AI Expert Panel - Resurrected George Carlin! Creative Industries - Artificial General Intelligence

today we have another AI expert panel I'm happy to have Sam seain the founder of Theos Seth early the founder of early information science and O Shota the founder of oilfield strategic Solutions so guys to jump in today there's a lot happening in the realm of the creative Industries with AI I mean they reincarnated George Carlin and they had the 1hour comedy special where it was scary how good it was like you could have sworn he was commenting on today's news so what do you guys think of what's happening right now in Creative space of AI I would call it try to impress people by injecting AI in it I saw that special because a lot of people were talking about it and what I thought about it who really did that what who is the human who did that because apparently what they did is voice generated by AI make it similar to the comedian also some one improved the text was the text generated by AI but hey not for us everybody is using C GPT and knows that it cannot be autonomous you need to to work it out to appear good so I think the the hidden artist here is a human who was using Ai and let's say it it's it's a show done by a human using AI so I wonder who was that human and why nobody want want to take credit for it yes it make a lot of noise when you say it's a I but let's be real guys we know all of us it's not pure a i it's a big Talent Amplified by AI hey it's interesting in in in the way of industrial world we don't have yet that threet I believe that we going to have some uh fake personas robots or would you have which all them the in your image that's going to be doing anything or telling you anything so it's not we're not there yet in in the industrial world it's a lot more I think prevalent in the politics and certainly in the entertainment industry where each person carries a lot of a lot of money behind in terms of BU products we don't have that it's the security is important so we need to know we talking to the right people so I think it's more a question of security for the natural resource to be sure you're talking to the right person but other than that I I don't see this being a massive threat in in all a lot more entertainment that's shter I agree with the comment that there there's a human behind something when anything truly creative has to have some human Direction and guidance if you look at Sora the video creation application someone has to give it ideas and props and direct the the application to perform a particular task or produce a particular output and I think what people are realizing is yes this is very powerful and it can provide a lot of capab ability but it's really for humans it's really tools in the toolbox I actually had a conversation with the former head of the motion picture Academy the ERS Andy MZ who is the former head of the Science and Technology division had just spoken for him many years ago and he was saying that look there the movie studios are certainly interested in this but we're not going to be seeing there's a lot of issues around character likeness but we're not going to be seeing a lot of movies that are going to be created strictly from AI there's a lot of interesting things but we can speed up the process of Motion Picture develop right you can create story boards and matics and all sorts of things that much more quickly you can tell the visual story in pre-production and that's a valuable tool for human producers directors writers and other creative set set designers and so on so there's a lot that can be done to speed up the workflows But ultimately behind the scenes there's they're humans and we have to look at this as a human creativity tool Echo what Seth said rather than fearing AI taking over the jobs of creators should we be looking at at this as more of a tool help get creativity flowing in fact I think it's the concept of for human intelligence augmentation we are heading toward a a new era when our intelligence will be Amplified by by these three TOS the same way in the industrial revolution a machine ified our capability and replaced our muscles so this time it will touch our recognition our capacity of thinking but the machine will not think and will not replace us will just amplify what we are thinking about going back to tools like Sora it's unpressed right now but just the word unpressed is it useful not yet but I'm doubting within a three four years five will have this tool or this engines integrating production of already existing platforms and here it will be the impact of amplifying the human creativity because what you used to do I don't know you have a Creator who want to create something in 3D it takes him I don't know to to put his um Imagination on on screen dream off maybe with this tool taking three hours and that will be the real effect so calling it um AI will replace creativity is just a way to attract the clicks and and attention but the reality is super tools that will ement our intelligence and wow I'm very optimistic for that I like the idea of having super human not not a super AI as as alone but AI that is under the supervision and the control of human intelligence it can change it is still the case in in the industry as well when you come to to Refinery I was working not long ago in in Eastern Europe we Refinery last month and they're looking at ways to identify specific adjustments to the protocol and the way they they're producing a specific chemical and and some augmented uh reality and some really good algorithm AI are able to find some new ways to do things but it's because and and I want to Echo a little bit what Sam is saying it's because there's a guy behind programed this who is extremely creative and he's already putting some seed on safety we need to look in this direction no the guy doesn't have enough time to do all these looks he doesn't have enough time to go through all these data but guiding a proper algorithm to go and do all these churning of data which is millions of data of course millions of line 20,000 tags that are immediately integrated in one second no one can do that but you can tell them where to look and they will find something so surprisingly sometimes that you didn't expect the outcome but it's exactly the same again if you don't have a good programmer or someone which is smart enough and understand the process well enough to guide this machine learning algorithm it's going to do something which is be will be irrelevant because the Insight will be not necessarily something you want but certainly something that's going to lead to no production or no extra extra production or Advantage for the people so you need the experts you need the human intelligence beiring artificial intelligence today you still do I haven't seen any artificial intelligence today that produce something that okay we didn't know that it made it go very fast May made it to a conclusion much much quicker but it it didn't really say when Chad gbt pulls me a paragraph together it's nicely written it's good 85 90% that makes the that makes sense but it's not something that said I could not have produced it if I work hard for 20 minutes so that's an example yeah and I think you have to think about the business problems that people are trying to solve where you do need a business use a business case and business justification and then use cases and then objectives and then clear outcomes and so on versus the creative arts where there's a business behind that of course but you're also looking at what can these tools be used for in the hands of the creative uh developers the artists and I also had a cover ation with the president of APM music which is a Lor of uh music for commercials and movies and so on and he was telling me a story about a young uh guy who had created just a very clever derivatives of an existing character and and he thought that they were he was going to get in trouble by using this and but the director actually saw what he had done and said oh that's really clever why don't we get together and then that kind of launches his career right so it's doing things that are unexpected and doing things that you couldn't do before but it also calls out the need to think about derivative work so when we start looking at creative works first of all large language models are great for starting point and starting research doing outlines even doing CP development or whatever that might be in the motion picture industry but in the in the music industry you're starting to see generative AI create and M llms create emulate certain artists like famous Drake song that was made up completely but with the style of Drake there's a fingerprint of what dress is Right In other words there's a genome there's actually a music Genome Project that kind of breaks down music into its different components and you can say how does that dress how is it how does it inform the model and then what the model creates is going to have those characteristics so these were all derivative works and Del derivative works also need cop right permissions right now it is the wild west where the AI companies have consumed all of this content without necessarily respecting copyrights so when people are doing things in the voice of a particular author or particular writer a singer or whoever it is those are derivative works right and that's where we have to have some mechanism to chain together the original work and then what comes out of that work even if it's through a trained model so there's a lot of really interesting thny details that we have to look at but from a p perspective one of the things I was that this guy was saying to me is gu generated content doesn't have a heart right it doesn't have but I think in year in coming years we're not going to be able to to tell that necessarily and it will still be the voice of the AER it will still be the voice of the person who is applying these tools using their creative energies and their creative thinking so I agree that it accelerates stuff I think that some of the the directors and some of the filmmakers who were creating the Sora Sora images and videos were saying this allowed me to do something I could never have done before some of them were pretty creepy but at the same time they were looking at this as a valuable tool here I think that moves us into the regulatory conversation around safety copyright regulatory all of that one of the things that just recently came out was the figure one robot from open AI where somebody's interacting with it and telling do a few tasks and can you give me something I can eat and on tray it identifies what is edible which was an apple and hands it to the human so are we getting closer to artificial general intelligence and if we are do we need regulation and safety standards in place so I think myself we are very far from AI I like a lot Sam Alman when he talks about AI he believes he will put it out in summer I will tell him good luuck with that and I would rather do other approach where we can obtain a lot from generative Ai and obtain even super intelligence level but not with a itself with combining human going back to regation is very important in order to be transparent with people we need regulations at two level first level at the data level we want to know where this large language model gots it's chaining dat it's very important to be transparent about it and the second level is algorithmic trp we want to know that their algorithm is respecting certain kind of fools now here come the tricky part because they want to be magician who will tell you hey W this is our large language well look what it does do but they will let us how it's doing that and sometimes they're trying to kick us by we just change the model and this is the out but it's not the truth they have a generative algorithm kind that it's just not algorithm so both side need regulations but we need to be careful also not to over regulate everybody was trying these days playing with generative and see okay I cannot say that this is not appropriate and for fun everyone can break the the AI to say it at the end there was a lot of videos separating the internet how you can make CH pronounce words that in the beginning it will tell you I cannot talk about it it's not appropriate so going back to regulation it's need to be a partnership between three parts of course The Regulators of usually they don't understand anything the industry who is programming and preparing the L language models and the tools but also the public need to have a say of this they need to see okay I don't want the large language model to filter I don't know a word like male organ I'm doing research on male organ I want to make a drug about it so when I'm using the large language what is not appropriate what's going on I'm not trying to do anything inappropriate I'm I'm trying to make that large language model go and analyze papers for me about this this Stu so it's a very complex but very important and topic regulations and it can be taken at various level and here our roles as industrial implicated people is to give our opinion and to tell Regulators please help us Dr please don't overdo it it's it's an interesting one but it it hasn't reached yet some of the Industrial Level because the regulation doesn't come within the companies companies are regulated by themselves and they do not let this go in there are some companies where they do not let their the the people their staff to go on Char GPT from the computer of the company because it's intellectual property going in because of course we don't know what CH GPT is doing with these questions and with these answers so if you are in a oil company and you're trying to look at some a very specific oil field or a specific resource and you asked on tgpt of store Indonesia on this block I found some o I want this I want that boom immediately there's a Le someone know onch GPT who gets the data that en the national company in of Italy has found some oil in there so that could leak out that could be used of different P so a lot of company are very worried of their employees using Char gbt because they don't know what those question will be it can yes enhance their work but the regulation in there is that you cannot use the company computer for specific questions about your work when it comes to the industry it's very similar there are some protection of of Ip or coduct there's patents so all these things are very I would say strictly regulated so charp doesn't go past that that firewall now it's of course those regulation open a brand new Leora of small satellite company that it takes inide GPT and are installing it behind the firew world within large companies and are doing elastic search and they're doing the proper Char within the old company that doesn't go out on the worldwide web but is behind this and and this is a brand new I would say opportunity for people that really know a specific industry would be be farma or supply chain to take CH GPT and adapt it to that very specific industry farmer or food industry or something and make a TP which is dedicated to this which has been specifically implemented for specific customer so regulation I think is it's not yet uh in inside of the companies they just say let's not look at it right now I don't want to use it they will come to regulate but it's iing the next yeah I think one of the things you're referring to before was a retrieval augmented generation keeping a local Lang language model and then using it for and then retrieving content from corporate data sources and knowledge bases and that reduces hallucinations and also protects IP there are some settings that that you can uh use to control whether your uh content is used to train the model some people don't trust that yet but it also you're supposed to use the API if you use the API levels you have more protection but a localized model is really the best way and as you say you can fine-tune that with your data than your content but the bigger picture around regulatory uh requirements I think there's going to be I'm actually speaking at a a regulatory uh conference on large language models but uh the whole idea is to say what do you need to do to keep your company from getting in trouble right and there's copyright issues but there's also transparency on how you're collecting the data ethical data collection removing bias and making sure you don't have bias in your data understanding what the process is that you're going to be applying large anguage model too and so all of that is going to still be as you say self-regulated or industry regulated as opposed to government regulat but when you start thinking about artificial general intelligence I think we are a long way from that I agree with that point but AGI has also been defined in some cases people say you're going to have narrow AGI narrow AGI is not AGI right so it's a contradiction in terms so you can say it's going to get better in these Nar domain Doms and then maybe you'll have a combination of Highly specific domains that are going to be chained together in some way but do we need regulation on AGI we need regulation on the data side on on transparency and privacy and personally identifiable information and that I think is well in process uh but on I don't think we're going to see AGI for quite a while and the human brain works in orders of magnitude greater complexity than anything we can do on Silicon we have 3 billion nerve cells and each one can be connected to 10,000 others and there's 100 neurotransmitters which are all uh analog right they can all vary they're not binary they're analog so there's a lot of complexity behind carry around with us in our heads but emulating that with AGI it's looking good in terms of what large language models can do in terms of emulation of what appears to be human intelligence but it is still a parate it's still probabilities and Mathematics in emulating human intelligence within the algorithms there is this is the concern of bias and fairness is as humans we have of course our opinions and our biases so how do we build these types of algorithms and large language models and not have them be biased have them be fair I think it's AOSS to not to be biased in this case because our knowledge and based knowledge as a human civilization is really bias we we have era where slavery was okay we have era where racism was okay I don't know we have terrible things in our past and pretend that we are not going to train our large language models on data on the Internet is hypocrisy because this is the biggest data set available that you can use and buas from the internet and then we can remove bias for algorithms good luck with that and also the refer the cultural reference like some stuff is accepted in certain cultures that is not accepted in other culture so here we move from buyers to civilization flash what are you going to do about so are you going to tell what I will give very dumb example in India or cow is is sacred so what are you going to remove all cow reference from the US internet when we eat cows and Stakes so it does not make sense so at the end bias will be always here what we need really is a human supervision we don't need to throw the bias and the problems on the machine and algorithms and the daa we need always have human supervision we need to know who is the human behind the comic show you mentioned in the beginning of the arode we need to know who is this guy because it's easy to say it's machine bias it's algorithmic bias and not to attach names but at the end of the day we know all there is a human behind the curtain yes for controlling the bias AR regulation and putting things behind it but the real approach to put human name always behind any algorithm or any it's called a tool and it doesn't need to be one person could be I don't know a te people need to assume their responsibilities in programming such sophisticated tools they don't need okay I did it but it's magical it's it can do things by itself this is too easy to do and by the way this is not really new people at the beginning of computers they used to go and program EXT system let's ask computer was not the time let's ask the computer whatever happens it's the computer mistake but they didn't fool us and people come again and say no we need responsibility we need verication for this software we need control to see okay if there is a bug who is responsible about the bug what's the limitation what's the problem I think it's appropriate to assign heal to any AI tool and bias is it's more complex than you think there was I went to a conference and they were talking about Talent retention for engineering company where I work the the P that they were trying to emulate diversity within the companies and therefore they were trying to have some blank CV so CV with no name because of course let's say you are in France and you want to Pure wi TV uh sometimes the the the first name Mohammad will very clearly say that this person is from Arabic background or North African backgrounds of course if you call William or if you call in US you would be called Washington there's a very strong likelihood that you have you're coming from afroamerican background so all these things are there so they tried to create an algorithm that was just eliminating the names from the series and what they were taking they were basically looking at other factors but the other factors were taking as well where you were located well where you were based and where your home was and there was a bias that was inherent to the home there because you could see on the let's say you live in Houston Texas and you're on the east side of Houston it's not as nice as if you live in the west irland area so you would already have some different in categories by knowing some people that went to this University or this one so if you really want to remove the buyers you need to remove the names of the University you went to you need to remove the place where you living you need to remove your age or because you can even if you don't put your age by looking at the CD you would say oh this person is above 60 or this person is about 50 so there's already a bias so to eliminate own bias there going to be AB extremely difficult and I totally agree we can't do that we can't eliminate all bias but we need to have sensible persons looking at it looking at the algorith and looking at the result of the algorith and see is there a bias or no or do I go through or not in India you're talking about the cast I was in India and I was trying to hire people for a local company in India and my supervisor told me I said that's the best CV oh you can't take this why look at his name he's from a cast which is above the cost of his supervisor he will never work for this guy let say it doesn't matter the guy got 15 years of chemical engineering background is brilliant his CV is tip Tok it's exactly what we need it's never going to work so even if you want to eliminate bias in that case you will not because it's not going to function so again sometimes the bias is there for a reason as well because it helps compy to function because if you hire this guy even with the best intentions in the world it will never work so again that's an example and maybe you understand that example some because you you were talking about India but yes there is bias that exist in there and you it's you're never going to take this bias from the cow yes remove all the cows from streams on in on internet and this no because people have different values yeah if you think about I I totally agree with that because different values and different ways of thinking are are biased because they're shortcuts they're shortcuts to understanding the world so So when you say bias what's the definition bias context and perspective because many times if you have a particular perspective yes you're biased in that this is your frame of of of of view this is your point of view this is your frame of reference and what that means is you're making certain assumptions about the world and it's like why are stereotypes stereotypes because they fit many times and I know there's all sorts of political sensitivities about about all that stuff but we are sensemaking machines and by by making sense of the world we're taking shortcuts and those shortcuts really are about judgments and judgments are biased and so when it comes to data and making sure your data isn't biased it's Fitness to purpose right it's what are we trying to use the data for and do we find that it's influencing the results in a way that is unintended so I think part of it is transparency part of it is being ethical and collection of that that data but it's also so looking at what are you trying to do with the data and what does the data contain in terms of context or perspective right and be because those things can become bias and completely agree with the cultural aspect right what is cultural bias it's how people look at the world it's how people think about things and we take shortcuts because it's they're always processing lots of information and we need to make judgments and and take shortcuts and that is bias yeah to have ai without bias is almost you're not building replic of a human brain yeah I think it's hard to think about um all those algorithms have to make assumptions in some way they're looking for patterns and within patterns there will be some bias so if any of our listeners wanted to get in touch learn more about your Ventures get your books how could they do I invite people to go to san.com S m.com uh there is a link for my book Singularity of hope uh The Singularity of hope it's all about balance and putting AI in its place it's a formidable machine not a replacement for our brain and our human Okay the other thing also our tool is coming soon in the and it will be the first AI tool that would be using human dentation as a base rather than the magic you can find me on my on my website OSS I4 meds and I usually put some of my projects in there but otherwise on LinkedIn I try to be active on LinkedIn I find so many very interesting interesting publication and I think it's Linkin has completely changed for the last for the last two years in terms of quality of what you can get and and and contact with the people so I'm always happy to meet people again the industrial world but it's it is it is a very fascinating place because we need to be extremely careful with AI and machine learning because again this is a place where it's not like the entertainment in entertainment yes there can be consequence if you fake the profile of some which you make a deep fake if you make so yes there are consequences but certainly not the same consequences when you just put a Refinery at a stock for 6 hours or 20 hours and it cost you $50 million again we see this in a very different way and yes it's not for entertaining but I still believe that a ml can break enough not today and we're not ready we're not ready for it there's still coming and everybody's going to be surprised Chad GPT 5 will be will be again something that everybody's excited about and maybe it's going to be contextualization between different industry there will be some changes we're looking at all the the new stuff coming but again it's it's very slow to adapt in our industry much slower people could contact me at south.com and it's e a r l e y don't forget the E before the Y and the website is early.com again EA r l e y and you can also find me on LinkedIn I'm just Seth early so as always as you can remember how to spell the last name you can find me Seth early.com or Seth early or early.com thank you I want to thank our AI expert guests for coming on today and everybody for listening to another episode of failing to success if you like the show make sure to subscribe we'll see you next time [Music]