News
ARTIFICIAL INTELLIGENCE IN PUBLIC ADMINISTRATION
Professor Iscte Technologies and Architecture
ISTAR-ISCTE Researcher
It is intended to understand the issues related to data, often large volumes of data currently not being used in several public administration entities, or to help with digital transformation and automation processes.
What is this project, and how is it connected to the AI Competence Centre for PA (IA>AP)?
The MAIPro project – Monitoring and Alert for Project Non-Compliance, based at ISTAR-Iscte, follows another, the IA Incentives (both with IAPMEI, the first also with the participation of AICEP). The initial project came to us through our colleague, Professor Ricardo Paes Mamede, who directs IPPS-Iscte (Institute of Public and Social Policies). It brought together a multidisciplinary team with professors from the Department of Information Sciences and Technologies, the Department of Quantitative Methods for Management and Economics and the Department of Political Economy.
During this work, the idea of creating a Competence Centre in Artificial Intelligence for Public Administration (IA>AP) arose because, on several occasions when we talked about the project, we were contacted by AP, either to implement similar solutions or with requests related to the area of text mining, for example, for email screening. Over the past two and a half years, we have been trying to give the initial impetus to this AI Competence Centre for AP to be able to respond to those requests.
What stage is this Competence Centre at?
This Centre still needs to get a physical location; we have an online page (https://iaap.iscte-iul.pt) with information, and, for now, we are a group of people with a common goal. In addition to the many collaborations of colleagues from various areas, it is essential to highlight the contribution of the members of the initial installation committee: Ana Almeida, Elsa Cardoso, Ricardo Ribeiro and Francisco Guimarães. How this Centre will operate after the installation phase (from January 2024) has not yet been defined. Still, it will inevitably maintain its nature of technology transfer and support for learning in this area. Fundamentally, it is intended to understand data-related issues, often with large volumes currently not used in several PA entities, or to help with digital transformation and process automation topics. Sometimes in the PA, there are a lot of questions about what can be done with the data you hold, and what we propose is: lend us the data, and we will see to what extent you can derive benefits from that data that will be useful to you in your mission. In this way, this Centre also contributes to improving teaching in AI by exposing students in the area of technologies at Iscte to real problems during projects and dissertations. This is also a relevant asset.
Is what you develop with IA>AP more the way of thinking and planning the operation?
Some call this the try before you buy. We have no costs other than some time to give us the information and explain a few things about the context in which the data is used. There are cases where mastery of knowledge is very, very important. Sometimes, it takes a long time for the team to be comfortable with the terms and processes in question, but "training" is the only cost that public entities have in these experiences.
This is what happened in the projects IA Incentives and MAIPro: based on the information from applications for European funds, from 2014 to 2019, we tried to train a system to predict the risk of project cancellation, i.e., whether they would either end correctly or be cancelled during the process. The forecasting capacity for some projects proved to be quite reasonable (around 70%). The result of this work points to the feasibility of a valuable tool to assist technicians in their work.
We also made an initial prototype for mail screening with the General Inspectorate of Justice Services (IGSJ). Many complaints are manually directed to other services, including external services, and based on the redirection history they provided, we tried to gauge how effective it would be to automate part of the mail redirection process. The developed component seemed feasible, but developing and testing other components to thoroughly evaluate a solution would be necessary. This is work that we hope to continue.
Creating functional applications integrated into the entities' IT systems is not feasible. It is something that is then left to the entities. That is why it is always essential to manage the expectations of PA entities in this regard. Usually, there is the idea that we will make an application functional, but that is different. A functional application requires maintenance and operational guarantees that we cannot provide. Still, the entity has an idea of what it can ask of those who come to make the application and can also contact us to help pass on knowledge to those who make that application.
Although we often call Public Administration Artificial Intelligence, we start with issues related to the information systems used, the quality of the information, and the processes that allow us to use the available data later.
MAIPro's team is multidisciplinary. What skills do its members have?
The initial team (in the IA-Incentives project) had Professor Ricardo Paes Mamede (from the Department of Economics and Public Policy), Professor Raúl Laureano (from the Department of Quantitative Methods for Management and Economics) and Professor Ricardo Ribeiro (from the Department of Information Sciences and Technologies), a specialist in Natural Language Processing.
Also, part of the team who moved with me to the MAIPro project was Professor Elsa Cardoso, who is more connected to visualisation, data quality and business intelligence systems; Professor Ana Almeida, who is (like myself) in the area of Machine Learning and its applications. In these projects of European funds, we also collaborated with Dr. Susana Fernandes, who, having much experience in the project area, gave an invaluable contribution to both projects. We also had grant holders from various areas, who formed a fantastic team.
Although we often call Public Administration Artificial Intelligence, we start with issues related to the information systems used, the quality of the information, and the processes
Have you seen any results from your work being implemented?
These projects take some time because there is the phase of experimenting with solutions, then a phase for the entities to understand how they will use these ideas, order the product, and put the product into operation. We may not have news of the end of this whole process. However, I believe it has been very little time to reach the implementation of a tool that directly impacts in the cases we have worked on so far.
In the presentation of MAIPro, it is stated that it intends to foresee in advance the possibility of non-compliance with temporal or financial goals through a system capable of generating alerts. Has this been possible?
We have been able to predict which projects have the greatest potential for approval and predict with some accuracy which ones are at greater risk of cancellation. As for predicting temporal and financial slippages, the initial results were disappointing. We still need the necessary data.
Does the Competence Centre for Artificial Intelligence for Public Administration aim to be a platform, an open door, for requests from PA in these domains?
Yes… We have tried to publicise the centre in academic circles and beyond: in conferences and workshops, which are somehow connected to the PA. We have tried to mark our presence and present what we do and how we are ready to receive these requests. This has generated several contacts that have borne fruit. We already have a large number of dissertations based on PA