Do androids dream of electric sheep? Does artificial intelligence dream of becoming general, strong, intelligent as a human being? Of course I do. From the Technological Institute of Aragon (ITA) the artificial intelligence (AI) of the present, applicable in many directions, but one that passionates Rafael del Hoyo, responsible for the research line of AI and Cognitive Systems, is the most complicated: the AI of the future.
"Normally, AI is taught to do something right, for example recognizing license plates or making summaries, but the challenge is to make machines completely intelligent, totally autonomous systems, the idea is that the general AI is more human, with the ability to learn on its own." And, to this day, this "is not only investigated in the United States, China or Germany, but from the ITA we are working on it, and bringing results, along with a team of neuroscientists based in the Pyrenees." This is the young company The Mindkind, located in Castejón de Sos (Huesca). Together, they seek to create video game characters that carry their own life within it, able to learn, reason, make decisions..., so similar to a human being that it makes it difficult to distinguish whether they are real. Behind the scenes: cognitive systems, human behavioral modeling, physical modeling of how the human brain behaves and new generative AI tools.
Twenty years ago, there were only two AI people across the ITA. Today, it's 30. "We are the weird team of the institute," says Del Hoyo, because our profile is mathematics, physics, computer engineering, some industrial engineering, telecommunications, we would like to get people of humanities... and we collaborate with the University of Zaragoza, Circe... and people from outside. The man-machine interaction is increasingly multidisciplinary."
Hybrid AI
At the University of Zaragoza, the new EINA chair in Artificial Intelligence and Sustainability has just been created, whose objective is to develop hybrid AI techniques applied to sustainability in two fundamental sectors for the economía Aragonese and national economy: construction (especially large infrastructures) and automotive.
Its director, Elías Cueto, explains that "hybrid AI uses a mixture (hence its name) of human knowledge, developed through centuries of scientific and artificial learning, generated by data. This ensures, on the one hand, that the resulting AI will be much safer and more robust, as it will always respect the scientific knowledge already available. On the other hand, the incorporation of existing human knowledge saves on the amount of data and computing that will be necessary to use, so that it is faced with a type of green or sustainable AI itself."
" AI is called upon to suppose a revolution in the industry of the future," he says, convinced that it will help use less materials (especially non-renewable ones), consume less energy and generate less waste. How's that? From the product design phase, "enteraly virtual, through an optimal useful life in terms of resource consumption (energy, materials...), to the extension of that useful life through intelligent monitoring."
A specific calculation cluster
At the same time, however, because of the need to train algorithms with millions of data, AI consumes a huge amount of energy and computing resources. At the Institute of Biocomputing and Physics of Complex Systems (BIFI) of the University of Zaragoza they already address this challenge, theoretically and very incipiently: "If the brain consumes the same energy as a light bulb, no more than 20-40W, there must be another way to do things beyond the use of brute force, other forms more efficient for "compputing," says Yamir Moreno, director of the BIFI.
But, as that new AI arrives, the calculation power needed to move forward grows on the Rio Ebro Campus. In November, the new Caesaraugusta IV supercomputer, the brain of Unizar, was premiered, which will be overcome very soon. The BIFI institute will invest one million euros from the Ministry of Science's funding for unique scientific-technical facilities to acquire and launch a calculation cluster specifically designed for the development of AI and its applications. This new calculation infrastructure should be underway by the end of 2024 and will serve the researchers of Unizar and the Spanish Supercomputer Network, as well as to companies that require these services.
"Without a calculation capacity like the one we will have, Moreno says, it is very difficult to move forward and be in the first division of AI research." The new cluster will make it possible to take "a significant step forward, because it will provide the university community with a first-class infrastructure for basic and applied research." But not just that. "It will also facilitate Moreno's generation of new projects with the industrial fabric, which will ultimately allow companies to optimize processes and resources and be more competitive. In the biomedicine sector, the new calculation infrastructure is expected to allow more data to be analysed and faster reach caliber targets for advancing the diagnosis and treatment of diseases such as cancer.
Applied to complex systems that are very complicated to analyze, AI is unique, it is able to detect patterns that would escape traditional techniques and also mathematically model the problems, to put them in the computer. Something applicable to the study of social systems, human behaviour, analysis of climate data and detection of patterns of certain diseases.
BIFI is immersed in the European Katy project, which uses data from patients with the most common renal cancer subtype and, applying AI transformative tools, aims to improve diagnosis and develop more accurate therapies tailored to each patient. On the other hand, the Clinical Diagnostic and Drug Delivery group of researcher Olga Abián takes advantage of the precision and efficiency of AI to process large amounts of experimental data so that it can "identify early signs of cancer by detecting changes in the thermal properties of biological samples, such as blood." By improving early diagnosis, we would see cancer locking up from its inception.
Markets, waste or fraud
The AI application field has no doors. In collaboration with a group of economists, the group of Networks and Complex Systems that Moreno is helped by the neural networks applied to graphs (GNN) to better understand the dynamics of the Spanish market of equity funds. Thus, "in addition to the quotation data, we can integrate additional information such as financial news, economic reports and market sentiment, to obtain a more holistic and accurate view."
From money to waste management. The Smart Waste Pick Up project, from the Computer and Data Science group, creates optimal waste collection routes based on big data techniques and models of "automatic learning" (automating learning) techniques. "The routes are dynamically generated daily according to the filling criteria of the containers measured automatically and in real time, as well as the capacity of the vehicles, the location of the landfills or the work shifts of the operators," says David Iñiguez. This achieves an estimated savings of up to 40-50% of the waste management costs of public administrations.
One of the things that motivates Rafael del Hoyo the most is when engineers are mixed in the teams with professionals from other fields. This is the case in an ITA project with law experts. "Together with the General Directorate of Taxes of the Government of Aragon, we are detecting with small AI and large frauds in the payment of taxes," he says. Natural language analysis technology applies here to scriptures. "It's complex to understand human language, in this case, legal, but we're trying to teach examples of machine fraud to find other similar cases."
The processing of natural language is the research field of the Vivolab group of the Institute of Engineering Research of Aragon (I3A) of the University of Zaragoza. Together with 13 other countries, they participate in the Esperanto project, which, they say, will contribute to the development of the next generation of AI. This is an artificial intelligence project for low-income speaking technologies. "The objective of the project is to use AI to reach those fields or areas that do not have sufficient resources for its development, that acts as facilitator and allows to reach audiences or sectors that are difficult to reach without this technology," says researcher Alfonso Ortega of the Vivolab group. This collaborative research program will establish metrics, protocols and standards applicable to the development of the next generation of artificial intelligence algorithms, more accessible, human and transparent, since "something crucial to the trust and social acceptance of AI, is to make these systems explainable and interpretable."
All I3A activity around AI is collected at one of the Vanguard Laboratories launched at this university institute. Among other groups, Howlab works primarily on the Internet of Things; the computer-by-computer viewing group is a leader in its field at the international level; while the work of the Graphics and Imaging Lab group, of graphics and computational image, has also gained a great international impact.
Cognitive twins
From the AMB group of Applied Mechanics and Bioengineering, they apply AI to scientific learning as part of the project "Discovering and learning models based on physics." This is "a fundamental aspect in the development of the so-called digital "digital" gels - highlight Elias Cueto, virtual copies of a given product or industrial installation." This copy is fed by experimentally obtained data on the really existing physical gemelo. "We are developing cognitive meme, that is, twins who are able to learn for themselves, point - and correct the models with which they were developed, if they began to deviate from the experimentally observed."
Going one step further in product design is what has been proposed another of your projects: Automatic learning informed by physics for the virtual haptic prototyping. "We are looking at an aspect that is rarely considered in the design of products: its appearance to the touch "recalca Cueto." We know how to make virtual prototyping by simulation in aspects related to product resistance, even with its physical appearance, but it is very difficult to simulate the sensation that will provide to the touch. However, this is fundamental in products such as cars - let's think about dashboards, for example, or cosmetics - where the sensation provided by a cream is fundamental to its acceptance by the consumer." In this project, artificial intelligence is applied to simulate the haptic sensation (i.e. tactile) of prototypes of future products.
Stuck to the manufacture of products are projects such as Data4zero, in which the Technological Institute of Aragon and Fersa participate. The goal is the zero defect and "now it is no longer just about detecting the error that makes a piece have a failure and predict when it will happen, but, thanks to artificial intelligence, to understand why a failure is going to occur in a manufacturing industry," says Del Hoyo.
To achieve this, they mediate models of machine learning and deep learning, tools that predict future and explanatory techniques that allow us to explain why the machine has predicted something. "It's interesting to take that algorithm and explore it, he insists, and turn it into something an engineer can understand to prevent failures or lengthen the lives of the pieces."
That part that might seem like magic of artificial intelligence also doesn't escape scientific study. In the theoretical field, the BIFI institute inquisses into "how the science of complex systems can help you understand how they work - many AI algorithms that, to this day, we know are good results, but we don't know why, that is, they are black boxes," Moreno says.
An opportunity for Aragon
After composing this panoramic view of what is cooked in the stoves of artificial intelligence in Aragon, it is time to make assessments as a whole. For Yamir Moreno, "Aragon is on the very right track; there are several very strong groups that are increasingly developing basic science projects and industrial applications with impact, although, despite its potential and future relevance for science, society and industry, it is an area that is still underfunded."
According to Rafael del Hoyo, our Community "is very well positioned in relation to its size and population and the investment that is made." He sees AI as "an opportunity to grow faster, we are in the middle of a change and the Aragonese are good at getting on the wave and seizing the opportunity."
For his part, Elías Cueto considers that "Aragon has a remarkable research activity in the field of AI, especially with regard to applications. Aragon's AI is an AI that, far from having the astronomical budgets imposed by the AI that is being developed by a handful of large companies, mainly Americans, is very attached to the ground, industry and livestock." Looking ahead, "the prospects are very good, value. Both government agencies and private industry are betting on AI, aware that we cannot depend on the United States or China and that Europe must have a way of developing AI."
With realism, he points out that, at the moment, we live immersed in a sense of excitement around AI "that may end up extinguishing in a more or less short time, but it is our responsibility to show that the AI that we develop serves society and is not realized for the benefit of a few, that it is worth investing in it and believing that Aragon, Spain and Europe have a role to play in a tremendously competitive international landscape."
