by The first thing one imagines when thinking about artificial intelligence (AI) is a robot or some sophisticated system close to the fictions that have accompanied the human imagination for decades. With the proliferation of generative artificial intelligences, such as ChatGPT, Copilot or Gemini, it is easy to conclude that this is what AI consists of: a system with which we can establish a kind of conversation and that quickly provides answers to the questions we present.
The interest in emulating skills considered human in AI has been present since the beginning of computing. What has changed is the impact of these systems on our lives. Today the availability of data and accumulated computing capacity makes it possible to have complex interactions with generative AI that were previously difficult to imagine. This is what Andrés Moreno, professor at the Faculty of Engineering of the Pontificia Universidad Javeriana, expresses.
It is not a single artificial intelligence
Artificial intelligence is the ability that machines or computer systems can have to perform complex tasks that would traditionally require human intervention. Phenomena such as learning, perception, decisions and problem solving are susceptible to being implemented autonomously by an AI-based system.
There are several forms of AI: machine learning, computer vision, robotics, and natural language processing. With the latter, we have interacted a lot in recent months.
Before the emergence of generative models, resources such as chatbots or virtual assistants have been used to improve the interaction experience with complex information systems, facilitating tasks such as scheduling a medical appointment or shopping online. Another example is search engines or databases; In the background we are interacting with the same information, we are simply using tools that make it easier to digest.
Generative artificial intelligence, the functional basis of ChatGPT and its analogues, consists of computer programs capable of understanding a large amount of data and responding to requests in a short time through an intuitive interface that facilitates consultation.
Unlike other AI systems, it is not easy to understand how generative models work. One can understand a cross-stitch embroidery or a crochet piece by turning the work over and reviewing the direction and meaning of each stitch; For experts, it is not so easy to follow the direction of the threads and understand how the system arrives at a certain reasoning, since the rules that lead to an answer are unknown.
Although it is possible to explain the model training process, just as it is possible to teach how to use the embroidery or crochet needle, we do not know the logic that the program uses to make specific decisions. It is a kind of black box that can worry many people.
This scenario has motivated experts to build practices on how to adopt and manage AI in a transparent and secure way. There are guides that advise on the monitoring questions that should be asked of a generative AI to ensure that the responses it produces are reliable and safe. Faced with the black box, guarantees are placed outside and not inside the system and thus ensure that it is being transparent and follows the rules that were trained to it.
While many security approaches view models as black boxes, there are also processes that examine them as white boxes, with the goal of understanding and possibly correcting some of the internal mechanisms of networks. However, these approaches are usually more complex and there is still much to learn about these processes.
AI adoption
Artificial intelligence applications are becoming more and more present. In the financial sector, they perform predictive market analysis. In transportation, autonomous vehicles use AI to improve road safety and traffic applications guide the best routes. Even in education, smart tutoring systems are implemented to personalize teaching.
Professor Andrés Moreno, who works in academia and industry, comments that, with the adoption of AI-based systems, business models are changing. For example, technology companies that manage documents and whose market has been to safeguard information, see business opportunities in understanding what is in the data and taking advantage of the insights that AI offers into large volumes of information.
Needs arising from AI
The adoption of AI-based systems requires human capital specialized in these technologies. The industry is increasingly leaning towards emerging technologies; However, the talent shortage in the country poses a challenge.
For Andrés Moreno, there is an urgent need to train people in the use, adoption and understanding of these technologies, since depending on the import of products is not strategic for the technological sovereignty of the country. Furthermore, the representativeness of the data used to train these models is an issue that must be present in all reflections on AI, since they mostly come from texts in English and contain the worldview of authors in developed countries. It is crucial, Moreno emphasizes, to develop our own systems adapted to the particularities of our contexts, biases and cultural perspectives.
AI-based decisions
Although technological tools can facilitate information analysis and decision making, there is uncertainty about the ability of these systems to make autonomous decisions based on data. The possibility of evaluating the appropriateness of responses is a challenge given the opacity of its internal functioning.
“I like to say that these systems answer any question like any person would,” says Andrés Moreno, and the criterion of whether that answer is right or wrong depends on the expert, so “the important thing is what is done with what is done.” says the machine.”
Decisions should not be left to the machine, comments the researcher; AIs offer a broader view of a problem in less time than what people could achieve using only the processing capacity of their brain. However, the responsibility for making decisions must continue to fall to trained experts, as they are the ones who have the necessary knowledge and experience to do so.
AI-based health
Health problems have a lot of data available, explains Professor Andrés Moreno. We all generate data when we go to the doctor, examinations and diagnostic images are added to each patient’s medical history; This provides many possibilities for systematization and analysis that can contribute to the functioning of the health system.
Artificial intelligence applications in medicine have shown a notable contribution, especially in the automated analysis of various types of examinations, such as diagnostic images. The adoption of AI can be a valuable support, particularly in places where the availability of specialized personnel is limited, points out Professor Andrés Moreno.
Medicine benefits from artificial intelligence mainly in terms of processing capacity and rapid analysis of large volumes of information to make decisions. The rapid integration of knowledge about the functioning of the human body with information systems and the use of artificial intelligence allows us to understand pathologies and make public health decisions.
AI-based risks
With AI there is a risk that trust in decision-making by specialists will be lost. Any decision, human or artificial, is subject to biases that can be difficult to identify and create a false sense of impartiality. Given this, it is crucial that people are trained to interpret the information from the models and detect errors and biases, since in health, incorrect decisions can be costly.
Caution in the adoption of new technologies is necessary, the transition from proofs of concept to proven and precise products requires time and resources. This process needs political and regulatory guidance that establishes the operating framework for new developments and the limits that AI-based tools can or cannot cross.
Data security
There is a real concern to prevent the misuse of AI in critical issues such as health. Although there is no clear solution, many people are working to find an answer to this challenge, says Professor Andrés Moreno. Collaboration between policymakers and system developers is needed to establish safe and ethical use protocols, to establish AI fair play.
The tension centers on the advances that artificial intelligence can offer people in the face of restrictions and obstacles in regulatory processes. It is crucial to find a balance between the safe use of these technologies considering the ethical guidelines established by entities such as UNESCO to ensure ethical use of technological tools.
Given the lack of regulation of the adoption, implementation and practice of these technologies, the outlook is not encouraging. The safe, fair and ethical use of artificial intelligence, especially in the context of data security, has a lot of work to do.
AI-based replacement
One of the concerns that arises from any revolution is the possibility of being replaced by a machine, as happened to the people who operated fax machines, the door-to-door encyclopedia sellers, or the designers of the yellow pages. For Andrés Moreno, the possibility of this happening is latent, work tools are always changing, and although there is a risk that the profession or work will be transformed with the use of these tools, it is also an opportunity to improve and enhance the job.
Technological advances have generated changes and led people to move towards new activities as machines take on tasks previously considered fundamental. This change has created opportunities for new human activities and although the primary concern is people’s financial security and livelihood, considering alternatives such as a guaranteed minimum income can alleviate these tensions. Ultimately, if work were to run out, the researcher argues, we could explore new forms of social organization that allow us to find sustenance and use our time in more satisfying and meaningful activities. Maybe have time to explore our curiosity and sensibilities and find new ways to inhabit the world.
