It is nothing new that neural network and machine learning tools have been in the hands of the qualified public for some time, as “challenges” have been carried out for years, where their own AI is compared to the performance of online players ( here ), but with computational capacity more than doubling in shorter periods. With the focus on AI by technological giants, we see that the capabilities and complexity of the functions performed are gaining prominence. What was previously considered a mighty machine, which struggled to create generations with more significant numbers of entities for deep learning, has become a typical computer that many of us have on our desks.
Simplistically, the process of creating and training a neural network consists of training through “rewards,” where data is provided as an input to the machine, which carries out the processing and presents a result or output; this post-processing data is compared to an expected outcome, whether right or wrong or a value stipulated as ideal, then with the acquired data the AI tries to get closer to that result. This process requires large amounts of data for greater precision; the processing necessary to supply the layers of a neural network is still considered significant and continues to be a laborious process, but more tangible than ever. (1)
The evolution of technology is one of the points in the battle for the popularization of this type of method; the other is the creation of more knowledge on the subject, facilitating the entry of new enthusiasts without the need for training or many years of study, to a certain extent. This process ends up supplying itself with greater access to technology, and more people work on this topic, creating more knowledge and reducing entry barriers so that even more knowledge is generated. (2)
Below are examples of how this tool can be used in different sectors:
Health:
- Medical diagnosis: Neural networks can analyze medical images, such as X-rays and CT scans, to help diagnose diseases more accurately and quickly.
- Drug development: Machine learning can identify new compounds that can become medicines, speeding up the development of new treatments.
- Genomic data analysis: Neural networks can analyze large sets of genomic data to identify disease patterns and mutations, aiding in personalized diagnosis and the development of gene therapies.
Finances:
- Fraud detection: Neural networks can analyze financial transactions to identify patterns that indicate fraudulent activity, protecting banks and customers from fraud.
- Market analysis: Machine learning can analyze large sets of market data, such as stock prices and financial news, to identify trends and assist in making investment decisions.
- Risk management: Neural networks can assess customers’ credit risk, assisting banks in granting loans and mitigating risks.
Manufacturing:
- Predictive maintenance: Machine learning can analyze data from sensors on industrial machines to predict failures before they occur, avoiding production downtime and unnecessary costs.
- Quality control: Neural networks can inspect products during the production process to identify defects and ensure their final quality.
- Logistics optimization: Machine learning can optimize delivery routes and manage inventory, reducing costs and increasing supply chain efficiency.
Retail:
- Product recommendation: Neural networks can analyze customer purchase history to recommend relevant products, increasing the likelihood of purchases and customer loyalty.
- Sentiment analysis: Machine learning can analyze online product reviews to identify customer sentiment and make decisions about products and services.
- Demand forecasting: Neural networks can predict product demand, allowing retailers to optimize inventories and avoid stock-outs or excess inventory.
Even with all the evolution and diverse applications of neural networks, an old AI problem reappears: the possibility of algorithmic bias. Because it is a complex process, it often becomes even more challenging to detect, and there are more possible reasons. For the existence of this bias, be it a biased database or an algorithm that has bias within itself, in more severe cases, even creating discrimination against groups of individuals, generating unfairness in the responses of a machine that should value efficiency. The process of algorithmic discrimination can affect areas, including healthcare, by perpetuating stereotypes or screening CVs in job recruitment, causing direct negative consequences in people’s lives, eroding trust in technology, and closing the eyes of many to possible functionalities. of AI (3)
The most common way to mitigate bias is to increase the scope of databases. Still, even with the broadest database, human error when creating the algorithm or defining the objective of Artificial Intelligence can easily “taint” the database’s functioning. A.I research by tech giants is still needed to avoid algorithmic discrimination as much as possible; however, based on the work of a human, the complete elimination of this problem appears to be quite complex.
With all these advances, which have brought the democratization of an entire area previously restricted to specialists in companies capable of giant investments, the role of researchers in reflecting on ethical action is vital, pointing out challenges whenever they are encountered, at least in an attempt to guarantee a more promising future regarding the activity of AI’s, which can be a path to solving complex problems and transforming different sectors of society, thus requiring enormous care when dealing with the topic, since without due attention the tool can be applied incompletely, or even to the detriment of human development and the quality of life of the general population.
References
(1) IBM Brazil. “Neural Networks: How They Work and Why They’re Important.” IBM. Available at: What is a neural network? | IBM. Accessed on: 17 June. 2024.
(2) AMAZON WEB SERVICES (AWS). Neural Network. Available at: https://aws.amazon.com/pt/what-is/neural-network/. Accessed on: 17 June. 2024.
(3) Algorithmic Bias in Data Interpretation. Smart Cities, 2024. Available at: https://cidadesinteligentes.org.br/artigo/ biases – algoritmico -na- interpretacao – de-dados /140 . Accessed on: 17 June. 2024.
