AI against harassment and depression: how bots make the world a better place.
Media says that artificial intelligence deprives of jobs, supports Adolf Hitler and discriminates women. But AI also helps people - and sometimes makes it more human than many of us. On the eve of AI School, we picked up four such bots and asked the head of the computational linguistics department of Nanosemantics company and course speaker Anna Vlasova to tell how to create a service that will make the world kinder.
“Russian users are still skeptical about bots and see artificial intelligence as a threat rather than a friend. For them, the AI service is less valuable than the advice from a human expert. However, the attitude abroad has already changed, and soon the trend will reach Russia, so the service bots have exactly the future, ”says Anna Vlasova. According to her, when creating a bot, the main thing is to find an idea that will be in demand.
We decided to consider four existing services.
Spot is a chat bot designed for implementation in companies. The bot allows the victim of sexual harassment at work to document what happened. Often, people who are faced with harassment, it is difficult to report this to another person. But complaining about the bot is much easier, say the creators of Spot.
The employee communicates with the bot in a free form, and the service compiles a report based on the information received. This document can be simply saved by the user: in the case of proceedings, it will be more weighty evidence than his own notes, the creators of the application say. In addition, an employee can anonymously send a report to the HR department of his company. An HR specialist can ask clarifying questions in a chat with an employee, while anonymity is preserved: the sender’s identity will not be disclosed. As a result, companies that use the bot will be able to form a panel of incidents.
Spot for companies up to 100 people costs $ 1,200 per year, service costs to large corporations $ 2 per employee per month. To implement Spot, the company must undertake to investigate the employee’s message within 10 business days. If the employer does not take any action during this time, the employee can contact the Spot, and the bot creators will contact the employer.
Spot is a product of a venture fund to support projects in the field of artificial intelligence All Turtles, which was founded by one of the founders of the former head of Evernote Phil Libin. One of the creators of Spot is a clinical psychologist and author of popular science books about false memory syndrome, Julia Shaw. She explains in a talk at TED about why it’s important to report her experiences right away.
The chat bot based on artificial intelligence Replika was created by a startup from San Francisco Luka. Its founders - a former columnist of "Posters" Eugene Kuyda and ex-employee of RIA Novosti Philip Dudchuk. The startup has already attracted more than $ 11 million investment.
The service was launched in March 2017, and at first Replika was just a digital copy of the user, imitating his style of communication. Now this is a full-fledged interlocutor who will support and listen for free. This is not a voice assistant like Siri: it does not call a taxi and does not keep a calendar.
Replika will be translated into other languages. “We have a large user base in Brazil, and for them the language is a serious barrier. If they could talk to Replika in Portuguese, it would greatly improve performance, ”said Yevgeny Kuyda. However, there are no plans to translate Replika into Russian from a startup.
“Russian is difficult because of grammatical homonymy, syntactic variability in the construction of a phrase and, of course, because of the ambiguity and ambiguity of meanings that can be extracted from the same phrase,” explains Anna Vlasova. “In general, a language with a developed grammar, a system of case endings, verb inflection is always more difficult in automatic processing. In this sense, the most convenient language for developers is Chinese.
At the same time, she said, language difficulties are not a critical factor that hinders the development and implementation of chat bots. “Now there is a wide range of different natural language processing technologies. You just need to choose the one that solves the necessary linguistic problem, ”says Anna Vlasova.
The DoNotPay chat bot was launched in 2016 and in the first month of its existence challenged 160 thousand cases. The program collects data about each case, leading a dialogue with the user. Based on the data obtained, the bot makes a complaint and helps to plan further steps.
Basically, the bot works with simple, typical legal problems. First, DoNotPay only processed parking fines, but then the service began to help people defend their homes, get compensation for the delayed flight, and then prepare papers for divorce.
At the time of the release of DoNotPay, its creator Joshua Browder was 19 years old. The idea of creating a service came to him when he was still in school. At age 17, he received a fine: the wheel of his car was a little outside the parking space. "The amount of the fine was 50 pounds, and I had only 100 pounds for pocket expenses per month," said Browder. He decided to challenge the matter: it turned out that the parking space was already what the law required. Having won the case, he decided to help those who faced the same problem, and created a bot. In more detail about its creation Browder tells in the performance:
Woebot was created by Stanford University psychologist Alison Darcy and AI experts in 2017. The bot asks questions about the mood, well-being, desires of the user and analyzes how likely the occurrence of depression.
Woebot is designed according to the methods of cognitive-behavioral therapy. Alison Darcy tested Woebot on her own students at the university. Most of the 70 students who passed the test, noticed an improvement in their mental state after two weeks of communication with Woebot. As in the case of Spot, the creators of Woebot believe that sometimes users find it easier to talk with a bot than a person.
Chat bot writes to his “patient” every day, sometimes he offers to watch a video or play a psychological game. He also draws up a plan of therapy and goals that the user wants to achieve.
Woebot is not the only bot psychologist, he has quite a lot of competitors. So, monitor the mood of the user and analyze the possibility of mental illness services Moodkit or Moodnotes. In addition, there are many services for meditation and anxiety relaxation.
According to Anna Vlasova, for the development of some bots, specially trained neural networks are needed that can determine the communicative intentions of the interlocutor. For others, a simple script is enough - a bot action sequence tree. If you already have an idea for a service that will make the world a better place, you will be able to learn technical subtleties on the AI School course .
“Russian users are still skeptical about bots and see artificial intelligence as a threat rather than a friend. For them, the AI service is less valuable than the advice from a human expert. However, the attitude abroad has already changed, and soon the trend will reach Russia, so the service bots have exactly the future, ”says Anna Vlasova. According to her, when creating a bot, the main thing is to find an idea that will be in demand.
We decided to consider four existing services.
Service to report Spot harassment
Spot is a chat bot designed for implementation in companies. The bot allows the victim of sexual harassment at work to document what happened. Often, people who are faced with harassment, it is difficult to report this to another person. But complaining about the bot is much easier, say the creators of Spot.
The employee communicates with the bot in a free form, and the service compiles a report based on the information received. This document can be simply saved by the user: in the case of proceedings, it will be more weighty evidence than his own notes, the creators of the application say. In addition, an employee can anonymously send a report to the HR department of his company. An HR specialist can ask clarifying questions in a chat with an employee, while anonymity is preserved: the sender’s identity will not be disclosed. As a result, companies that use the bot will be able to form a panel of incidents.
Spot for companies up to 100 people costs $ 1,200 per year, service costs to large corporations $ 2 per employee per month. To implement Spot, the company must undertake to investigate the employee’s message within 10 business days. If the employer does not take any action during this time, the employee can contact the Spot, and the bot creators will contact the employer.
Spot is a product of a venture fund to support projects in the field of artificial intelligence All Turtles, which was founded by one of the founders of the former head of Evernote Phil Libin. One of the creators of Spot is a clinical psychologist and author of popular science books about false memory syndrome, Julia Shaw. She explains in a talk at TED about why it’s important to report her experiences right away.
Replika Virtual Friend
The chat bot based on artificial intelligence Replika was created by a startup from San Francisco Luka. Its founders - a former columnist of "Posters" Eugene Kuyda and ex-employee of RIA Novosti Philip Dudchuk. The startup has already attracted more than $ 11 million investment.
The service was launched in March 2017, and at first Replika was just a digital copy of the user, imitating his style of communication. Now this is a full-fledged interlocutor who will support and listen for free. This is not a voice assistant like Siri: it does not call a taxi and does not keep a calendar.
Replika will be translated into other languages. “We have a large user base in Brazil, and for them the language is a serious barrier. If they could talk to Replika in Portuguese, it would greatly improve performance, ”said Yevgeny Kuyda. However, there are no plans to translate Replika into Russian from a startup.
“Russian is difficult because of grammatical homonymy, syntactic variability in the construction of a phrase and, of course, because of the ambiguity and ambiguity of meanings that can be extracted from the same phrase,” explains Anna Vlasova. “In general, a language with a developed grammar, a system of case endings, verb inflection is always more difficult in automatic processing. In this sense, the most convenient language for developers is Chinese.
At the same time, she said, language difficulties are not a critical factor that hinders the development and implementation of chat bots. “Now there is a wide range of different natural language processing technologies. You just need to choose the one that solves the necessary linguistic problem, ”says Anna Vlasova.
Free DoNotPay bot lawyer
The DoNotPay chat bot was launched in 2016 and in the first month of its existence challenged 160 thousand cases. The program collects data about each case, leading a dialogue with the user. Based on the data obtained, the bot makes a complaint and helps to plan further steps.
Basically, the bot works with simple, typical legal problems. First, DoNotPay only processed parking fines, but then the service began to help people defend their homes, get compensation for the delayed flight, and then prepare papers for divorce.
At the time of the release of DoNotPay, its creator Joshua Browder was 19 years old. The idea of creating a service came to him when he was still in school. At age 17, he received a fine: the wheel of his car was a little outside the parking space. "The amount of the fine was 50 pounds, and I had only 100 pounds for pocket expenses per month," said Browder. He decided to challenge the matter: it turned out that the parking space was already what the law required. Having won the case, he decided to help those who faced the same problem, and created a bot. In more detail about its creation Browder tells in the performance:
“At the core of DoNotPay is the IBM Watson supercomputer's AI: during operation, the bot itself accesses external intelligent services that help it recognize the meaning of the interlocutor's statement. Using an external AI service built on machine learning or on the classic rule-based technology is a good option for developers. This allows you to focus on the purpose of the bot, and not on the toolkit.
Not only IBM offers technologies that allow you to analyze a phrase and highlight entities (objects) and intents (intentions) in it. Such services are in many companies, leading R & D development in the field of AI. This includes, for example, Microsoft, which has an intelligent Luis service on the MS Azure platform. You can get acquainted with Microsoft's intellectual services at various courses for developers organized by the company. On the AI School course in the Binary District, you can try creating a simple bot on this platform yourself. ”
Anna Vlasova, Head of Computational Linguistics Department, Nanosemantics Company
Bot psychotherapist Woebot
Woebot was created by Stanford University psychologist Alison Darcy and AI experts in 2017. The bot asks questions about the mood, well-being, desires of the user and analyzes how likely the occurrence of depression.
Woebot is designed according to the methods of cognitive-behavioral therapy. Alison Darcy tested Woebot on her own students at the university. Most of the 70 students who passed the test, noticed an improvement in their mental state after two weeks of communication with Woebot. As in the case of Spot, the creators of Woebot believe that sometimes users find it easier to talk with a bot than a person.
Chat bot writes to his “patient” every day, sometimes he offers to watch a video or play a psychological game. He also draws up a plan of therapy and goals that the user wants to achieve.
Woebot is not the only bot psychologist, he has quite a lot of competitors. So, monitor the mood of the user and analyze the possibility of mental illness services Moodkit or Moodnotes. In addition, there are many services for meditation and anxiety relaxation.
According to Anna Vlasova, for the development of some bots, specially trained neural networks are needed that can determine the communicative intentions of the interlocutor. For others, a simple script is enough - a bot action sequence tree. If you already have an idea for a service that will make the world a better place, you will be able to learn technical subtleties on the AI School course .