Estonia is trying to use AI in justice

The government is usually the last place to look for IT innovations, or new technologies like artificial intelligence. But Ott Welsberg may be able to change your opinion on this matter. He is a 28-year-old postgraduate student and director of data from Estonia, and he is leading the introduction of AI and Moscow Region in this tiny Baltic country to the services provided to citizens of the country , which number 1.3 million.

“We want the government to be as economical as possible,” says Welsberg, an Estonian citizen who works on a dissertation at Umeå University in Sweden, who is wearing glasses, whose subject is the use of the Internet of things and sensor readings in government services. The Estonian government hired Welsberg last August to start a new project on introducing AI into various ministries in order to speed up the provision of services to residents of the country.

He says that AI will be critically important. “Some people worry that if we reduce the number of officials, the quality of service will suffer. But AI will help us in this. " About 22% of Estonians work for the government; For European countries, this is an average, but it exceeds 18% in the United States.

Siim Sikkut, director of information for Estonia, launched several AI-based test projects in 2017, a year before he decided to hire Welsberg. Welsberg says that Estonia has introduced AI or MO already in 13 different places where the algorithm replaced the officials.
')
For example, inspectors are no longer engaged in checking farmers who receive government haying subsidies every summer. The satellite images received by the European Space Agency every week, from May to October, are fed to the deep-learning algorithm that was originally developed at the Tartu Observatory. The images are superimposed on a map of fields, according to which farmers receive subsidies so that these places do not turn into forests.

The algorithm evaluates each pixel of the image, determining whether this portion of the field is beveled. Feeding livestock or partial mowing can confuse image processing; in such cases, the inspector leaves the scene. Two weeks before the end date of the mowing, the automatic system notifies farmers by e-mail or SMS, including a link to the satellite image of the field in the message. The system saved € 665,000 already in its first year of operation, allowing inspectors to drive less in the fields and concentrate on other matters of enforcement of law and order, says Welsberg.

Another machine learning application processes resumes of people who have lost their jobs and looks for employers. 72% of people who got a job with the help of this system remain working after 6 months - before the introduction of the computer system, this figure was lower and amounted to 58%. In the third example, children born in Estonia are automatically enrolled in the school at birth, so that parents do not need to enroll manually in a queue or call with school principals. The system works by automatically sending records from hospitals to local schools. It does not require AI, but it shows the distribution of automatic systems.

In the most ambitious of the existing projects, Welsberg and his team are developing a “robot judge” at the request of the Estonian Ministry of Justice, who is able to make decisions on minor requests that do not exceed € 7,000. Officials hope the system will help clear up a line of accumulated by judges and court clerks.

The project is in the early phase of development, and is likely to start at the end of this year in a test mode, starting with disputes over contracts. Theoretically, the two parties will be able to upload documents and other relevant information, and the AI ​​will make a decision, which can then be challenged by a human judge. It is still necessary to deal with many details. Welsberg says that the system may have to be adjusted after feedback from lawyers and judges.

Estonia is not the first to try to cross AI with the law, although, perhaps, it will be the first to be able to give the algorithm the right to make decisions. In some US states, the algorithms offer a length of criminal sentence. A chatbot from Britain DoNotPay helped to challenge 160,000 parking fines in London and New York several years ago. Tallinn law firm Eesti Oigusbüroo provides free legal assistance using a chatbot and generates simple legal documents for sending to collectors. She plans to expand her legal aid service by the end of the year, intending to find lawyers for clients in Warsaw and Los Angeles, said company director Arthur Fyodorov.

The idea of ​​a robot judge can work in Estonia, in particular, because its 1.3 million citizens are already using an electronic state certificate, thanks to which you can get a lot of electronic services, for example, sending tax returns, or vote in elections.

Government databases are interconnected using X-road digital infrastructure, which facilitates the data exchange process. Estonian citizens can also check who exactly received information about them on the government digital portal.

The transition of Estonia to digital government services has not been without glitches. Third-party experts found in 2017 vulnerability in the Estonian ID card system, which led to rather unpleasant consequences; it was fixed, and the cards [almost 750,000 pieces] were replaced. However, government officials have argued that there has not been a single major data breach in the country since the country began to digitize in the early 2000s. In 2016, more than two thirds of the country's adult residents sent various documents to officials via the Internet, which is almost twice the European average.

“Access to the most private and confidential things is not with the government, but with banks and telecommunications companies,” says Tanel Tammet, a computer science professor at Tallinn University of Technology. Tammet is a member of the Estonian government's AI implementation team, which is due to release a report on this project in May, and propose another 35 demonstration projects related to AI by 2020.

David Angstrom of Stanford University, an expert on digital government, says that today Estonian citizens may be able to trust their digital data to the government, but this may change if one of the AI-based decision-making systems suddenly goes bust.

Some agencies in the United States, for example, the Social Security Administration, use AI and MO to speed up the sorting and processing of data, and the Environmental Protection Agency uses them to determine which industries need to be checked for environmental pollution. However, Engstrom says that working together between different branches of the federal government in the field of AI is too slow, mainly because each agency has its own database format and it’s not so easy to transfer data between them. “We have not yet reached this level,” he said.

Engstrom with a team of law and computer science students at Stanford are exploring how best to use AI in US government agencies. Soon, they will report to the United States Administrative Association, an independent federal agency, to which recommendations have been received on improving administrative processes.

He does not see opportunities for the emergence of a robot judge with AI in the US in the near future. There is no national ID card system in the US, and many Americans are afraid of an “all-seeing government.” “The proper legal proceedings are spelled out in our Constitution, and it has something to say about the fully automatic decision-making by a government agency,” said Angstrom. “This can be an obstacle even if there is a possibility of an appeal to a person.”

Nevertheless, Engstrom believes that the time will come when the AI ​​assistants will be able to issue the relevant laws, precedents and all the necessary information for making a decision to the judges. “AI promises a more consistent approach to solving problems compared to the current one,” he said. “And, perhaps, the system under AI control will work more precisely than the system where people make decisions.”

Its disadvantage is that the quality of AI will depend on the quality of its implementation. For example, algorithms that offer terms of imprisonment have already been criticized for bias towards blacks.

“You also have to worry about the bias regarding automation,” says Ingstrom. The more decisions cars make, the less people will make their expert contribution to the system, he says. "This is one of those frightening things that privacy advocates and supporters of quality government are worried about, seeing how the state is gradually turning into a figure."

But for the time being, Estonian officials like the idea of ​​AI, which is able to solve simple disputes, and leave animate judges and lawyers more time to solve more complex problems. The introduction of AI in public services "will allow us to specialize in what will never be available to cars," said President Kersti Kaljulaid at the recent North Star AI conference held in Tallinn. “I want to specialize in being a sympathetic and warm person. And for this, our AI must be safe, and this security must be provable. "