[00:00.000 --> 00:05.060]  I will tell you about this experience I did last year named my presidential Twitter bot.
[00:07.640 --> 00:12.420]  In the mid-90s, I was working for a firm in Buenos Aires doing computer programming.
[00:12.760 --> 00:18.980]  I was in charge of a content management system for a woman's news website.
[00:19.220 --> 00:20.080]  Not a good one.
[00:20.760 --> 00:29.500]  All the articles were like finding your Prince Charming and summer diets, winter diets, I don't know, sex move, stuff like that.
[00:30.420 --> 00:33.000]  And we all work in the same place.
[00:33.680 --> 00:36.960]  The journalists and the people in the computer department.
[00:37.200 --> 00:39.840]  And I think this triggers some kind of reaction on me.
[00:40.040 --> 00:40.940]  Not a good one.
[00:41.440 --> 00:45.720]  Because I started to work in this side project of becoming a porn millionaire.
[00:47.560 --> 00:49.220]  So I started to work in this website.
[00:50.520 --> 00:53.760]  The idea was a front page with different women to choose.
[00:54.000 --> 00:55.980]  Each woman with different features.
[00:56.280 --> 00:57.100]  Let's put it that way.
[00:57.880 --> 00:59.820]  So you choose a woman.
[01:00.760 --> 01:02.980]  Then you pay with your credit card.
[01:03.820 --> 01:06.400]  And then you start to have a conversation.
[01:07.760 --> 01:11.700]  If you say the right things, the woman will take her clothing off.
[01:12.080 --> 01:12.260]  So...
[01:12.820 --> 01:14.200]  And you know how it goes.
[01:15.520 --> 01:18.020]  That was the use case and business plan two in one.
[01:19.620 --> 01:20.900]  But I had a trouble.
[01:21.100 --> 01:23.260]  I didn't know where to find those women.
[01:24.020 --> 01:26.440]  And worse than that, I was a little short on money.
[01:27.660 --> 01:30.080]  What I did have was programming skills.
[01:30.320 --> 01:32.560]  So I decided to make a chatbot.
[01:33.220 --> 01:39.080]  If you don't know what a chatbot is, a chatbot is an application designed to conduct a conversation.
[01:39.080 --> 01:43.980]  Maybe the most famous one was done at 1966 MIT.
[01:45.220 --> 01:47.260]  Eliza, the computer therapist.
[01:47.760 --> 01:48.900]  That was the basic idea.
[01:49.420 --> 01:56.000]  And a friend of mine, working for the same company, provided pictures from a webcam stream with striptease progression.
[01:56.360 --> 01:59.000]  So the chatbot and the pictures should make the trick.
[01:59.220 --> 01:59.800]  I thought.
[02:01.040 --> 02:02.940]  The website was a complete disaster.
[02:04.240 --> 02:05.660]  It didn't last a year.
[02:05.660 --> 02:09.300]  And I talked with some of the users and they gave me feedback.
[02:09.620 --> 02:13.960]  And it turns out that they always recognize the bot behind the woman.
[02:15.500 --> 02:18.740]  So whatever, whatever, I put all the files in a backup.
[02:18.960 --> 02:21.660]  The database, the code, the templates, and the pictures.
[02:21.940 --> 02:25.700]  And completely forgot about it until last year.
[02:27.800 --> 02:32.280]  Last year in Argentina, we changed a 12-year monarchy-style government.
[02:32.280 --> 02:36.640]  First of all, this guy over here, Nestor Kirchner, won the presidency.
[02:38.000 --> 02:39.700]  Well, in fact, he didn't win.
[02:39.900 --> 02:41.860]  He came out second.
[02:42.480 --> 02:46.780]  But we have in place something called ballotage, which is just another terrible French idea.
[02:47.420 --> 02:48.480]  Like rotten cheese.
[02:49.320 --> 02:53.240]  With ballotage, if you lose an election, you make everyone vote again.
[02:53.240 --> 03:00.140]  And then maybe the second time you win because your opponent will not be present because of the polls that are not so convenient.
[03:00.700 --> 03:03.380]  I think here you call it a two-round system.
[03:04.400 --> 03:08.200]  So this guy was president with 22% of the votes.
[03:09.020 --> 03:17.000]  And after that, his wife, right here, Cristina Fernandez de Kirchner, won the presidency for two pariots.
[03:17.180 --> 03:21.720]  So four years, eight years, 12 years, the same family in the government.
[03:23.020 --> 03:27.600]  The man-in-the-middle, in this case, is not a communications attack.
[03:28.140 --> 03:32.480]  That is just Hugo Chavez, the former dictatorship of Venezuela.
[03:33.140 --> 03:34.620]  That's a nice threesome.
[03:35.720 --> 03:36.200]  And...
[03:37.440 --> 03:40.280]  I don't know who that woman is with the funny hat.
[03:42.460 --> 03:49.480]  So in Argentina, we have 41 million people, as you can imagine, having the best two for office in the same house.
[03:49.700 --> 03:50.820]  The odds are really low.
[03:51.260 --> 03:57.120]  And maybe you know that because you had like father and son, and now maybe you had the marriage as well.
[03:58.360 --> 04:00.800]  But that was not even the problem in Argentina.
[04:00.800 --> 04:05.680]  The problem was that during those years, we had like huge corruption.
[04:06.540 --> 04:09.000]  It is pretty common in South American countries.
[04:09.420 --> 04:17.340]  If you want to work with the government, let's say you want to make a highway, you have to put a 10% under the table, like a sweetener.
[04:19.880 --> 04:23.800]  During those years, that 10% went to 50%.
[04:25.020 --> 04:28.740]  And of course, if you don't want to make the highway, that's okay as well.
[04:29.920 --> 04:33.820]  We also had nepotism, friends and family working for the government.
[04:34.620 --> 04:40.040]  We had statistics manipulation, the government made up all the numbers of economics.
[04:40.600 --> 04:47.580]  And if you have a company, you work in that area, and you want to publish your own numbers, they will put your fines and, well, anything.
[04:47.580 --> 04:53.960]  Then exchange fraud, they use privileged information to exchange pesos for dollars, and they won like two or three million dollars.
[04:55.160 --> 05:00.320]  And if you want to change your own pesos for dollars as a common citizen, you weren't allowed to do that.
[05:01.540 --> 05:02.300]  What else?
[05:02.560 --> 05:04.640]  A little drug money financing campaigns.
[05:05.440 --> 05:06.320]  Ephedrine smuggling.
[05:06.520 --> 05:09.140]  Ephedrine is a drug used to make another drugs.
[05:10.040 --> 05:12.440]  And imports of ephedrine were really low.
[05:12.600 --> 05:14.220]  During those years, went to the roof.
[05:14.620 --> 05:16.100]  And ephedrine is used to make drugs.
[05:16.300 --> 05:17.000]  Let's remember that.
[05:20.660 --> 05:35.560]  Last but not least, a prosecutor investigating the connection between the government and people from Iran involved in bombings, attacks in Buenos Aires, was found dead just the night before going to make an accusation against the president.
[05:36.480 --> 05:39.420]  As we say in Argentina, we know it was a suicide.
[05:39.760 --> 05:40.860]  We just don't know who did it.
[05:46.110 --> 05:51.370]  And two months ago, this guy over here was a former public works secretary.
[05:51.370 --> 05:56.170]  He was found trying to hide nine million dollars in cash in a convent.
[05:56.550 --> 05:57.790]  Nine million dollars.
[05:57.950 --> 05:59.310]  Bags and bags and bags of money.
[05:59.430 --> 06:02.410]  He was also carrying a huge military weapon.
[06:02.530 --> 06:03.630]  That's understandable.
[06:03.910 --> 06:04.590]  Nine million dollars.
[06:05.410 --> 06:09.450]  And of course, he never heard about bitcoins and blockchain.
[06:10.650 --> 06:14.310]  And, well, God forgives, but it doesn't always help.
[06:14.650 --> 06:16.870]  He's now in prison, so that's good news.
[06:18.130 --> 06:24.710]  During those years, the last eight years, Cristina Fernández de Kirchner, the president, didn't give press conferences.
[06:24.990 --> 06:30.410]  After all, press conferences are annoying events where you have, like, journalists making awkward questions.
[06:31.210 --> 06:36.450]  Instead of that, she used a Twitter account to provide skewed pieces of information.
[06:37.090 --> 06:43.910]  Also to harass the opposition, to excel fanatics, and even to make racist Chinese jokes like this one.
[06:44.550 --> 06:47.970]  This is an actual tweet from the president of Argentina.
[06:50.470 --> 06:55.230]  So, after reading... I was one of the followers of the president's account.
[06:55.750 --> 07:01.110]  After reading that, I thought, well, my porn bot is able to tweet like that.
[07:03.210 --> 07:07.190]  So, I went to... I found that backup folder.
[07:07.530 --> 07:10.690]  I extracted the code and tried to look what was in there.
[07:10.970 --> 07:13.450]  So, I found ESP code.
[07:13.650 --> 07:15.290]  That's active server pages.
[07:15.650 --> 07:17.490]  It was a programming language by Microsoft.
[07:17.670 --> 07:18.510]  No longer used.
[07:19.070 --> 07:20.790]  It was pretty common back then.
[07:21.350 --> 07:22.870]  Connected to an access database.
[07:22.870 --> 07:29.250]  So, I migrated the code from ESP to PHP, a great programming language with all the features I needed.
[07:29.930 --> 07:34.010]  And I migrated the code from the database from Access to MySQL.
[07:35.430 --> 07:41.330]  Then I wrote an important piece of code to connect the chatbot with Twitter.
[07:41.510 --> 07:48.530]  Let's remember that the chatbot originally obtained all the inputs using a web form and all the outputs using HTML.
[07:49.790 --> 08:00.290]  So, I wrote this code and now all the inputs were provided using Twitter mentions and all the outputs were answered using Twitter status update.
[08:01.970 --> 08:03.290]  Then I had the database.
[08:03.930 --> 08:08.070]  In the database, I have words and expressions from the porn industry.
[08:08.410 --> 08:11.030]  You can say that porn and politics have lots in common.
[08:11.230 --> 08:12.430]  You could be right.
[08:12.430 --> 08:18.850]  But you won't find in porn, not even in sadistic porn, somebody so angry as our former president.
[08:22.410 --> 08:28.610]  So, I went to the database and started to load the database with words and expression from Cadena Nacional.
[08:29.190 --> 08:33.750]  Cadena Nacional were long, really long, two hour, three hour long presidential speeches.
[08:34.950 --> 08:39.910]  That by law, all TV stations and radio networks should supposed to broadcast.
[08:40.430 --> 08:43.830]  Of course, the law also says that you need an emergency to make Cadena Nacional.
[08:44.130 --> 08:48.170]  But, you know, for them, like emergency was a broad concept.
[08:48.510 --> 08:53.070]  And they used Cadena Nacional even to announce when the president got a new dog.
[08:55.650 --> 08:56.450]  And that's true.
[08:56.570 --> 08:57.130]  You can look it up.
[08:59.930 --> 09:03.470]  So, let's take a look at how this chatbot works.
[09:03.650 --> 09:07.670]  First of all, there's a piece of code monitoring the Twitter account.
[09:07.930 --> 09:10.030]  I registered this one, CFK Responde.
[09:10.390 --> 09:12.330]  The original was CFK Argentina.
[09:12.850 --> 09:13.970]  That's the initials.
[09:14.070 --> 09:18.930]  And I thought, well, maybe the people will think that the president used an institutional Twitter account.
[09:19.870 --> 09:20.910]  Maintained by a CM.
[09:21.130 --> 09:24.250]  And then, with this account, she want now to answer the citizens.
[09:25.070 --> 09:27.490]  So, a piece of code monitors the account.
[09:27.670 --> 09:34.130]  When a new mention is found, that mention is saved into the database, into a log, for three reasons.
[09:34.490 --> 09:41.590]  First of all, I wanted to have, to check what's going on in there, to have a log to tweak the application.
[09:42.870 --> 09:46.090]  Then, I don't want to answer twice the same mentions.
[09:46.390 --> 09:53.430]  And then, I was going to use that log as well to provide answers, to use that, and I will talk about it later.
[09:54.510 --> 10:02.390]  The next step was to parse, to remove the account from the tweet, and to parse words and phrases, and make a database comparison.
[10:03.090 --> 10:08.370]  And finally, when an answer is found, it's sent to Twitter, using Twitter status update.
[10:10.210 --> 10:12.190]  So, now, let's talk about the answers.
[10:12.690 --> 10:14.310]  I have three levels.
[10:14.470 --> 10:18.250]  Level one is when the chatbot finds a complete phrase.
[10:19.230 --> 10:23.850]  If I have a complete phrase, chances are that the answer will be pretty good.
[10:24.410 --> 10:27.950]  If I'm not able to find a complete phrase, I try to find just a word.
[10:28.350 --> 10:32.850]  If I have a word, the answer is good, but not as good as level one.
[10:33.550 --> 10:35.470]  And then, I have all-purpose answers.
[10:35.650 --> 10:42.670]  In those cases, I'm not able to find a phrase or a word, and I'm using just generic strings, like, tell me more about that.
[10:42.670 --> 10:45.110]  I don't agree, let's agree to disagree, whatever.
[10:46.450 --> 10:50.870]  And a nice trick that the chatbot also does is a mix of online and offline content.
[10:51.950 --> 11:02.810]  In this case, the chatbot is able to go to the main opposition newspaper, get headlines from the politics section, and mix those headlines with offline content.
[11:03.350 --> 11:08.770]  An example would be, well, look what the newspaper is saying about me, I don't agree, what do you think?
[11:08.930 --> 11:12.110]  So, it's a nice trick, and usually people get impressed about that.
[11:13.270 --> 11:16.570]  And another thing is, it's able to use information from Twitter.
[11:17.110 --> 11:23.550]  For example, the chatbot is able to say, well, you could be right, but you only have five followers, so who cares?
[11:25.630 --> 11:38.250]  And then, the chatbot is also able to use previous mentions that are now in the database in the log, so it's able to get something that anybody says, and put that as a Twitter update.
[11:39.990 --> 11:49.630]  So, I put everything in place, and I thought, well, maybe a few people will start to use this, and they will say, well, I know you from that stupid porn chatbot in the mid-90s.
[11:51.170 --> 11:55.210]  But, to my surprise, things started to work really well.
[11:56.650 --> 12:00.010]  Here, we can see an example of an actual tweet.
[12:00.630 --> 12:05.570]  A guy named Joshua says, I'm going to make a real-time translation.
[12:06.870 --> 12:12.910]  You're a criminal, assessing, you're going to be in jail for the death of Alberto Nisman, that was the prosecutor.
[12:13.530 --> 12:17.910]  You're going to have a fair trial, you're assessing, you're going to be in jail for life.
[12:17.910 --> 12:19.010]  That's pretty intense.
[12:19.910 --> 12:25.010]  And the bot says, anyone who does sometimes makes mistake, you never did anything.
[12:25.890 --> 12:33.170]  And the guy answers, I did some things in the past, and I paid for them, but I never stole like you, you're going to be in jail.
[12:33.550 --> 12:36.390]  So, the conversation is taking place.
[12:36.390 --> 12:45.670]  In this case, a girl named Lucrezia says, since 30 years ago, I'm building democracy day by day, from my works in the opposition party.
[12:45.950 --> 12:50.010]  Other ones makes wealth using the votes.
[12:50.630 --> 12:55.650]  And in this case, it seems that the bot detected this word, years.
[12:55.990 --> 13:00.530]  And it says, your notions of timing are a little bit confused, so middle class.
[13:05.280 --> 13:11.140]  And the conversation goes on, and the girl says, is it wrong that everyone has the same opportunities?
[13:11.340 --> 13:12.360]  You are the confused one.
[13:13.140 --> 13:15.620]  And here, the answer is, you shouldn't be so extremist.
[13:19.650 --> 13:22.470]  So, it is working pretty well.
[13:26.120 --> 13:30.720]  In this case, the bot says, by the way, how do you reach here?
[13:30.900 --> 13:34.520]  Well, a question, and a guy named Leonel says, I'm Kirchnerista.
[13:34.580 --> 13:37.500]  That means that he's a follower of the official party.
[13:38.000 --> 13:41.940]  And I'm writing because of workers at sea are without jobs.
[13:42.120 --> 13:42.840]  Nobody listens.
[13:43.080 --> 13:43.760]  Cristina, please.
[13:43.920 --> 13:48.220]  So, this guy over here thinks that this account is actually the president.
[13:48.220 --> 13:51.220]  And the chat bot is not allowed to give jobs.
[13:51.460 --> 13:55.340]  So, he says, until what point the job is just an income source?
[13:55.540 --> 13:58.040]  It's just detecting one word, and it answers.
[13:58.580 --> 14:02.180]  I'm sorry, that guy, well, I hope he has a job by now.
[14:05.100 --> 14:12.740]  Well, here, the bot says, you have to know that we are leaders in human rights for everything we have to live.
[14:12.740 --> 14:16.720]  This seems to be actually a phrase obtained from Cadena Nacional.
[14:16.940 --> 14:19.920]  That is a propaganda made by the government all days.
[14:20.920 --> 14:24.080]  And a guy here, all in uppercase, is really pissed off.
[14:24.440 --> 14:27.880]  He says, I doubt that it's you in person, the one that is answering.
[14:28.100 --> 14:32.120]  And if it's like that, you give yourself a moment and thinks, what can you do better?
[14:33.200 --> 14:39.080]  So, he considers that maybe this is not the president, but he thinks that there's someone there answering.
[14:40.360 --> 14:46.100]  And here you have a guy named Dosto, a clever one, this is like LOL, he's laughing.
[14:46.560 --> 14:49.220]  And he says, this is not the official CFK.
[14:49.560 --> 14:54.000]  That is just a dumb CM who's giving scripts to talk to the people.
[14:54.240 --> 15:01.160]  So, he's clever enough to understand that this is not the official account, but he never imagined that this is just a computer application.
[15:02.620 --> 15:06.320]  So, it's important to know that this is not real time.
[15:06.320 --> 15:12.100]  This PHP script is executed using a cron shop in a Linux box once per hour.
[15:13.120 --> 15:18.460]  So, the illusion is far from perfect, because if you write at 2 in the morning, the president will answer at 2 in the morning.
[15:19.380 --> 15:22.520]  Of course, I can change that, but that was not necessary.
[15:24.480 --> 15:27.720]  And the account, anyway, started to have real engagement.
[15:30.440 --> 15:34.580]  So, I was able to identify three main groups of people interacting with this bot.
[15:35.060 --> 15:40.000]  First of all, we have people that think that this account is actually the president, the former president now.
[15:40.740 --> 15:47.220]  Some people that think that this account was maintained by an employee, like the official institutional account.
[15:47.680 --> 15:48.680]  Lots of people here.
[15:48.860 --> 15:52.680]  And some people that think that this is just a fake, somebody there answering the question.
[15:52.840 --> 15:54.300]  But there is no group D.
[15:54.420 --> 15:55.920]  There is not enough people.
[15:55.920 --> 16:00.360]  In fact, for months and months, nobody, not even one person says, well, you are a bot.
[16:00.580 --> 16:01.340]  This is a computer.
[16:01.460 --> 16:02.040]  This is stupid.
[16:02.560 --> 16:08.480]  But then I wrote an article in the magazine, and some people tried to trick the bot, and that is okay.
[16:08.580 --> 16:09.440]  And of course, you can do that.
[16:09.500 --> 16:10.280]  It's not perfect.
[16:10.540 --> 16:14.280]  Let's remember that this is the same chatbot from the mid-90s.
[16:16.120 --> 16:20.660]  So, my reactions with this project were like a roller coaster.
[16:20.920 --> 16:22.440]  That's supposed to be me, there.
[16:24.040 --> 16:26.180]  I don't have... my teeth are not so white.
[16:28.220 --> 16:30.240]  At first, I was a little bit scared.
[16:31.780 --> 16:34.180]  Because it all started like a small experiment.
[16:34.420 --> 16:35.000]  It started to grow.
[16:35.760 --> 16:38.900]  And sometimes you can imagine all the things your government is able to do.
[16:38.900 --> 16:44.160]  In my case in Argentina, I was very aware of the things that this government have done.
[16:44.420 --> 16:52.780]  So, like maintaining violent groups, and putting fines, and sending the IRS to investigate you, if you have an opposition voice, or something like that.
[16:53.460 --> 16:55.960]  But then, last year, there was this situation.
[16:56.360 --> 16:57.560]  Nestor Kirchner was dead.
[16:57.900 --> 16:59.760]  Cristina Kirchner was president for Chuper Riot.
[16:59.940 --> 17:02.240]  So, she wasn't able to run again for president.
[17:02.860 --> 17:08.040]  And they hadn't any good candidate, and they were a little bit worried about how to...
[17:08.580 --> 17:11.040]  and get out of the power without being prosecuted.
[17:11.260 --> 17:13.700]  And so, I thought, well, maybe I was safe now.
[17:13.920 --> 17:15.640]  And I started to have some fun with this project.
[17:16.460 --> 17:17.800]  So, I made this little box.
[17:18.000 --> 17:19.620]  The box is over there, right now.
[17:21.580 --> 17:23.280]  Inside it has a Raspberry Pi.
[17:23.480 --> 17:29.700]  It has some USB Wi-Fi dongle, LCD screen, and two buttons, I don't know.
[17:30.740 --> 17:33.300]  And it monitors the account in real time.
[17:33.340 --> 17:37.360]  So, I'm able to see what the people is telling to the bot.
[17:37.360 --> 17:40.580]  And I have that on my desk, and every day I go there, and I read.
[17:40.760 --> 17:41.820]  It's really fun.
[17:43.420 --> 17:50.980]  But then, I started to think, how easy is to cheat people in the main communication platforms we use today, like Twitter and WhatsApp.
[17:51.440 --> 18:00.120]  Both use, of course, you can use links and videos and stuff like that, but mainly both use small phrases, small bursts of text.
[18:00.120 --> 18:02.020]  And that is really easy to cheat.
[18:03.620 --> 18:06.460]  And then, I started to think another thing.
[18:06.620 --> 18:10.320]  Why is that we cannot tell the difference between a person and a bot today?
[18:11.720 --> 18:16.520]  And, like 20 years ago, this same application didn't come...
[18:16.520 --> 18:19.480]  The average porn user wasn't convinced about this illusion.
[18:19.700 --> 18:20.880]  And then today, it's really easy.
[18:20.880 --> 18:28.580]  And I thought that maybe we cannot tell the difference anymore, because we behave today as bots.
[18:29.000 --> 18:30.520]  We write like as bots.
[18:30.780 --> 18:32.040]  We use small phrases.
[18:32.040 --> 18:34.100]  We don't pay attention to the other part.
[18:35.440 --> 18:41.640]  So, we cannot tell the difference anymore, because a bot write just like ourselves.
[18:43.400 --> 18:47.220]  Then I started to think about the connection between politicians and social networks.
[18:47.480 --> 18:54.240]  I don't think that they both share the same agenda, but they both have something important in common.
[18:54.380 --> 19:00.860]  They both need your contributions really badly, and they both want to put your contributions in a sandbox.
[19:01.140 --> 19:04.600]  So, there's nothing you can do that will be able to harm them.
[19:06.160 --> 19:17.060]  In Argentina, we have a very popular social network that is only popular there, and once I wrote an article, it went pretty well with comments and stuff, and then they delete the whole thing.
[19:17.260 --> 19:19.280]  And I ask, what's going on here?
[19:19.440 --> 19:24.640]  And they ask me, they reply, well, you're not respecting the protocol.
[19:25.120 --> 19:28.480]  And the protocol was a fancy way to talk about terms and conditions.
[19:28.840 --> 19:37.560]  I went there, I read terms and conditions, and they say there, you cannot make criticism against this social network.
[19:37.560 --> 19:44.320]  Of course, you can, if you're happy, you can say that, but you cannot talk about how we are running those things.
[19:44.680 --> 19:50.360]  And I think this could be hidden, this could be explicit, but it's the same in all social networks.
[19:51.880 --> 19:54.720]  And in Argentina, voting is mandatory, you have to vote.
[19:54.860 --> 20:02.300]  If you don't vote, you're not allowed to renew your passport, they will put you fines, so you have to vote.
[20:02.300 --> 20:04.400]  And you have to vote pretty often.
[20:04.600 --> 20:10.300]  You have, like, primaries, and you have the voting for this and that, and you have the ballotage, you have to go again and vote.
[20:11.480 --> 20:13.980]  And so they need your contributions.
[20:14.480 --> 20:18.560]  In social networks, they need your post or anything, and here, they need your votes.
[20:19.620 --> 20:22.700]  But there's nothing you can do that will be able to change the system.
[20:22.840 --> 20:25.140]  And that's a sad reality.
[20:27.200 --> 20:31.500]  So, as it is common, this type of projects, you end up with more questions than answers.
[20:32.100 --> 20:38.460]  In my case, I started to think about how many times do we interact with bots without even knowing.
[20:39.140 --> 20:48.660]  So, sometimes we are chatting with someone or asking something about a pre-sales question using a chat system, and that could be a bot, and we don't know about that.
[20:49.240 --> 20:53.400]  Then I started to think, why do we use Twitter to discuss politics?
[20:53.400 --> 20:58.740]  It's a place that, I don't know, maybe for politics, you need to elaborate more.
[20:59.020 --> 21:00.300]  And then we use Twitter.
[21:00.480 --> 21:01.040]  I don't know why.
[21:01.480 --> 21:03.480]  And maybe because the people is there.
[21:03.780 --> 21:09.380]  And then I also started to think, why the politicians feel so comfortable in Twitter?
[21:10.160 --> 21:16.140]  Maybe the answer is that you put a monkey there with slogans all the time, and that works just well.
[21:16.220 --> 21:17.640]  You don't need to elaborate nothing.
[21:19.840 --> 21:22.940]  So, maybe we'll have a little question and answers.
[21:22.940 --> 21:24.020]  I have the language barrier.
[21:24.180 --> 21:29.960]  So, if you're going to make a question, please make it short, slow, and clear.
[21:30.660 --> 21:33.240]  But, first, I wanted to ask to this guy over here.
[21:33.380 --> 21:35.760]  I received this tweet while being here at HOPE.
[21:36.060 --> 21:38.520]  He says, his name is Ruben Carranza.
[21:38.800 --> 21:43.300]  He asked, does HOPE conf care about Argentina Junta's impact on politics?
[21:43.300 --> 21:46.960]  Ronny Bandini panel oversimplifies Kirchner's record.
[21:49.120 --> 21:50.740]  I respect all opinions.
[21:50.980 --> 21:53.000]  I think Ruben has a point.
[21:53.220 --> 21:56.380]  I'm, for sure, oversimplifies Kirchner's record.
[21:56.560 --> 22:01.040]  He means, about this, Junta's impact, we have in the 70s a military government.
[22:01.420 --> 22:02.860]  Any military government is bad.
[22:03.020 --> 22:04.980]  This one also killed people and torture.
[22:04.980 --> 22:06.360]  Well, it was really bad.
[22:07.360 --> 22:12.860]  And the Kirchner's wanted to present themselves as some kind of human rights superheroes.
[22:13.660 --> 22:25.100]  But, in fact, if you take a look at the records and you read the facts, you will understand that the only thing that was common in all the history of this family was greed and money, and not human rights.
[22:25.100 --> 22:27.740]  So, that's my answer to this tweet.
[22:28.840 --> 22:35.720]  If you're interested in the code, I published some pieces of this code in this magazine.
[22:35.880 --> 22:39.400]  You can find it on page 15, right outside.
[22:40.480 --> 22:42.420]  So, again, thank you for being here.
[22:42.580 --> 22:43.980]  I really appreciate this opportunity.
[22:44.220 --> 22:44.800]  I'm from Argentina.
[22:45.080 --> 22:48.360]  It's really far away, so being here in New York is a great experience.
[22:48.780 --> 22:53.180]  There you have my Twitter account, Facebook, and, again, thank you very much.
[22:53.180 --> 22:55.240]  Here's a phrase from a writer from Argentina.
[22:55.540 --> 22:55.820]  Thank you.
[23:06.160 --> 23:09.000]  Well, if there's no questions... Okay, yeah.
[23:09.920 --> 23:12.360]  How many followers did you have?
[23:12.940 --> 23:14.160]  Okay, how many followers?
[23:14.420 --> 23:16.200]  And did you use both for followers?
[23:16.880 --> 23:17.880]  If I use, sorry?
[23:18.340 --> 23:19.900]  Both for followers too?
[23:20.200 --> 23:20.660]  No, no.
[23:20.740 --> 23:25.500]  This account, in fact, has, I think, 8,000 followers.
[23:25.740 --> 23:27.920]  It just started, like, last year.
[23:29.900 --> 23:34.240]  But lots of the people that interact with the bot is not following the account.
[23:34.540 --> 23:36.580]  It's just conversations that they don't follow the account.
[23:36.700 --> 23:43.920]  So it's not a huge success in terms of followers, but it is a success in terms of conversations and taking place there.
[23:45.720 --> 23:46.460]  Anyone else?
[23:47.320 --> 23:47.720]  Yeah.
[23:48.100 --> 23:53.660]  Are you thinking of actually doing this with other accounts with other politicians as well, or...?
[23:54.460 --> 23:57.040]  Yeah, if I'm interested in doing this with other politicians.
[23:57.240 --> 24:00.440]  Yeah, sure, that could be a good experiment.
[24:00.600 --> 24:05.500]  Right now we have another president, Mauricio Macri, and it could be interesting of replicating this experience.
[24:05.500 --> 24:07.500]  And maybe doing this in English as well.
[24:07.640 --> 24:08.780]  Maybe we can do it here too.
[24:09.060 --> 24:09.420]  Who knows?
[24:09.600 --> 24:10.300]  It's interesting.
[24:10.520 --> 24:12.860]  If you're interested in this, I have the code available.
[24:13.580 --> 24:13.760]  Yes?
[24:14.100 --> 24:16.500]  You should do a parody account of Donald Trump.
[24:17.160 --> 24:17.600]  Okay.
[24:18.720 --> 24:19.780]  That would be fun.
[24:19.940 --> 24:22.640]  But I think the original is pretty fun, so...
[24:23.820 --> 24:24.560]  I don't know.
[24:25.720 --> 24:27.500]  What can I add there, so...
[24:28.520 --> 24:28.780]  Yes?
[24:28.780 --> 24:29.620]  Did you just...
[24:29.620 --> 24:33.280]  When you started the bot, did you just pick any random...
[24:33.280 --> 24:35.140]  Like, how did you start conversing with people?
[24:37.260 --> 24:39.600]  How did the conversations start?
[24:39.760 --> 24:41.080]  That's an interesting question.
[24:41.320 --> 24:44.880]  Well, I have a piece of code that I did not explain here.
[24:45.280 --> 24:48.580]  I have a script running that identifies one word.
[24:49.060 --> 24:50.220]  The word is kernerismo.
[24:50.480 --> 24:50.900]  That's...
[24:50.900 --> 24:54.600]  When you're opposition, you use that word to talk...
[24:54.600 --> 24:55.720]  It's like...
[24:55.720 --> 24:57.200]  To talk bad about the government.
[24:57.200 --> 25:01.320]  So, I try to find that word, and I start with a tweet from this account.
[25:01.560 --> 25:03.360]  So, that usually starts the conversation.
[25:03.560 --> 25:06.840]  Of course, if you go there and you start yourself, it will answer as well.
[25:07.040 --> 25:08.340]  But usually, that...
[25:08.340 --> 25:11.100]  That script starts many conversations.
[25:13.160 --> 25:13.560]  Yes?
[25:14.140 --> 25:16.940]  You said you run it with a cron job once an hour.
[25:17.160 --> 25:20.700]  Do you only respond to one tweet, or do you respond to every tweet in the account?
[25:22.440 --> 25:26.760]  The question is, there's a cron shop executing this, if that cron shop only answers to one.
[25:26.760 --> 25:32.760]  No, in fact, it answers to all the mentions without answers until that point.
[25:32.920 --> 25:37.080]  It used like a timer, not all the tweets at the same time, like seconds later.
[25:37.260 --> 25:45.400]  But usually, if the cron shop is executed right now at 17 minutes every hour, you will get between 17 and 18 minutes all the answers there.
[25:47.420 --> 25:47.860]  Yes?
[25:48.520 --> 25:55.000]  So, last year, there was a chat bot, I think it was by Microsoft, I'm not sure that people put up.
[25:55.580 --> 26:02.520]  And within about a day or two of it going up, people were able to manipulate the bot into saying really hateful, racist messages.
[26:02.520 --> 26:09.840]  Is there any chance of someone hitting your bot and manipulating your bot into saying things that you would not want it to say also?
[26:10.600 --> 26:11.220]  Of course.
[26:11.400 --> 26:16.180]  Well, the main difference is that that bot was developed by Microsoft, so there's an interest there.
[26:16.180 --> 26:22.000]  So, if you do the same here, I don't know, it's nothing terrible will happen.
[26:22.360 --> 26:25.160]  Of course, you can trick this bot as well if you have that intention.
[26:25.380 --> 26:32.520]  But mainly, the people that is interacting with the bot doesn't think that this is a bot, so they don't have that intention.
[26:32.740 --> 26:34.260]  But, of course, you can do that as well.
[26:35.960 --> 26:36.620]  Anyone else?
[26:37.440 --> 26:37.840]  Yes?
[26:38.500 --> 26:44.460]  Can you go a little bit more into how deep the answer file is?
[26:44.640 --> 26:49.460]  I mean, are you just... how are you coming up with the answers, the responses that you have?
[26:49.600 --> 26:50.080]  Okay.
[26:51.040 --> 26:56.020]  So, this chat bot is exactly as the original I did in the mid-90s.
[26:56.260 --> 27:00.180]  The main difference that maybe right now I have a huge database.
[27:00.840 --> 27:07.120]  The database, when you have a huge database, chances are that you're detecting phrases, complete phrases and words.
[27:07.120 --> 27:10.000]  And those answers are pretty good.
[27:10.200 --> 27:12.520]  So, the procedure was like that.
[27:12.720 --> 27:14.960]  I try... I parse the original text.
[27:15.080 --> 27:15.860]  I try to find words.
[27:16.060 --> 27:18.560]  If I'm not getting word... phrases, sorry.
[27:18.920 --> 27:21.080]  If I'm not getting phrases, I go with words.
[27:21.200 --> 27:25.600]  If I'm not getting words, I use that mix of online and offline content.
[27:25.860 --> 27:32.320]  And sometimes I use... I put random things, so you're not always getting the same responses for words and phrases.
[27:32.540 --> 27:36.360]  And that works pretty well, basically, because the database is a big database.
[27:36.360 --> 27:44.820]  But is the code actually recognizing the grammatical structure in a sentence?
[27:45.140 --> 27:50.660]  I mean, like, are you just picking out single vocabulary words?
[27:50.820 --> 27:56.500]  Or you actually have already attached grammatically correct... I'm trying to get up the natural language of the process.
[27:56.720 --> 27:58.100]  No natural language here.
[27:58.220 --> 28:02.420]  Just a huge database and a dumb script and the effect is really good.
[28:05.830 --> 28:06.730]  Any other question?
[28:07.810 --> 28:08.770]  Okay, that's all.
[28:08.830 --> 28:09.630]  Ah, sorry, yeah.
[28:10.590 --> 28:24.750]  So, when you're doing it as far as language-wise, is it different as far as the intonations in your language as far as how things are represented versus what people are actually asking?
[28:25.050 --> 28:25.850]  Is that an issue?
[28:26.070 --> 28:27.430]  Is the database the same answer?
[28:28.010 --> 28:31.350]  Um, if you have the same answers, if you write the same thing, that's the question.
[28:32.310 --> 28:36.950]  Um, chances are that you're not getting the same answer because, uh, there are random factors there.
[28:37.170 --> 28:42.610]  So, if you're putting twice the same thing, the vote will detect it's exactly the same thing.
[28:42.750 --> 28:45.810]  And it will give you an answer because you are repeating things.
[28:46.190 --> 28:48.110]  It's not going to find the same phrase.
[28:48.110 --> 28:51.650]  So, you need those small things to avoid, like, cheating.
[28:52.450 --> 28:52.510]  Yeah.
[28:53.710 --> 28:57.030]  Did you get any sort of an official response?
[28:57.250 --> 29:01.290]  Did anybody go over and ever reach out and say, if you want to stop, is that a good idea?
[29:01.950 --> 29:02.810]  No, no.
[29:03.050 --> 29:04.990]  Because I, I think because of the timing.
[29:05.250 --> 29:11.010]  I, I started with these things just a few months before, uh, change of government.
[29:11.010 --> 29:18.810]  Of course, if I did this, like, uh, one year before, I'm sure that I, I, maybe I had troubles.
[29:18.990 --> 29:19.290]  I don't know.
[29:19.430 --> 29:21.390]  But the timing was really important here.
[29:22.610 --> 29:23.430]  Any other question?
[29:23.650 --> 29:23.770]  Yeah.
[29:23.990 --> 29:29.190]  Were you, uh, was the account used as, like, as reference in other articles?
[29:29.870 --> 29:33.350]  Was it used as, like, documentation or stuff like that?
[29:33.670 --> 29:39.470]  Um, I just started this, I think, in September, November, uh, like, a few months ago.
[29:39.470 --> 29:45.550]  And, uh, it has some appearances in articles and stuff, but I'm usually not monitoring, I'm not even monitoring the, the account as well.
[29:45.710 --> 29:53.750]  Maybe, uh, once in a month I check there and I put a little code or phrase, but I'm not really, it's, it's just working by itself and it's working well.
[29:55.710 --> 29:57.090]  Well, thank you very much.
[29:57.210 --> 29:59.170]  If you're interested to talk about this, I will be outside.
[29:59.270 --> 29:59.490]  Thank you.