national court register
about project
The project aims to create AI which allows a computer to "understand" the representation of a legal entity and subsequently apply it in automated system's workflow.
What is legal representation? Specified in the law and statutes/contracts, the way in which an entity can effectively make statements of intent on its behalf. In the Polish legal system, in the vast majority, the representation of each legal entity is recorded in the National Court Register. The record of representation is freely established by the persons forming the entity within the limits of the law. This record is captured in natural language. The computer does not understand records of natural language. The project presented below intends to create conditions under which a computer can "understand" it and, consequently, apply it.
Why should the computer understand the representation?
Imagine a system of automated administrative decision-making supported by artificial intelligence: A request for a decision on behalf of a company arrives electronically at the office. Based on its content, without human involvement, a draft decision is drawn up. Then, based on the draft, the authority issues the decision. In such a procedure, human involvement would be necessary only during verification of the automatically prepared draft decision and for possible analysis of unusual situations. The use of AI to process an application in an administrative proceeding could significantly reduce the time it takes to receive a decision. Of course, a system of automated verification of representation can be useful not only in administrative proceedings, but also in automated contracting or litigation.
A system that drafts decisions would have to include at least two elements: 1. an electronic form of verification of formal requirements of the request, 2. an automated form of substantive analysis. This project focuses only on one of the elements of electronic verification of formal requirements - representation of legal entities. In order for the automatic decision-making system to work, it has to "know" who is capable of acting on behalf of a legal entity, it must "understand" what consequences the representation record in the National Court Register brings. In other words, it should apprehend whether person X bringing an application on behalf of entity Y can act alone or whether the application must in this case also be signed by another person Z. Of course, such a system of automated verification of representation can be useful not only in administrative proceedings, but also in automated contracting or litigation.
AI development
Understanding how presented AI works requires comprehending how it is created: AI trains from tens/hundreds of thousands of real data from the National Court Register. For each type of legal entity (e.g., a general partnership), it is necessary to properly process the representation records leading to multiple categories of representation. Each category of representation has its own features. E.g. category 13 for a general partnership has only one feature, i.e. joint representation of two partners. Ultimately, in practical application, the features of the category will determine exactly what person(s) can act on behalf of a legal entity.
What if the AI makes a mistake?
Let's remember that the whole system requires human supervision - ultimately human is the one who makes the decision based on the automatically prepared design. Artificial intelligence is only supposed to help him in this, relieve him of the burden of tedious handling of typical cases. There is also question for the future - is a person without legal training more effective in interpreting the representation record than artificial intelligence? It also seems that while dealing with legal norms achieving 100% effectiveness of AI is impossible because, ultimately, marginal cases will often be interpreted differently by different lawyers. In other words, for one lawyer the system will be 100% effective, while for another its' correctness will only be 99%.
scheme of preparing data for training
Familiarize yourself with the next phases of data preparation.
unusual forms of representation - selecting category
There are sometimes records in the National Court Register that are ambiguous to interpret even by a human. For example, record "board of directors" in the representation of a general partnership. The problem is that under the law a general partnership cannot have boards (including a board of directors). There are also single-word provisions in which representation has been regulated in an extremely original way. It seems inefficient to create additional categories of representation for such cases. We assume that these are those cases that require human intervention and are included in the feature of representation - UNCOMMON. The task of the system is to create a separate category for them, which will alert the user that in this case human interpretation of such description is necessary. As it was presented above, the assumption is that AI is subsidiary to humans. Of course, this category having the feature of UNCOMMON should be as narrow as possible.
When will it be possible to use this project in practice?
The project is under development. I hope it will be completed in 2023. Once the work is finished it will be possible to purchase a license to use the models or they will be made available as an API. As AI is developed for more categories of entities, it will be possible to run tests for them on this website. If you would like to discuss details, talk about cooperation or give me any suggestions about the project I invite you to communicate with me via social media.
test ai - general partnership
At this point, the test is only possible for a general partnership.
features characterizing the categories
Individual representation.
Joint representation of two partners.
Joint representation of three or more partners.
Joint representation of all partners.
Representation is regulated by naming the partner(s).
Joint representation with proxy.
With the advent of the condition, the form of representation changes. The condition is not financial in nature.
With the advent of the condition, the form of representation changes. The condition is financial in nature.
A resolution is required before any legal action can be taken.
An unusual form of representation.
categories
What does it mean that the representation has a "named" feature?
This means that the representation provision indicates the entitlement of a partner by referring to a specific name. According to the views expressed in the doctrine, representation in a general partnership can also be established by indicating the scope of action of a named partner.
Joint representation in a general partnership vs. third parties
According to the currently prevailing view of the doctrine and the case law of the Supreme Court, contractual regulations aimed at determining the method of representation are permissible. At the same time, the establishment of joint representation is permitted and binding in the company's external relations (among others - the resolution of the Supreme Court of May 30, 2008 (III CZP 43/08 Biul.SN 2008, no. 5, p. 5).
What is the effectiveness rate of AI for a general partnership?
The effectiveness of AI for the general company is over 96% on test datasets and improving all the time.
A POOL OF EXAMPLES FOR RANDOMIZED TESTING
~ 34000
category breakdown
The number of types of representation belonging to each category in after analysis of about 20 thousand entities.
Use AI to predict a category
AI will determine the category for the representation based on the register number.
Random general partnership
AI will determine the category for the representation based on the register number.
select category example
See the prediction for the selected category example. Use the slider to make your selection.
entity
data of the registered entity
prediction result
Who can sign the automated application/contract ect.
WHAT DOES AI ULTIMATELY "UNDERSTAND"?
Who can sign the automated application/contract ect.
created by Tadeusz Mięsowicz | 2022 | All right reserved