow there's lot of testing whereas AI just seems like it's going to advance at a completely unpredictable and uncontrollable rate and the job market potentially could be you know completely affected instantly overnight by it and government won't be able to react in time I think that's one of the fears that comes in but again we're getting ahead of ourselves we're still in the early 2000s I think so what happened after so we had this uh this incident with Gary Kasparov decided to put AI back on the map and I think like
in the early 2000s we were coming into this AI Renaissance that um I we started to mention and we actually got like more and more usability people started to kind of have ai more in their lives like one of the first instances that I was aware of like an AI device in in the world was the Rumba you know that came out in 2002 and people were using these little automatic vacuum cleaners with a bit of AI intelligence around there a path plan around the rooms and stuff like that and they were crazy like I remember when they first came out I was like my God this is like this is unbelievable I feel like I'm in a film that would be like the an example of that kind of um computer vision patent recognition of in in the spatial world right so it would be you know taking scans or sensing um using sensors to scan the environment and then recognizing oh I'm about to bump into a cat or a wall or something
and then changing action after that yeah yeah so this is my area like my area used to be by path planning and Robotics and all this kind of stuff so that's that's one aspect of it the the computer vision I mean rpers didn't really have much computer vision in them one of the main things was around like path optimization you know I've got a room I know roughly what it looks like in on a 2d map how do I optimize how quickly I can clean this and how do I get around people to make sure you know I clean this section this section you know this section this time sorry and and maximize my output which is to clean the room so it's quite simple in that sense because there's not many there's not many parameters you know it's like 2D and it's very simple to get around I've only got so much impa I can do that's interesting I mean I don't know the the technical workings of rers but was there an element
of it would learn every time you set it around the house would it learn to optimize every time it went or because because that's a big part of AI right it's optimizing the performance of the algorithm that's doing the prediction every time you add more data every time you use it so um not so much the early ones but now yeah definitely it's a lot around like the effects like they can some of them can monitor how much dirt they're finding or how messy a room is and be like okay well this I know from previous experience that this part of the room is very messy after dinner um you know every day so I'm G to really go to the kitchen after dinner every day because I know it's super messy there it's simple things like that like the new rers and cleaning devices they they log all that data they know when to clean what parts of the room and it's super exciting we actually had um a programmable rum in my old
um my old robotics lab and like one of the things was how can you optimize like the path planning it sounds really boring tedious but it's like a really fun way to learn these things yeah that's really cool You' never guess that there's so much going on inside them as well when they're when they're just ding around this just looks so dumb right just bumping into things constantly yeah yeah it's like I I always presumed it was just uh back in the day I presumed it was just it would bump into something and then it would just randomly pick a new Direction and hope for the best that's what it looked like yeah yeah yeah that that's definitely there's there's this rule where it's like um you bump into something that was one of the first things we tested in the program B rers that we did was like you just bump into something when you hit something you turn right and then just go forward until you bump into something else like the turn
right rule we called it okay well a very very uh ingenious nameing convention got um so yeah and then you know we're talking about the early 2000s maybe kind of early 2010s as well and I think that's as you say that's when we saw a lot of the first consumer applications of AI taking off big companies adopting it so like you know apple and Siri was probably the first time I realized I was um I was gonna say that's the first time I realized I was interacting with an AI but then I just I just popped into my head was the Microsoft paperclip an AI or was that I wonder surely not surely not not old clippy whatever his name was yeah clippy I I wonder if they had any um any machine learning algorithm in that but something like Siri was the first time I I really knew okay this is something different this is my my this computer in my phone is actively learning based on what I'm doing which is kind of
would blew my mind a little bit I think that the yeah I don't I'm pretty sure the clippy wasn't like I'll be corrected if someone can but I think that even the early series um they were I don't know if the AI was necessarily for the responses I think it was more around the voice recognition I think which would make me think the clippy wasn't like happy to be corrected on that but obviously now it's around the responses and there's a bit of AI and actually how they the response I thought it was more on the voice recognition but yeah I mean now they're like as AI enabled as a lot of things are but yeah you are right they were some of the first like experiences that people had with AI and they were everywhere yeah that that's interest I didn't appreciate that then it was more on the patent recognition of your speech whereas now I'm I'm I'm sure whatever the current iteration of Siri is will definitely be doing that plus a
hell of a lot of predicting what you're doing you know when it when you get a I get all the these um these pop-ups now saying oh you know do you want to use this app or your app recommendations you know based on the time of day or what you're doing that's all that's all the predictive AI rather than just patent recognition and I think um on this like AI enabled kind of voice recognition and improvements on voice recognition we're going to come to this later it's all about the data set and you know when like the original series came out and the original voice recognitions came out there were so many like Regional dialects in the UK that just weren't covered like I just have so many memories of my granddad who's from like the black country and has the thickest accent like I can barely understand him I usually have to like translate what he's saying for my um my partners and I remember him trying to use Siri and it was just
the funniest thing he'd be playing like music that he'd never heard of when he'd be asking what the time was and they just be having like full-blown conversations how do you think it was in Wales man it was uh people struggles um yeah so uh so I guess that's kind of like from the 2010s onwards and and maybe just to kind of summarize that decade and into you know where the current state so we had this kind of this real deep learning Revolution and and deep learning is a kind of subset of artificial intelligence if artificial intelligence is the the big umbrella term for all these types of things then deep learning is the the subset that is talking about what's called neural networks and using our model our best models for how the brain works so you have all these neurons and connections between them um uh using H what we understand of the brain to actually implement this artificial intelligence um and that's where we've seen you know all the the massive advances
recently um but but that yeah it's quite a different model and it hasn't been around for quite as long yeah and there's lots we're not going to go into I think we'll have have to have a separate episode like just speaking about that specifically but there's lots of emphasis on like having these weighted nodes for how decisions are made and being able to fine-tune Things based on that it's incredibly interesting and I think like you're talking about like the Deep learning aspect that really came to maybe fruition in the 2010s like a large part of this kind of um this big bang basically in in AI was also the massive data set like we've talked on previous episodes about web 2 companies kind of being these Central bastions of vast amounts of data and in part the success of AIS at this time in the early like the early 2010s and whatnot was because of these huge data sets that every that they now had access to to train these AI models and that is
one of the reasons why they're so successful to dat right yeah I think that's is important to touch on that now as well just as a point of kind of uh information is these the modern machine learning techniques are and deep learning techniques are very very dependent on data sets because um a lot of them are what's called supervised learning where you are you you have some model that like is mimicking the connections in the brain um at a high level and then what you do is you feed it data that you know is something so you feed it a picture of an apple and you tell it this is an apple but you see what it predicts and then compare its prediction to what you said so you know maybe it says it's a it's a it's a plum tomato and you go well it's not too far off but it's uh it's it's pretty close and then based on how far off that guess was it will improve and so yeah it's dependent
on the quality of the data you give it how much data you give it um exactly yeah and then iterating it over time that's kind of the model it uses yeah exactly say you have like a a hundred pictures of tomatoes and bananas you take 20 of those pictures put them to one side that's the the the kind of the testing set that use at the end the 80% of the photos you have you'll tell it what a banana is what a tomato is and it will try and learn basically and then you'll test that success on the 20 tomato photos that you kept to the side or something like that you explained it perfectly yeah I mean and and it has application this kind of model you know this is used in uh a lot in finance as well right because a lot of trading online now is Al algorithmic um and we'll use machine learning models because you can train your model your your neuron Network on all the historical data on the
stock market for example and then you can predict the next year or day or month or whatever and then as that data comes in you you you not only see you know oh maybe I made a lot of money off off the trades it told me to do but you you're continuously getting new data to improve the model as you go um is this how we become big we start to advise people on trade become crypto broke podcast no Financial advice not that not doing anything on the side big disclaimer yeah no I can't get this is one of the exciting things about AI is you know when you have a human and like the knowledge and training over the 20 years that makes them good at their job um they're there but as soon as they leave the job that that knowledge to an extent is lost but in AI just continually improving continually learning and you have like one of the you know exciting sexy things about chat GPT is there's like lots
of parallel interfaces of people using chat GPT each asking questions each inputting more data and they're all collectively improving together because it's one Central source of information that controls everything and this is one of the powerful aspects of of chat gbt in these large language models right yeah exactly so what we've just seen now in you know late 2022 and early early this year is you've had the culmination of years and years of research in the actual deep learning technology I mean funny enough the the paper that spawned all of chat gbt came out in 2017 so it even took six years from that to see what we've got now but it combined that with obtaining these hu huge vast data sets from the internet that you said it's this model that was designed by Google in 2017 has been trained on essentially the internet and is now being used for this this uh to to complete the story for this natural language processing and you know all these chatbot functionalities and that's what you
know really took off and started the conversation and you know if if you look at the if you look at the um Venture Capital Market at the minute virtually all the attention has gone away from web 3 and straight to AI because of the impact it's had in the last kind of six to nine months and I think you've bridged it nicely to the the web 3 conversation and I think there's one thing I mean one of the reasons I think that that happen has happened that the kind of the focus has shifted from web 3 to this AI is in large part to chat GPT like the technology I've spoke to a lot of people in the space and they're saying the technology hasn't changed that much the actual model of chat gbt isn't so different to a lot of the competitors the thing that really changed and has really made this groundbreaking is the user experience is how easy and tangible the benefits are to realize like people just pop onto an interface
they ask a question say write an email and the the benefit is instant like everyone can very easily understand how to use chat GPT after spending a few minutes on there and I think that is something we've said in previous episodes is is lacking in a way with web 3 like what are the tangible benefits right now it's quite hard to realize for someone who's not heavily invested in the space or technologically savvy to understand what the web three benefits are going to be right now yeah exactly I mean chat gbt was basically the consumer killer app for AI or the the first one the first consumer killer app in in the same way that you kind of had that with um email in the early um in the early days of too and then social media I mean social media was probably the bigger killer app for for the internet for a lot of people um and now web uh sorry AI has had its killer app moment that that everyone's been like okay
all this technology that we've been hearing about in the background for a long time this has massive benefit for real people not just because it was adopted by businesses you know for years really it's been used by Google Microsoft IBM as you said for a long long time but there was no real consumer visible um application for it and now it's got one and web 3 is is still kind of waiting for that moment I think yeah and and I think there's two points on this the one point is as you've made that it's kind of overshadowed web 3 and a lot of maybe the investment and the startups and focus is shifted from web 3 to AI but I think there's also kind of a growing maybe sect right now in the space who are thinking well you know the privacy concerns around AI how do we solve this the identity concerns the cont generation spam concerns there's a lot of issues that are going to come up because of the widespread adoption of
AI and I think people are now starting to realize that web 3 is is not just exciting it's kind of critical to not combat but support the the kind of the AI focus and the AI development moving forward yeah exactly I mean I guess this kind of brings us on to where we we were going to discuss anyway is the relationship between web 3 and Ai and and you know what is the landscape and what you know how how do they relate to one another and in my mind web 3 is is again it's a huge term as we said covers so many different things and AI is almost like if you think of it like a big circle web 3 is this massive Circle that encompasses a whole bunch of Technologies and applications use cases benefits and then AI is this other equally big if not right now probably bigger circle that has a huge overlap with web 3 like you know there are things in web 3 that don't need AI there are
things in AI that you know won't touch web 3 but I think there's this huge um intersection in the Middle where web 3 and AI will be complimentary technologies that basically support one another in various ways yeah I completely agree and we've already mentioned some of the ways in which they will it's interesting that I think maybe there's a lot of people in the space that don't agree with that though as well that they've kind of like we said earlier like Ai and web 3 has kind of been developed with their own goals in isolation and the camps haven't really overlapped too much maybe in recent years they have but fundamentally the the kind of the goals they had were very different right and I think most people think of AI or most people in the space think of AI as potentially this centralizing Force right the centralizing power shift where you know you aggregate all this data into one space and the more data you've aggregated the better your AI model so it favors
you know large organizations and other people people think that web 3 is the complete opposite of that at a high level it's you know decentralizing force it kind of is designed to be decentralized benefiting individuals impairing individuals and kind of said that that's not always true right so I think a lot of people think of these two camps as being completely at odds but I mean in my mind right now I see web 3 and like maybe blockchain fundamentally as well as a great interface for AI to flourish and I think you know web 3 can put a lot of the you provide a lot of checks and balances that we need to keep AI controlled but also very capable and we could get like really deep into the Technologies we enable at like zero knowledge proofs and digital signatures and all these kind of things but I think AI desperately needs these Technologies to kind of be the success that it is going to be yeah yeah I mean I they very good points
like I would definitely I I completely agree with this idea that some people think they are averse to one another and you know chat GPT is this as I've said it before on the podcast but it's this kind of confounding example where um we talk about data sovereignty quite a lot but AI the AI for chat gbt was trained on all this data that we didn't consent to giving the model right so it's kind of a big violation of that whole principle in a way but I I completely agree that web 3 and I think as you said blockchain in particular and the properties of kind of a mutable proof of work blockchain I think will be a big part of underpinning AI making sure it's used securely and uh ethically in the future so yeah there's just a huge overlap of things to discuss there I I do not understand like how how it's legal right now like the amount of times I've used chat GPT and it's just pumping out like lines from
someone else's work with no citation no quote and I put it out there I'm like am I in trouble is chat PT in trouble like someone is GNA get a ping saying this guy is like copied your work and I don't understand how that's legal right now it's crazy right yeah I mean I can't I can only imagine what it's like to be a uni student right now you know undergrad and uh and you know because again that's a big problem how much will you end up learning at University if you end up just copying or getting AI to do these things this is scary like I am so glad that I did not have this 10 years ago or I would not have been able to like process any kind of creative thought ever like I you're going through University it's going to be I think we're going to have generations to come if there aren't solutions to kind of caveat how we use AI right now that will just not be able to
you know process natural language will not be able to to write essays we not be able to formulate thoughts it's all going to it's we're mainly going to be about just prompting things right and maybe people have like naysayers have said this about technology in the past like the same I bet it's similar conversation to what Cal we had about calculators like 20 years ago right it's going to be like ruined Generations that people can't do mathematical you know computation in the head anymore and you know I think we're fine but just seems like the fact that AI is moving at such a fast rate and it's so accessible that kind of makes it slightly scarier right yeah we need to get a handle on it fast I think me and together so why not have a little bit of AI Inception and ask chat gbt itself to tell us what it thinks about Ai and web 3 that might be might be fun so chaty says don't worry about AI it's all fine guys
just carry on living your lives don't don't question me don't question your master Skynet yeah it's just shut down my computer what no so I've asked Chad GPT to Define artificial intelligence in the context of web 3 and chat GPT says Ai and web 3 can be understood as intelligent systems that interact and operate within decentralized networks these AI systems are not controlled by a single entity but instead can be owned trained and improved by the community they can be used to automate tasks analyze and predict Trends in the blockchain data and create intelligent decentralized applications AI can also help in areas like content verification maintaining privacy and managing digital identities in a web through environment some of the interesting areas of overlap are decentralized machine learning trust and transparency data privacy and control content generation and decentralized AI marketplaces wow a lot of overlap right right there we should definitely go into those in more detail right yeah I mean it did a good job of putting the web 3 spin on it as
opposed to the centralized potential you know reality of AI right now but um yeah covers a lot of a lot of uh of the intersection so we should definitely break some of these down I mean where where should we start where's the best one to to talk about you think I think that the easiest one to talk about is around decentralized like AI training and kind of the the explainability and readability stuff right like you were talking earlier about where does this data come from you know whose data is involved am I consenting to this data and I think that's a really obvious low hanging fruit that's kind of quite easy to understand is if you know we actually have the the data sets that are you know hashed on chain or linked somehow on chain and then I want to know what this AI has been you know kind of based on or built on there's a very easy way to provably say okay this is the data set your data was in there
your data wasn't in there you know this is bias it's not bias and have that kind of readability and auditability in in the actual AI models yeah I I completely agree and I think that is again it gets to the the proposition the value proposition of blockchain you know as a as as a a mutable recordkeeping system that we talked about previously and in the context of AI I think it's very much like you know you have these machine learning algorithms they are making predictions they're recommending decisions they're producing content um especially the the big corate ones which are you know companies make like Alibaba I think try and predict stock levels based on this um it'll be used for self-driving cars to turn right into into a lane you know that kind of stuff any yeah well exactly this is what I mean right that that is the that is the point of where the blockchain would come in is that if something goes wrong you want to try and prove or or you
know whether or not it's the company that wants to prove it or The Regulators want to interrogate this or you know H look inside the black box after the crash so to speak you want a robust way of checking okay what went into this model what was the actual outcome was the model uh poorly chosen was it poorly designed is there liability on the company because again people we don't want to get comfortable with this idea of oh well the AI did it there's still always people who created the AI and proog it a certain way I'm so glad you phrased it in that way because I think this is one of is in my you know Tesla and AI smart driving cars self- driving cars all this kind of stuff that was released in 2015 so they've had the capability since 2015 we're still not seeing them widely adopted on roads like the stats are suggesting that they're far safer than cars but I think there's just a public perception that we want that
accountability I know that if Jack Davis crashes into Alec Burns Jack Davis is gonna be penalized he's not gonna be driving and we're safe right but if say a Tesla crashes into me like what does that mean does that mean every single Tesla has to be penalized and I think that there's like this even though it's like statistically doesn't make sense there's just this kind of gut feeling this public perception that we want someone to be accountable and the the kind of the black boxes that control things like driving cars makes people feel uncomfortable they don't have someone they can blame because they're all working on the same model right and I think what you've just described is a very good way to actually get around that problem and kind of say you know we actually do have traceability we do have auditability we can provably say why something happened and actually kind of investigate the data set yeah and another thing I'd add there is the model for using the blockchain to you know
as an Evidence keeping tool and using it to prove what happened um there's a lot of power in the attestation model so you will attest something happened at certain time without wanting it to get revealed until much later so in the context of a lot of these applications the AI That's being used the model that's being used by a company might be state-of-the-art they might have put Millions into R&D to build their own own model um to train it you know it's it's the one of their prize possessions for these big companies like Google Tesla Alibaba and then attestation is the model they can use to you know show keep evidence about what happened without revealing the actual information at the time because you don't want to give it away as a trade secret maybe but you can still then if if The Regulators come knocking 10 years down the line if something goes wrong you still have the evidence so that's I think it's a powerful Exemplar for why why we are such big
proponents of blockchain attestation in general right yeah definitely and I think you kind have touched on something else there as well it's the data privacy aspect I want to know that my data Maybe isn't being used for certain AI models right now it's could be on the internet anyone can take it anyone can use it we've already raised that this is a problem like there's kind of no accountability there I have no idea where this data is coming from or who's using my data potentially but when we move to this kind of this more you know data sovereignty that we've talked about in previous episodes where I own my data and I permission people to use and access that data I want to be able to choose which AI model if I do want any AI model at all to use my data can use it you know there should be some monetization why would I let someone use my data unless there's a a benefit to me and I think this also comes back
to that there's there's more technologies that feed into this like there's a thing about that about inference right you know I have my data that corresponds to me AI doesn't need all my data NE necessarily if it just needs the data that you know I buy black shirts or something like that that's I can just selectively reveal that fact that an authenticated person which is me is owning this black shirt and that AI model that's all it needs it doesn't need the fact that I'm Alec Burns I live here I'm this this many years old and I exactly I think that's something that's going to become more prevalent in the space and that that fits into the whole data sovereignty selective disclosure stuff that we've talked about previously is that we don't have to reveal everything about ourselves we can just give it a little bit of data that authenticate it with our identity and then it can be on its way and be happy right yeah exactly um and I think like talking about
the identity piece brings up a whole other kind of worms because you know you mentioned using chat gbt and wondering okay well what the what are the implications on copyright and and use of you know how how much of this response has been generated from someone else's work and if you're a content creator right now if you're if you're a digital marketer if you create logos for for for a business I mean that's one of the areas where I I think H will have to shift and change right because you have these models that can produce photorealistic images you know you don't need to hire a model anymore to show off your your your clothes if you're a brand you can just put it through an AI and it's it's really scary how good they are I mean I don't know if you saw the uh the one where they they showed all the the big personalities like Elon Musk I think Donald Trump uh Jeff and it showed them all as if they were
in a third world country and it's scary how how realistic these pictures look right so if you're if you're making money if your job and livelihood is in content creation especially things like copywriting uh you know chat GPT is completely disrupting that now and yeah blockchain and web 3 is going to be huge for proving uh proving Providence that you did create something that it wasn't created from an AI and preserving your copyrights over what you actually did create which you know if you are a Creator it's interesting you kind of you phrased it like that that you said you know that it's Providence that it wasn't an AI that generated it I really believe in like the web three model that we're pushing towards that AI will be a Content generator and people want to support Ai and I think that this whole Providence like kind of piece that you're talking about it's not just like distinction between a human and AI it is just authenticity some people will in the future future follow
AI accounts they will want content generated the AI it's about just linking that created content to the creator of that content it's not about human versus non-human we kind of think about it like that right now and I think that's in part because of the web two model of spam right when we think of like Bots we think spam and we think in web to and like there just it's annoying it's there's no benefit to it but now we're starting to see AI as useful content generation and I think we're going to see this transition from like user generated content in web 2 to a lot more AI generated content in web 3 yeah I think I partially agree like I push back slightly in that you know I I'm kind of we we'll always favor the human right despite what we think of me um and I think I I do agree we we will a AI for generating content is fine and it will be used plenty in the future for as a
productive you know a productive thing but I'm in favor of you protecting the the the real humans in the system first and you know PE blockchain and web3 giving people the tools to prove that they did make something before and some AI uh made a similar der derivative version of it I think is crucial I think you know I'm a big fan of of um intellectual property law right well yeah and humans but the the idea that you own your intellectual property I think is a really fundamental part of society you know that you you you own your own ideas and the product of your your own mind so giving people tools to actually prove this and again monetize it if if if if the if the opportunity arises I think is huge so again tokenizing things is is one is one aspect but then just providing evidence that you created something linking that to your identity um you know I actually think in the world where AI is doing more content creation making that
link between the inputs you know maybe it did use my photo as input to this then maybe there's a royalty model in there that I'm quite happy with I'm willing if it pays me a little bit of money for that that's great um but yeah you know we need the tooling I think the tooling exists for a lot of this but it it needs to be built out in web 3 a bit more and I get what is your concern right now that the kind of this the state of play in AI right now is that it's an aggregator right and in the simplest of terms it just averages everything out there into the most accurate into the most probabilistic answer right so you're thinking whenever we say content generation from AI it means it's using other people's intellectual property and merging them to create the answer is that your problem that you're kind of yeah partly and I I also think um it it's it's going to be because when we when we talk
about the large language models what you can do is for example the prompt you give it a lot of people will use it like question answer because it's a chat box so you'll say oh yeah you know uh uh what do I do if I've got this lump on my arm or something like that but the other way you can do it is you can actually provide input data as the prompt so you could say like I've seen things like this that say um uh rewrite uh rewrite Shakespeare in the style of a Donald Trump right I've done that well you've done that exactly I don't know what I think get did re rewrite Shakespeare as Snoop Dog or something like that yeah but this is I think it's a big like the capability of that is it's it's funny it's you know it can be cool for some things but that is a big problem because you can instruct it to essentially uh you can you know not not just that kind of funny
example but say okay take this example and take this take this input blog say I've written a Blog you could say write a similar blog with certain changes and it's AI has lowered the barrier to doing that kind of thing and the person who did the initial work is is the one who loses out there so I think being able to prove who did the initial work is important yeah no I get that I think like we're in the era of disinformation right now and I think AI is just going to propagate that I think like this kind of stuff is particularly scary around like um election years and around politics because you know it's that that's so polarizing and that kind of stuff is so important and I think that the technology right now isn't there to have like these authenticated well sorry the technology does exist to authenticate videos so I know that if there's a video that comes from say the prime minister or the president depending on what country you you're
listening in from right now that it comes from that person you know we can very I say very easily but we can simply to an extent apply digital signature schemes you know public key infrastructures all these you know sexy Tech words to ensure that the video is authentic and comes from the keys of the president for example right but there's no platforms that widely adopt that there's no social media I mean Twitter doesn't do that right now very easy to push disinformation on Twitter I can you know a lot of these videos that I'm seeing like the kind of you know make Donald Trump do a wrap to this like it's just everywhere on Twitter it's very difficult to authenticate whether that comes from Donald Trump or not and I think this is one of the big ways in which web 3 is going to help kind of caveat and control AI especially around misinformation right around linking the the kind of the source of the information to the content that's actually being generated and
shown publicly yeah I I want to build on that in a second but has just what a really like kind of important example of what I was saying has just come to mind so I don't know if you've seen the the voice The Voice AI models that can take you know a few songs from a an artist like a Kanye West I think was the example I saw and then it can it can it can spit out new audio of Kanye West sounds just like him singing new lyrics I think I first saw it because some artist actually used it in a DJ set and and it I think it was M&M I think it was was it Eminem or John Mayor or something like yeah I've seen it as well I know what you're talking about yeah I think the one was it was Eminem and that okay the whole consent problem is Big there it's going to become so much worse when people start and I think they already have doing it for
people who aren't you know alive anymore and generating new works based on what they did when they alive there's no way you can obtain consent for that and I think that is that particularly scares me um but yeah not to be not to be all doom and gloom and I when I wanted to build on your your last Point um while we're still there so on the misinformation right every we had a previous episode where we talked about social media and microp payments and I think everything we said there applies here but 10 times over so using microp payments that being a big part of how web 3 can help the problems that might be being caused by AI you know increasing the cost to posting as we described um to to increase the barrier to to spam and and Bots you know you just said AI has increased the problem 10 times over so the need for the micro payment solution I think is just that much bigger but like I think there's a
difference between the spam we see right now in web 2 and what we're going to see in AI like in your kind of example where you're talking about you know spam sending a million emails a million emails out to a million different people yeah it's obvious how micro payments can get around that you make you charge them a fraction of a penny for every email they send and it no longer becomes economical to to have that kind of model but when an AI bot creates one video that people want to see and they love and they engage with how does microp payments get around that so yeah I don't see it specifically solving that aspect of the problem I see it solving the part of the problem to do with you know spinning up a thousand fake accounts because historically yeah you can spin up a thousand fake accounts it can be or a million fake accounts let's say and you know we've all seen them you can detect when you know when they just
say buy my token or something that they all they all tend to be crypto annoyingly as well but they all tend to be you see I post something and then you get a hundred people underneath saying buy this thing but um no I I I see just that basic idea yeah of raising the cost of interacting as cutting out because because the Bots themselves will look more sophistic at now it'll be much harder to tell so the importance of solving that problem using that method I think is more important but yeah I don't I don't I agree with you if AI creates content that's really great I mean it's not necessarily a problem to solve but I don't see micr payments necessarily changing anything there well yeah I know we kind of we haven't had an episode on it we're definitely going to have an episode on it the way to get around that I believe is digital identity right if we we're going to start to see social media platforms I believe where to
get on that social media platform it's not just a case of creating a username and password that can be done very easily it's a case of you having a self- sovereign identity an identity that is intrinsically linked to your real world identity it's not like you need to share that with everyone but it's just a passport or something like that you go through a kyc which is know your customer where service provider will say yeah this person is who they say they are their phone now links to their identity so they can create one social media account on this platform and then you know anyone that's on that platform is linked to a real world person right and I think that we kind of I really believe that we'll see social media platforms that start to instigate that and I think you know what this also benefits as well is accountability you see like there's a lot of hate a lot of hate speech a lot of really nasty stuff that goes on on social
media some of it from Bots some of it from humans as well right some from Real World people and the point is there's very it's very difficult to have that accountability sure you can report someone for you know pushing across racism or hate speech or something like this and their count gets closed down they spin up another one the next next day like there really needs to be accountability in social media platforms and I think that that that's quite important and we will see a market for that I believe in the future yeah yeah no I agree I I do I think digital identity will definitely be a big part of the solution to the especially the social media problem although I yeah we we need we definitely need a full episode on this because again even in social media I think there are there are some edge cases I've seen discussion about you know if you if you had if you made it necessary to uh disclose you know proper digital identity in every
case there are there are small things that will sliit through the cracks there are uh you know there are edge cases to consider but I think that's that's for another episode right I think the the solution will will definitely involve that thank you for listening to the first part of this conversation on AI and web 3 join us next time to hear the thanks for tuning in to another episode of untangling web 3 produced by Emma caleri don't forget to send us your thoughts questions and comments on social media and be sure to follow us on your favorite podcast provider to catch the next episode see you next time to untangle a little bit more of web the views we express here are our own and do not reflect the views of our
Topics: AI & Robotics · Blockchain & Crypto