Scale By The Bay 2020: Panel Discussion: Will AI Kill Programming?

Scale By The Bay 2020: Panel Discussion: Will AI Kill Programming?

Recording: Scale By The Bay 2020: Panel Discussion: Will AI Kill Programming?

[Music] um hey folks um thank you for joining us this is uh i don't know like say the first panel of uh the session or kind of uh we're the first awesome so this is where the experiments uh begin and we'll hopefully you know the people on after us will learn from this experience and know what to do better but this is the first time for i think all of us and yeah so this is my third time off kind of a hosting uh this uh debate um and i've done it kind of on previous occasions and we have you know four wonderful debaters uh here with with us and the motion that we will be debating um is the following is will ai kill programming but more specifically what we will be debating is the falling motion is sorry is er yeah let me find it be prepare is ai will become essential to programming over the next decade as critical to software engineering as development in source code control continuous integration and programming languages have been over the past three so we have two teams of debaters with uh brian cantrell and helena edelson arguing the side against and omar alonso and jin linetsky i'm sorry if i butchered anyone's name apologies um yeah we'll be arguing for the fourth side so um instead of maybe um kind of me introducing every one of them before you start and we'll have like five minutes on each side to kind of start with the opening statement maybe start by 30-second introduction uh of uh you know who you are and from that um then gonna continue uh into her points so i'll just introduce myself and like i said vitaly goren i'm the ceo of a company named ferris ai we're um basically you know have a slight horse in the race and it says we're trying to use ai to automate a lot of software engineering tasks but um i will be trying to be as neutral as possible um in this regard um also just a kind of a note two notes one is our wonderful debaters are arguing the points but please you know don't hold it personally against them we just kind of randomly split it into teams where they you know it's for the sport so they don't necessarily you know represent uh very strongly the point that they will be debating about and the second this will be an oxford style debate which means we will take a vote at the beginning and alexi just posted the link on this course so please go in and vote so we will uh tally the votes at the beginning and then we will do another set of voting at the end at a second link that alexey will post and the winning team is the team that was able to swing the most votes and you know from this gene i believe you are the first debater arguing for the fourth side so please introduce yourself um and kind of you know start a promotion and you have kind of five minutes on the clock and starting now thank you thank you vitaly thanks everyone who's watching and who's participating my name is jean i'm the cto of the company called true accord it's the first company in this country definitely or maybe in the world and that is redefining the what it means to collect debt what it means to help consumers close their late payments late debts we're using the combination of machine learning and the personal conversations with our agents who never collect never collect on consumers just helping them get out of difficult financial situations in uh in the last millennium i was just straightforward coder first in assembly then in c plus plus and starting with this millennium i started doing startups did them for a while one of them was actually with alexi of all people and right now i'm the cto of this company of for almost a year and it's definitely the best gig of my career so on the motion will ai kill programming obviously without strict definitions of what it means to kill something what what it means to program things i would say that um the even the definition of what it means to code what it means to program had changed so dramatically over the last let's say 30 years that i've been doing it it used to be that people coded things pretty much from scratch and and even before that people coded things in assembly with no benefit of accessing anybody else's code with no benefit of reusing somebody's building blocks to build build their solutions the biggest innovation i think in that direction was first the symbolic languages and then almost everyone is coding in symbolic languages right now although there are still people who are coding coding in assembly very close to metal and then what i'm seeing and and maybe your perception is different what i'm seeing in the last 10 years is that the the type of activity that the typical coder typical programmer is engaged in every every day is very different from what it was even 15 years ago mostly now coders build glue between the existing components to build a working system today means something entirely different from what it meant 15 or 20 years ago and so in that same sense i think what it will mean to build the working system 10 years from now will be very different and uh the biggest difference will come not from symbolic languages as it did before but in reliance on more and more of the rules to be produced automatically by intelligent systems i would say that for several companies for many companies that we all know who they are that that time had already came you wouldn't imagine facebook um employing a million engineers to to write rule if then rules that govern every single um inclusion of every single post on on every single person's feed right we have two billion people interacting with those systems billions times times a second every single day and um i think this is really the terminal state of any cloud company for lack of a better definition it will have some kind of a black box that is learning much faster than any human or any group of humans can learn about their particular industry and coders will be relegated to writing code that connects that black box to the outside world because the outside world is not yet fully up to speed with with let's say automatic processing of stuff and so we will still need people to connect the brain of that company to the outside world by what i would say for channels data coming in data coming out money coming in money coming out and so um in that sense it will definitely kill most of the programming that is previously would go into that central brain of any company that's my thesis okay and right on time gene thank you and our next speaker to go arguing against is brian cantrell from the outside computing company so brian please introduce yourself and you know let jin have it hey everyone i'm brian cantrell i'm the cto of the oxide computer company um we actually still do program in assembly um but not just assembly obviously higher level languages as well we in particular doing a lot of things in rust uh and i am just to be clear um ai is not my domain um i have been an ai malcontent and conscientious objector for my entire 30-year career um but the i have done systems programming a lot of systems programming and i've done a lot of systems debugging and i've seen as as part of that how frail our systems are these are mathematical systems and a a single defect in a complicated software system can induce really outsized pathologies these systems need to be exactly correct they're very finicky it is garbage in and garbage out that still remains true i think as we talk about how can ai effect programming will ai become as essential to the programmer as things like source code control um and these other innovations i think the term we actually most need to define um is ai itself um ai is a terrible term it has always been a terrible term um and i think that we are much better off thinking not of artificial intelligence which immediately thinks of us to to think of the computer as being like a human and much more in terms of automated pattern recognition that is what what we think of ai is actually automated pattern recognition and if you substitute automated pattern recognition into the sentence and the sentence does not make sense then it's probably not going to be true so i think that what we're actually debating here is will automated pattern recognition become as essential to the programmer as source code control continuous integration and so on these other developments and i think automated pattern recognition can be helpful but i think when you think of it that way you immediately begin to think of it not as replacing the task of programming but as dealing with those particularly data intensive aspects of programming there do exist data intensive aspects of programming but in my experience they are few and far between and certainly there is a plethora of data that exists out of our systems that we don't actually really process this kind of internet exhaust that we don't really meaningfully do anything with from a programmatic perspective and maybe we would do some things um with that but as i look at the systems that i develop the questions that i have are often much more precise than automated pattern recognition would afford um i am not doing the kinds of things um that certainly when i'm in the act of programming i'm in the act of cutting code it's very high consequence it has to be correct it can't be approximate this is not identifying a picture of a hot dog where the errors are are not necessarily a problem i also when i am writing code i need to have my own explainability i need to be able to explain why i'm doing this and the current state of the art in automated pattern recognition is very little explainability um now this is becoming i think increasingly a problem historically when you're doing image recognition um which something that ai has absolutely excelled at its ability to misidentify an image has not necessarily been high consequence that is becoming increasingly high consequence as time goes on but we don't really have the ability to understand why did you get this one but missed this one when we are programming those systems are so precise we really demand that explainability and that is not something that ai is really on a trajectory to deliver so i i from from my perspective um where we can expect um ai to actually impact programming it is where there are patterns to be meaningfully recognized where there are large data sets that can be trained and so on and i simply don't see that in most of what we call programming today so i do not think it is going to have this kind of outsized impact awesome and right in time thank you brian and the next speaker to argue for is omar alonso um just before you begin omar um please if you have questions that you want to submit at the next the next round after the all the speakers are done arguing their point they will be using your question to ask them to clarify their positions and omar please go ahead introduce yourself and kind of argue your point hello folks thanks for watching my name is omar i work at instagram i just joined instagram a few weeks ago background is mostly information retrieval known as graphs and large scale labeling before that microsoft amazon and oracle all right so why i'm bullish about using ai for improving programming at large i think there's tremendous opportunities to build new tools that are going to augment programmers and there's going to partition the set of parameters who are very skilled and the ones that i think are going to disappear uh what i'm trying to do here what i'm trying to say here is that there are certain things that we can always improve and if you can build tools that recognize these patterns as you know brian was mentioning before if you think of managing a file you're going to read the file you're going to open the file do something and then write back if you're thinking in terms of a database you open a connection you lock a record you do something you you know release the log and so forth and the same with memory management you allocate a pointer do some stuff release the memory back to to the hip a lot of the bugs that you can see encoding today is because people do not follow these protocols right so i think there's tremendous opportunities for building tools that can learn the behavioral patterns of programmers not just looking at the code and see the behaviors and we know that as you know you know microsoft acquired git some time ago so there's plenty of data to build new tools so i'm really hopeful that the new set of ids are not going to just suggest you syntax but i think we're going to have better tools that can predict potential behavior common behavior and just hopefully help the developers why i think there's going to be a break in two type of developers because the ones that are well trained and no argos and all those things can benefit from these new set of tools and the ones that are just doing initial hacks are going to struggle because you may not be ready to use these sophisticated tools that you know require serious coding so yeah that's my perspective and now i think i'm more like just to summarize i'm more hopeful on tools that are going to augment developers the good developers and i think those tools are going to be a very high bar if you are not a very skilled programmer that's my argument okay and thank you um omar and kind of the last but not least arguing against the pro position is helena adelson um please introduce yourself and tell us why you don't believe that you know ai will kill programming okay hi everyone um helen adelson um quick intro about me i've been i started out mostly as a you know i've been a senior cloud engineer for years at various companies down silicon valley worked in many different domains uh cyber security cloud infrastructure um been a senior data distributed data engineer and platform engineering at apple and a vp of platform engineering at a startup and principal engineer at lightbed so i've seen a lot of different domains and applications of machine learning and ai and those different places and at different approaches and different experience levels of the teams and also very importantly different funding of those teams and the different uh importance of what those teams are doing and the the business plan of those teams so um so this is a really interesting topic um uh what i'm doing now is that i'm sort of bridging the i i thought i left tech it turns out i really didn't i'm kind of bridging academia and and tech right now i'm starting a startup but i'm also sort of part of the involved with the um oregon state university but working a lot with the endangered species and conservation groups working on issues related to climate change and things like that so i'm looking a lot very different types of data very different constraints um and how to pull this all together taking what we all know how to do in tech and sort of layer everything with climatic data and species data and image data like hundreds of thousands of images of species and trying to identify the species and identify the individuals and you know all this um sort of ai for good um and so how i feel about this i think well first i have to thank my esteemed colleague and teammate brian excellent totally agree with everything that you said that was really well put um i also see ai largely as an optimization tool um particularly from where i'm coming from now which is sort of an and to what brian was talking about um i think we can all agree that ai helps us do more with less and so a lot of what i'm dealing with is a collaboration with scientists for example where i see well first of all ai in terms of how it's its relationship to uh will it become essential to programming with regard to control continuous integration languages that's already happening so i see it kind of as a foregone conclusion for example microsoft has that as a service um seems pretty mainstream to me um essential i think also is a strong word i think caffeine is essential to programming and so since it's already in place i think what's really going to happen is a transformation of programming with ai and so what i mean by that is the boundary first of all between programming and ai is falling away as similar things have happened between sort of big data and data science and programming but really the the boundaries of all this is just disappearing and i can go into that a bit more in the bulk of what we're going to talk about but i mean it's data scientists and machine learning experts learning to code on researchers and scientists learning to code programmers learning to do data science and you know it's it's just all starting to blur but there very definitely is things that are specific to our domains where um a data scientist and a researcher out in the field is you know studying uh ice melt for example is not going to be an expert in scaling up or productionization of these systems but also it's going to be opening up when i say transformation i mean it's going to be really really interesting to see the scientific collaboration and the innovation around that between programmers working more and more with scientists to try and help with some of the larger more urgent issues and i think there's going to be more of a i think over time we're going to see more of an exodus of people like myself that are leaving industry and going to see what they can do how they can apply their skills to help solve some of these problems globally but i have to bring in this one last point which i think is sort of a reality check which is something i've seen year after year everywhere that is of money in funding um and for engineering for infrastructure like we're talking about the specific context to me i i think sounds like technical debt to a sales team and sales teams are largely drivers of budget allocation and resource allocation so unless they can sell it it's really not going to ever be important enough to be something that you're going to throw an engineering team on it's always going to be like hey we need to put our resources towards these new functions or what our customers want or you know how are we going to solve our scaling issues better it's always seeming to be that versus doing this infrastructure work and awesome uh thank you uh elena it's unfortunately there is or i just don't know how to kind of give you signals uh around time so we'll just have to jump in at certain points but this kind of a concludes our kind of first round of the debate where um the debaters kind of made their opening statements the next um kind of portion i'll give you some questions both from um myself and the audience but kind of just to recap we heard kind of gene and uh omar um kind of argued that really um we already seeing ai pretty much everywhere whether we kind of call it programming or not um is a is a moot point and if we think about programming as kind of a set of small deterministic tasks that humans tend to err on actually this is something that is suitable for machines and there are already plenty of examples on the other side um the team arguing against um mentioning that you know ai is really kind of an optimization tool it's more about pattern recognition and really the software engineers work is way beyond you know just taking these small determination tasks and you know kind of re repeating them at the intimate and there is a limitation of what can we achieve uh using kind of these ai techniques so maybe kind of a first question to go for the the team arguing again so part of the proposition is saying that ai is not necessarily will replace um all of the engineers but will be common essential uh tooling like we said memory management source control and so on so forth so why kind of don't you believe that maybe ai or you know systems that have kind of ai's integral part of them will not be kind of these auxiliary systems that are essential part of the kind of tool chain that engineers use on a daily basis so that's a question for the against side so yes the first swing at that um well so i think that and um my teammate um uh haunted made a very uh i think compelling argument about how uh more and more data intensive domains are going to be engaging in programming which is kind of the the inverse of the proposition here um the and i i think that i emphatically agree um i and i also especially agree with the point that more and more folks are looking beyond the traditional kind of tech domains and looking for larger impact they're going into the world looking at things like climate change and finding a data intensive world that they need to understand programmatically so they end up applying all sorts of pattern recognition tools and so on to that data intensive world but that's the inverse of the proposition here the proposition here is not that we are going to provide doing more and more programming in a data intensive world but rather we will be applying these techniques to the actual act of programming itself and what i just i do not see the big data in programming i don't see it um now admittedly on the systems that i write i am very concerned about their their resource utilization and so on um but i don't routinely process gigabytes and terabytes and petabytes of data as a programmer in the act of programming um i don't think you can call a google search on stack overflow big data that is not actually um and that is about the time when we use the most amount of data i don't see us doing that so um i think that i would kind of flip it around what when a programmer is programming what is the large data set that they are querying to actually engage in that active programming i don't see it yeah so gene and omar brian makes a compelling point uh sorry brian and elena where they argue that you know as we see an increase in increase in ai capabilities we actually see more and more demand for programmers like if really the proposition was correct and you know not to pull it towards your mouth we would expect to see the opposite um impact on the demand for programmers and priorities so how would you respond to this statement do you want me to start whoever wants to go first well i mean as we all remember 100 200 years ago in new york there were predictions about corset filling all the streets to the tops of all the buildings um so that's the equivalent of too many programmers and then the more and more programmers needed um i mean if we take the slightly longer view and acknowledge that um the physical system is capable of thinking the proof is in each of our heads and then we look at the density of is available today and notice that it's probably two or three orders of magnitudes away from the density that's realized by neurons in our heads the argument becomes if not if but when um to redefine ai as pattern recognition is is convenient for the argument obviously but look at gpt three look at alphago look at image generation it's it's way beyond pattern recognition and of course they're always going to be something like the swiss watch today there's a mechanical system where one one tiny mistake invalidates it but already the cheap quartz watch is is so superior functionally to to the 50 000 piece of jewelry i would say that um we will probably not even see the inflection point between the the need for more programmers to to deal with horses producing produced by all those horses 200 years ago in new york and the situation when um many people suddenly find themselves not not needed to code even the most mundane tasks so um i don't think it's a valid argument to redefine ai as something it's it's it's been maybe 100 years ago or 50 years ago it's so much more than that um maybe you don't know but the number of knowledge in in the in the space doubles every three months right now in artificial intelligence publications and then advances and new models were even five years ago it would double maybe every year the humanity is notoriously bad estimating exponential processes and i think we're just starting the inflection point so what looks to many people as a linear growth is is nothing but okay that's also an interesting point maybe helena you could respond let me kind of paraphrase a little bit of gene statements so and maybe provide a little bit of context we already seen demos from gpt free where people can input literally a phrase of i want you know a website with two text boxes and you know a red button and things like that and gpd3 goes and generates kind of the front end and you know some back-end kind of prototype so what makes you and kind of your team um have a high degree of confidence that you know 10 years from now which is kind of the time frame we're talking about uh we will not see this kind of you know simple demonstration to become actually something very real well i've not i have to caveat i have not worked with gpd3 myself but i think the danger we could get in here with these discussions is the oversimplification of this so a lot of the examples um are games complicated games with complicated decisions that it has to make but still simple or if we're talk we're talking about things like pattern recognition and yeah like you know spongebob or something i mean that is not actually how it goes when we're trying to do a large-scale image recognition project humans are largely involved in that they have to go through and pre-identify when you're trying to first put together the images to train the ai so that they can learn how to do the job itself and so you need programmers too or you know interns or whatever but you need humans to go in and first work with the data and teach it what it needs to learn but you know what i mean pre-allocate the what it's going to learn from sorry and um and i my whole um premise for this of saying no is that it's it's not going to kill programming it's not going to take away the jobs it's just going to be more of a transformation on what we're doing as programmers awesome and maybe kind of back to you omar one also point that was made and kind of uh by the team arguing against and kind of appeared in the question is um really if you look at the programming job today if anything we go into more and more kind of higher depth of understanding what is actually going on if you look at the kind of the observability kind of movement that has been happening in the last uh several years and you know now we have a company like datadog worth like 20 billion dollars people really want to understand every single bit of buy while on the other hand ai kind of takes us in almost like the opposite direction where we now have these extremely complex black boxes that you know generate output but no one you know no human can say understand what actually is happening between so how can we actually replace programmers where you know it seems like we're trying to get a you know a higher degree of confidence around how our system operates while it seems that we actually kind of losing and losing more ground to understanding you know how ai works uh a number of observations here the first one is i want to address the the issue that why we're getting more parameters instead of less parameters first and one of the reasons is i think we're doing more computationally thinking if i want to borrow a phrase from jeanette wing where there's a lot of problems today that can be solved by just perming um this is not just for computer scientists like ourselves but for a lot of other disciplines as well so the use of a machine to code to solve a problem is pretty much mundane anybody now stakes programming class because you have to solve a particular problem so foreigners the number of primaries actually increases that doesn't mean that you know ai is not going to help you or there's no you know good act techniques that we can use today explainability well again depends on the kind of computation only thinking problem that you're you're using there are certain problems that you have to explain there are other problems that are going to be more difficult to explain hardcore you know ml your network's very hard to explain others you have to explain if you're going to build an autopilot which is basically ai for flying your plane you know pretty critical that you explain what's happening on the last point i want to make which i think elena was saying that yes there's always a need for a human that's just not for just labeling your images but there's a lot of techniques that you need a human to actually produce the computation first so then the machine can learn probably one of the biggest examples and this is not ai but well we want to call it ai but if you think of like parser generations like you know yuck the famous tool this were done because someone actually wrote one and then they wrote a prompt that actually writes that thing so it's kind of two levels kind of meta right and those things are happening so the ability to generate code not just as a compiler but also you know configuration files and little languages are all examples of different type of ai that you need to acknowledge you absorb you synthetize the knowledge and then you write a program that actually behaves like a human so from that it's pretty much everywhere well awesome and you know this concludes kind of our second round of this debate uh where um kind of the panelists answered the question both kind of for me and the question that was uh were posed by you the audience we're going to switch into the kind of third and final stage of the debate where um the debaters will kind of make their closing statements alexi if you can please uh share the link for the voting um so while people kind of um you know making their statements people if people already decided that kind of one way or another they can start start voting in case we have some technical issues around it and let's uh start the opposite way this time with the team kind of arguing against brian maybe start off with you know how you want to conclude your position on it and here is a kind of a a tip for all of you folks your statement should end with something along the lines with and this is why you should vote for or against the proposition so go ahead brian i'd like to start can i start sorry go ahead talk about that black box idea that really struck me i mean realistically you know first of all i have to say that this is a really hard thing to do because i'm able to argue both sides and so you know it's uh it's very touching but um on the black box idea first of all think you know who do we think is writing the box and writing the code that's inside that box and who do we think is writing every single logic change that has to be rolled to that box tested and deployed to boxes all over the world um and so what was the closing we're supposed to use that tag line and that is why i think you should vote uh no okay awesome and uh who wants to go between omar jin to kind of if you're closing remarks first go i go first um i think for for developers in general i think the tools have gotten better uh your ids now have auto complete they will just highlight the changes that will tell you not just their sin syntax bugs but also a few other things there's a lot of accumulated knowledge not only on obviously stack overflow and core and all those places but also in source code and there's a lot of potentials here for new tools that are going to help you not alone in terms of productivity but also do the important things that nobody wants to do and i think the machines are going to be great for that so remember to vote yes ai will dominate awesome and uh brian um how about you going next well i i think in those arguing for this we have seen the the classic trap of ai for time immemorial which is uh because our brains are physical stuff and we clearly have uh thought and intelligence and computers are physical things we must be able to make computers intelligent and it ignores the fact that the computational model is entirely different between the the silicon and our actual brains we look at the we have had since the the 70s we've gone from thousands of transistors on a die to tens of billions and over that time we have not fundamentally changed the computational model computers do loads and they do stores and they do branches that is what they do it's a von neumann model and the the the tricks that we have pulled which are very impressive around automated pattern recognition are exactly that they have not made us more intelligent and helen made a very important point about the hidden humans behind these systems these are the humans that actually tag data sets if we do not tag data sets we the computer actually doesn't have the capacity to reason about it it is not intelligent and if the data set is not tagged it doesn't know how to identify it if you have a toddler if you've had children and you watch the way they learn you know how fundamentally different it is we do not teach a two-year-old what a cat is by showing it 10 million images of different cats a two-year-old cat around the house will be able to identify a cat that they have never seen before that looks different from their cat but is clearly a cat a two-year-old is able to do this because the computational model in our brains is entirely different we are not on a trajectory to do that with silicon and ai continues by the by the arrogance of its name it continues to deceive the general public about its capacity we are not building intelligence we are building automated pattern recognition that automated pattern recognition is important but it depends ultimately on humans to see those data sets and i don't see that happening in the act of programming itself the act of programming itself is it like like math like theorem proving like writing is an act of honestly of extraordinary intelligence not to not overly raise us in the room but programming is hard it is essentially human in its endeavor and i do not see automated pattern recognition assisting or becoming essential in the act of programming um in the next right and this is why and this is why you should be against the proposition awesome and wrapping up the closing statements uh jin all right well i'm i'm really glad that you mentioned the baby um if the baby didn't have a family it was raised by wolf wolves it will never recognize the cat in fact pointing out that the computational model at the base of modern computers is not the same as for human brains is really mixing up the level levels of abstractions it's like saying look at neurons you will never be able to explain human thought if you just count the in the impulses that the the neurons send to one another in fact the way we're teaching intelligent systems today to do what they are capable of doing or maybe in some cases we make them teach themselves like in case of alphago is very analogous to the way a human baby is growing and the role of those programmers who need to code the initial data set is the previous evolution that created the hardware inside their brains that is capable of being taught in the in the gestation period of human mind which starts before birth and ends in some cases at 30 years old or older sooner or later the the computational intensity of any physical system will be sufficient for the emergence of thought what you said the very high intelligence all that's missing really in that area is goal setting and and the um i would say the survival goal of the system as soon as that as soon as that as soon as we teach the system some overarching goal and let it develop and absorb the information from the outside world it will definitely become as intelligent or more intelligent if it's possible actually than humans or any other life form and that is why you have to vote for our proposition well awesome uh thank you folks uh i really enjoyed the debate here you know and thank you so much for you know making the most out of it and these kind of unusual circumstances so i wish there was a live audience and we can kind of get a big round of applause but i guess we'll just have to imagine uh that you know it happened um it is happening so now we're gonna kind of tally the vote and i'll share with you uh so every hey everyone who hasn't voted uh yet uh please uh please vote for uh you know your position uh again i don't know if it's too soon but we won't have absentee voting in this um kind of in this debate so please make your voices heard so um kind of will ai kill programming at the beginning there were 39 responders and actually 82.1 percent um answered no so it's clearly kind of a highly technical uh kind of audience that is not necessarily my kind of my guess that is in an ai conference the the results would be somewhat different but here i see where people kind of um can see they were gone so the four team had really you know the work cut out for them to kind of try to swing some of these votes um but if we see here and there are still votes coming so i'll give it to you like it's actually see it like updating live and it's actually fairly close so i don't want to uh i'll let people to kind of finish out recounting yeah yeah finish the voting uh um yeah so you know we want to have like the msnbc you know the number guy yeah we need to worry about it so you know by the way if anyone uh wants to reserve the four season total landscaping uh for you know making some statements about uh you know how the vote should have been then maybe that's a good time to give them a call because i think they're probably very popular so let me i'll put a timer in kind of kind of one minute and then i'll stop uh i guess you know we're at furry responses now it's oh my god so it's uh 40 responses already um so please keep them coming i'll put a timer 60 seconds and after this timer whatever the result is i'm gonna read it um it's uh so by the way congrats to you know both uh both teams who probably did kind of great job on um making uh making the topic interesting and maybe in the second 60 seconds um who wants to kind of maybe how uh how was it for you folks did you enjoy the the format it was good awesome yeah i think it was hard especially like if you could argue both sides and the way i don't know if anyone everyone realizes that some of us argue um you know not necessarily how we feel strongest so it's really it was pretty fun yeah um some of us are unable to argue both sides yeah let me guess let me guess your reviews myself though so actually i don't know if i can share maybe my screen and actually like show you so we're done like people not get accused of kind of vote so this is the this is the before vote and now my timer just rang so uh can you see my screen here or not so this is before what 82.5 i guess some people added the before vote even after um this and the way we ended at 49 uh sorry this is the oh wow so folks i think it will be only fair if we conclude it as a you know maybe kind of a draw it's like actually like it it is within people oh i guess people are now seeing the screen and adding more but you know what um you know we can now split hair and claim that this you know one percent but i think if you know we can call it a as a first time in scale by the bay something that is coming close to a draw or you know people coming with vote after but like i said and we put a timer it was like literally less than one percent uh change on each side so i'd like to thank again everyone who participated in it folks you've done you've done a great job i enjoyed it greatly and alexi would you like to add kind of final words yes first of all uh i must report certain voting irregularities we detected certain suspicious patterns so what happened is uh uh i kept track of everything right so after so so before it's supposed to be really before we we we asked folks not to use the panel inputs uh uh to sway you know to to to change their state so when vitaly and i don't know it was a voting mistake or not but basically the original vote was 77 no roughly 23 yes right so after the most powerful statements uh for the no side i would say uh we have about five percent so so you you what what happened is when the after what was declared the counts started going on both so you might argue depending how you see it right maybe people are recording the original impressions from the best of the ability uh going back in the time machine uh but there is a you can interpret it as a slight uh change in favor of no party right so so i would say that that's one way to interpret it uh but again uh it is pretty close right and and uh uh yeah it's it and it's actually now going to almost 83 percent uh the good thing is that we have about 48 votes after versus 40 votes before so we certainly increased the uh voter participation which is a great achievement for our democracy yeah so i would say that people's minds are already made up and both teams did a good job throwing red and meat to their base yes that is probably you know a good way to summarize and i have to say that also since kind of somehow the there was kind of this weird part where at some point uh actually the red part was which represents the no and again this is uh not to draw any political conclusion this is how the google forms decided to color it it was 100 and then there was like this then there was this blue wave that you know came in and you know swept the votes to the other side which was kind of a little bit um you know too soon and kind of too close and uh like i think now people will start seeing uh these patterns um everywhere but yeah i had to like it was kind of so funny as there was like 100 percent red and then like the absentee votes started coming yeah the million votes arrived and the blue section started growing and growing growing uh but i think still uh like alex said maybe not enough to actually kind of swing but if yeah if the original one was 77 i think now people uh kind of uh relax on 84 with 50 responses so again both teams did a great job probably like one of the things when you do it in live you can actually see the difference between the absentees as well like because usually large portion of the room does not kind of vote and then you can start seeing here it's hard to say how many votes have been actually swung but there is at least 10 more votes that didn't appear in the initial vote but the problem with them they kind of fell very similar to the uh original voting um awesome so you know why this is happening right the uis ai is smart enough to keep most of humanity in the dark yes um so yes alexis what's next uh so uh so first of all uh we have q a in this format right so after we are done here there is a link uh double click it yet but there is a link in the zoom chat if you click on it you will get into another zoom which is open to the public this is the room for the panelists and speakers which is not open to the public but after the panels we have an open zoom so you are all welcome to go there one thing i wanted to ask you vitali if you want to read some of the final thoughts because the the vote forms we shared um the before forum had a question for the for the panelists and the after form had some final thoughts i wonder if you if you want to read yeah so let me read them sorry i didn't realize that these are kind of final thoughts yeah they're our final thoughts anyway so um kind of very quickly so will ai be trusted and actually used by humans question mark if they're aware uh open source software is an example requires several skilled programs to keep it usable and desirable my argument is that humans will not accept in parenthesis knowingly ai generate machines slash codes so it will not kill programming then someone said i'd like to understand how ai how is capitalized ai can assist us not the super fetch sci-fi idea that's uh interesting i wonder which side this person was on and but it will definitely assist programming that is something um another comment then there was another comment i expect to see some displacement programming jobs in the next 10 years i expect to see an order of magnet more displacement after the stock that's uh you know very interesting i think developers will figure out how to make ai explain and slash reasons behind its decisions baby learning can be compared with reinforcement learning techniques then there is a it will change the nature of programming in some areas but not kill it and in parenthesis although gene does make a convincing argument so that's our first call by name so good job gene and the question is to binary i'm uh well we have two teams debating so i'm not sure how we could have like improve improve it by making it less binary and the other it will kill some jobs but will create more and the you know evergreen anti-lodding argument uh there as well um and then someone said but it will take a long long time and that's it and then the final uh i guess now in the coffin is humans will die first that's uh somewhat a morbid view um on humanity but you know who knows with covet and all that then it might take a decade but i don't see a reason why not and i think yes i see there is less common simple things can be can be done made by ml but simple programs can be written as well yeah it's kind of um unfortunately it would be nice if these were tied into the votes but uh oh i guess yeah i i can't see them individually but it's kind of well i guess google is not very known for their top ux chops so it's very hard to kind of triangulate with the comments to how people voted but you heard here so now we're going to this full zoom inception alexi and going correct let me just read one last question from discord which i think is a very good angle on this vote it basically says the world just confirms that 80 percent of the people need to work harder on making the ikea programming so with that let's proceed to the q a let's just click the link in the chat [Music] you