A look at the discreet and indiscreet applications of AI and the baggage of challenges that comes as a bonus.
By Renata Neves
Source: Freepik
Lately we have been hearing a great deal about Artificial Intelligence (AI), which is apparently a technology built to revolutionise the world.
We are also witnessing a huge movement of developers and companies using the power of AI to improve internal processes, understand their customers and develop creative solutions for every problem.
We already have AI to talk, create, learn and organise. Yet we still know little about it, and it is unsettling that the technology can imitate a human being.
And, like a human being, AI has shown some flaws that can produce setbacks even as it surprises the market with new possibilities.
There are many controversies we need to put on the table, and so in this article we will set out what Artificial Intelligence is and which questions need to be considered. Shall we?
What is AI?
Autonomy is the key word for understanding what Artificial Intelligence is and how it works.
In other words, we can say it is a combination of computer science, data and the simulation of human behaviour to create an autonomous technology, capable of creating, suggesting, correcting and even conversing.
AI has in fact been under development for some time and is now becoming popular as we discover its full potential.
It is present in assistants such as Alexa and Siri, in social media algorithms, in Google’s search engines, in apps built to make a job easier, and so on.
These are systems in constant development and learning, so the more a user feeds data into them showing what is “true” and what is “false”, the more skilled they become.
Artificial intelligences can be shaped until they respond as expected and, for that to happen, they can be developed through two learning processes.
The first is called machine learning. In this method, the AI works with large quantities of data, collected or entered, which are analysed to generate new hypotheses.
The answers are not always precise and correct, but the more quality data is entered, the better the machine’s answers and generalisations become.
Then there is deep learning. Here, the technology uses Artificial Neural Networks (ANNs), and yes, they are similar to our own. It is a method with more knowledge power and mathematical models.
Through ANNs it is possible to solve more complex problems, and the learning process is carried out with a training phase and a testing phase. In training, various examples and adjustments are applied. In testing, the system is evaluated.
If you are interested in exploring machine learning from a more technical and mathematical viewpoint, we recommend the article “Conceitos sobre Aprendizado de Máquina” by M. C Monard and J. A. Baranauskas.
Applications of Artificial Intelligence
Today, the application of artificial intelligence is increasingly wide and varied.
As we said earlier, AI has been supporting the performance of certain functions in the market, and has also been developed to deliver major results with speed and convenience.
Starting with healthcare, AI solutions have been presented that make diagnoses and suggest possible treatments for doctors. These are tools to speed up health processes, from the suspicion of a problem through to the best way of resolving it.
It is worth stressing, though, that searching the internet does not replace consulting a specialist. AI needs to work alongside doctors, so do not look for solutions on your own. All use of technology in healthcare is supervised and evaluated.
Speaking of security, you have probably heard of technologies for simple home monitoring.
But beyond that use of Artificial Intelligence, major projects are being carried out based on failure analysis. Some AIs work in cybersecurity, predicting possible failures and attacks.
One of the great challenges of education is the lack of individual attention. That is why some systems are being developed to help students study more productively.
Teachers can also benefit from AI through support in preparing lessons, organising activities and improving distance communication.
In retail, Artificial Intelligence comes in strongly in the collection of quality data. In this way, retailers can better optimise processes, improve the customer experience and even price their products effectively.
The same applies to the business field. But we also have studies using AI to analyse the current market.
For example, the article “Explorando a forma fraca da (in)eficiência de mercado por meio de algoritmos de inteligência artificial” is a study using stock exchange indices to make forecasts with AI.
The study “Identificação de falhas por meio de inteligência artificial em dados ecobatímetros”, meanwhile, is of interest to agribusiness, which also uses the potential of the technology to solve problems, support processes and optimise resources.
If artificial intelligences are capable of assisting processes, then the field of Human Resources can also be — and is — helped in managing people. The technology can analyse employees’ development, export reports and create creative solutions for managers.
A study that illustrates the use of the technology in finance is “Identificação de evasão fiscal utilizando dados abertos e inteligência artificial”, which reports on the solution used in Goiás to identify tax evaders.
In Law, too, AI is seeking solutions for those who need accessible justice and for all the intermediaries.
Legal processing, service automation, regulation, decision-making and other functions are being carried out with the help of AI, if not entirely by it.
Meanwhile, in marketing and communications, even creativity can be automated.
Creation and production work can be developed by AI, as can automations, social media algorithms, process organisation and platform integration.
See also this study “RetificAI: Um verificador de notícias falsas com base em IA e Aprendizado de Máquina”.
Even transport and cars are using artificial intelligence. Autonomous vehicles and automatic braking, for instance, are AI-based technologies combined with other tools to provide more safety and comfort on the road.
Finally, we are also talking about thousands of platforms with specific (or not so specific) purposes, developed to perform any task. See more examples below:
Source: Instagram [Free translation of the left-hand column: Text; video; audio; face; legal; voice; meeting; productivity; code]
What are the limits of AI?
It all looks very fine and interesting, but since nothing is perfect, AI has its low points too.
It is important to remember that our reality is made up of various structural problems, such as racism, racial mixing, inequalities and other issues that can be amplified by Artificial Intelligence.
This generates a series of ethical questions and problems in the system that will only be noticed late on. After all, AI works with constant data collection and loses its neutrality as it is fed particular patterns.
We have already seen deep-learning AI acting with racial discrimination, for example, because its training is probably disproportionate and unfair, in the same way our social processes are.
AIs are acquiring knowledge from biases — that is, parameters based on injustices for or against something.
So if it is trained under a bias, it stops being neutral and starts making decisions that can harm users. This is a recurring discussion about ethics in technology.
AIs that work with facial recognition, based on deep learning, are the first to run the risk of being developed under skewed biases.
They depend on a database, training and testing to identify human characteristics that developers and platforms decide upon. So these problems arise from the coding in computer language right through to the collection of data in a reality full of social prejudice.
After all, who is the “standard human” defined for the AI? Or which characteristics are most seen on social media?
It is not only facial recognition that faces this difficulty: any system that has to define a “standard human” in order to make decisions can be undermined by biases of race, gender and ethnicity.
Recruitment processes, diagnoses and beautification will always run the risk of using a perfect model to justify their answers.
Amazon had this problem in an AI-based recruitment process, in which it excluded women from selection on the grounds of the company’s pattern and context, in 2014.
Data also has a shelf life, which means collections are made in different conditions and contexts.
It is therefore easy to teach a machine to make decisions that are not of interest today. This is another way of widening inequalities or generating answers considered immoral.
Moreover, dealing with the problem of bias in AI is a challenge for developers; many problems are identified only after years of use, or only when a considerably large volume of users has fed the system.
When they are found, they may even require shutting the AI down because of harms from use that were not anticipated. Such was the case with Tay, a Microsoft chatbot developed in 2016 that was released to the public on Twitter and was completely corrupted in less than a day.
Another problem that arises with Artificial Intelligence concerns the principles of data collection and use, running directly up against the data protection set out by the LGPD (Lei Geral de Proteção de Dados)
The more data collected, the better AI works; however, that data may be personal and sensitive, and if it is not handled properly it causes harm to users.
So data such as name, email, telephone number or the most precious kinds (such as CPF, card number and so on) are gathered by artificial intelligences and become liable to malicious use or to a leak.
The LGPD requires companies to be clear and transparent about the use of each piece of data requested, and obliges them to provide security.
The care brands need to take when using Artificial Intelligence has to be maximised in order to prevent people’s privacy from being weakened.
The future of Artificial Intelligence for us
The truth is that we are already embedded in a world automated by Artificial Intelligence — in other words, there is no going back.
We use, or are affected by, AI in our professional and personal lives, so it falls to us to shape its use so that it is more beneficial than risky. Whether through more training or more analysis.
Today, AI needs monitoring and care, whether for ethical reasons or for privacy. Even so, the idea is that it will keep developing and widening its reach.
There are also many opportunities to explore in every sector, cutting the time, energy and money spent.
The ease and autonomy of Artificial Intelligence are the most attractive points for anyone who needs results.
The use of AI promises improvements across sectors, the development of work and even the replacement of roles. Before long, results will be of higher quality and more reliable.
Now, as regards jobs, the use of AI may have a negative impact on the economy or open space for new business models, tasks and roles. Both may well happen.
Hence the need for companies to reinvent themselves constantly within their sectors.
We have picked out this study by Google on “Startups e o uso de inteligência artificial”, which reflects our social and economic questions in the market and their impact with AI.
In general, we hope that our society can develop and build awareness before applying a distorted reality to a technology. That would be the first and best step to take before any innovation.
Referências
-
MONARD, M. C.; BARANAUSKAS, J. A. Conceitos sobre Aprendizado de Máquina. In: Sistemas Inteligentes: Fundamentos e Aplicações, cap. 4.
-
Explorando a forma fraca da (in)eficiência de mercado por meio de algoritmos de inteligência artificial. Revista Gestão e Secretariado (GeSec).
-
Identificação de falhas por meio de inteligência artificial em dados ecobatímetros. UFRGS.
-
Identificação de evasão fiscal utilizando dados abertos e inteligência artificial. Revista de Administração Pública (RAP).
-
RetificAI: Um verificador de notícias falsas com base em IA e Aprendizado de Máquina. SBC.