The Modal Institute specializes in artificial intelligence solutions
The Modal Institute's Action Cases
A X-SENSORS is developing a digital platform for simulation and complete immersion in augmented reality of production environments for manufacturing industrialists. This environment requires the digitization and rendering of the real industrial plant (factory), as well as all equipment (machines), including its technical specifications. Based on these specifications, X-SENSORS 'proposal is to automatically configure the production line in the simulation environment and, ideally, to implement this configuration in factories, in an automated manner. This solution has the potential to reduce the time required to set up the production line and, consequently, increase the productivity of the industry, reducing downtime, eliminating configuration errors and improving efficiency. The number of variables to be considered for such an approach makes it impossible to use conventional solutions.
The Modal Institute used a combination of symbolic and connectionist artificial intelligence approaches to build the best solution to the problem, which presents itself in a deterministic way for certain cases and probabilistic for others. This situation led to a series of innovations to deal with a set of phenomena in the industrial area, among which stands out an ontological approach coupled with the development of specific solutions, implemented in logic programming language (Prolog). This approach was registered under the name of Ontoprolog (TM). The project continues in progress, with the optimization of solutions and the incorporation of new techniques and results. The number of variables to be considered for such an approach makes it impossible to use conventional solutions.
The NGO End Violence Against Children launched a call for projects that meet the following objectives:
- Detect, remove and report images and videos with sexual content or acts involving children and adolescents (often referred to as child sexual abuse material, or CSAM).
- Block adult access to children on digital platforms designed to sexually abuse them (often called online sexual solicitation or treatment).
- Stop the live stream of child sexual abuse performed in front of a camera (often called the live stream of child sexual abuse material).
- Prevent sexual abuse of children online before this happens, including prevention and solutions aimed directly at online sex offenders and adults with a sexual interest in children.
The number of variables to be considered for such an approach makes it impossible to use conventional solutions.
The Modal Institute, in partnership with the University of Ottawa, developed a methodology to identify patterns of textual approach to messages between harassers and children, which helps to prevent child sexual abuse online. The developed solution uses artificial intelligence resources (connectionist approach) to build the personality profile of the people involved in the chat, generating as a result behavior patterns that can be compared with those of known abusers. If the AI realizes that a certain message exchange presents a certain combination of risk factors, a human operator can be called upon to evaluate the messages and confirm (or not) whether it is the case of a police intervention.
In industries, one of the relevant cost factors is the electricity bill. Factories consume large amounts of energy for their operation, with the almost uninterrupted operation of large machines and equipment. In some cases, energy can represent up to 5% of the costs of a production line. Brasal Soft Drinks asked the Modal Institute for an exploratory analysis of the power consumption information of its soft drink factory in Brasília, focusing on the water cooling system used in the industrial plant. For this, it provided the electrical consumption records and the measurements of the sensors referring to the water cooling system of the last 12 months. The question to be answered was: is it possible to reduce the consumption of electrical energy used by the cooling system without it reducing the production of the factory?
From the records delivered, Modal's Artificial Intelligence team used a connectionist approach to discover patterns and relationships between the various events captured and available in the data. The exploratory analysis (first stage in the search for an answer) listed different behaviors that identified the moments of higher and lower consumption due to different aspects and situations verified by the sensors, leading to the conclusion that:
- Yes, it is possible to reduce the energy consumption of the cooling system without reducing factory production; and
- Continuous monitoring of the cooling system combined with the installation of new sensors will allow the automation of various actions that may lead to both an even greater reduction in energy and a lower need for maintenance of cooling equipment, reducing plant down-time and increasing service life equipment. The project moves to Phase 2, which consists of creating tools to automate these measurements and actions.
Concrete buildings (bridges, viaducts, structures, etc.) located in cities with certain climatic patterns are affected by diseases caused by these conditions - hence the term “concrete pathology”. This situation requires periodic monitoring and maintenance to ensure that the structures remain solid and without risk of collapse. The evaluation of these structures is done through the analysis of microscopic photographs of sections of the structures. It takes an average professional two months to identify and diagnose a possible concrete pathology, at a high cost and an even greater risk depending on the number of buildings that need to be examined.
In partnership with researchers from University of Ottawa, the Modal Institute was able to develop a Proof of Concept (PoC) with Artificial Intelligence using machine learning techniques (specifically, deep learning) and supervised training. The solution is able to analyze and identify if a series of pathologies occurs in minutes, reducing the workload of the expert and enabling better conditions for maintaining the structures. With this, it is possible to carry out preventive maintenance and identify possible diseases while they are still in the initial stage, significantly reducing maintenance costs and preventing tragedies. The result of this PoC was presented by the researcher Romualdo Alves at the 9th International Conference on Bridge Maintenance, Safety and Management (Melbourne, Australia, 9-13 Jul 2018), under the title “The use of machine learning techniques to assess damage in critical infrastructure”
Brazil - like many other countries - suffers from an excessive amount of laws, rules, decrees, ordinances, etc. that regulate, at different levels, the same things. It is common to find different acts applying different rules to identical or very similar objects, even when the acts are issued by the same level of power. When walking in a transversal way, the problem expands exponentially, to the point of contradictions, inconsistencies and voids in subjects of the most diverse natures. How can an organ:
- minimize the probability of legislating on something that already exists in a normative act so as not to contradict that act?
- know what is necessary to be revoked (totally or partially) for a new act to be effective?
- identify regulatory deficiencies that need to be addressed without contradicting other regulations that already exist in different bodies and levels of government?
The Modal Institute dedicated itself to this problem and developed a Proof of Concept based on the application of artificial intelligence algorithms that, using unsupervised textual analysis, can verify the semantic similarity between different acts, highlighting points of convergence and identifying acts that deal with the same object. With this, the legislator is able to be more clear about which points should really be addressed, which are already satisfactorily addressed in the existing rules, and which need to be revoked. This solution helps to reduce the time spent on research, improve the assertiveness of results, increase clarity and reduce legal ambiguity, increasing legal certainty.
A common problem for public figures is to assess the impact and quality of the message conveyed in a public speech or speech. In the case of parliamentarians, this situation gains an even greater dimension due to the impact that a statement can have. The challenge is to know, in advance, what impact a speech may have on the public and, later, what the real repercussions are with the press.
Artificial intelligence techniques aimed at textual analysis allow coupling a series of inferences about the text of a speech (or news, advertisement, tweet etc.). Among them, the following stand out:
- scentic analysis (detects the predominant feeling in the text)
- contextualized questions and answers
- keyword phrase detection
- detection of named structures (proper names, entities, acronyms etc.)
In addition, it is possible to use cross-references from the automatically detected key phrases to assess what has been discussed on the subject in newspapers, social networks and other sources of online information.
Instituto Modal carried out an exploratory analysis of the speeches of parliamentarians in 2018 in the Chamber of Deputies as a Proof of Concept of this approach, with results that can be seen in https://institutomodal.org.br/inteligencia-artificial-para-analise-de-discursos-parlamentares/.
When working on a large project that involves many people and different opinions and points of view, it is a relevant challenge to know and reconcile the different perspectives and interests. In the political arena, this is especially relevant when discussing bills, amendments and other parliamentary activities.
How do you know who thinks the same way? What are the political tendencies of each parliamentarian who approach or depart from the topic under discussion? What does the press reflect from the position of each one? What is the perception of a certain article proposed for the Law?
Using a combination of textual analysis and datamining, coupled with innovative artificial intelligence techniques, the Modal Institute put together a Proof of Concept to analyze the Proposed Constitutional Amendment to Social Security while its text was still under discussion. From the text proposed for the Law, the PoC automatically identified the theme of each article and correlated it with the speeches of the parliamentarians at the time of the procedure. Simultaneously, it also correlated both the themes of the articles and the speeches as it was broadcast in the press. The result brought the availability, in real time, of crossing themes proposed in the amendment's articles, what each parliamentarian said on the subject (and possible changes over time) and how the media reacted to each new pronouncement or alteration of the text.
How to monitor the development of the contagion of a worldwide pandemic, extracting information that helps to contain the outbreak and save lives?
With the unprecedented health crisis, the Modal Institute has built a panel with a daily mapping of confirmed cases, deaths and recoveries in the world. The approach is exploratory and built “on the fly”, that is, at the same time that new information is made available.
Check out the panel at: https://institutomodal.org.br/covid-19-world-panel/