Ethical and reliable Artificial Intelligence : what can we learn from the European proposal ?

AI has evolved rapidly in recent years . It is already being used in the most different areas of the economy and we are only at the beginning of its implementation that will change our lives in countless aspects. The potential of AI, therefore, is enormous and largely unknown. It is a fact that its benefits can already be seen and reach society as a whole.

The s citizens can be benef iciados, for example, with the improvement of medicine , with the comfort provided by appliances connected to the network (internet of things) , with transport systems intelligent - safer and less polluting - and with public services more efficient. The companies can increase their productivity by adopting automated decision systems that reach areas such as services, industry and agribusiness. The public sector certainly It could be improved with the adoption of AI : for example, reducing costs d the services public such as transport, education and energy . In addition, the judiciary will be able to ensure that cases come to an end more quickly.

On the other hand, the use of IA brings a number of risks to fundamental rights such as the protection of personal data and d privacy , the protection against any kind of discrimination .[1] Recent research showing the deficiencies of facial recognition systems in correctly identifying black women is an example of such problems. [2] While any decision making can be misguided and prejudiced, the problem with the decision made by AI is that it has a much larger scale.

Due to the existence of these risks, which are not exclusive to AI, as they accompany any type of human activity, it is that society is concerned with establishing norms . In fact, the concern with legislating on AI is evidence of its central role for today's societies.

AI regulation is extremely interdisciplinary, as it involves working together with professionals with knowledge from different areas that complement each other. The legal professional should also be called upon to contribute. The law is, to a large extent, an instrument for dealing with risk, avoiding it or reducing its consequences. No company would like to have its brand related to news in the press claiming that the algorithms it uses are discriminatory. This produces a huge reputational damage, causing, consequently, the loss of customer s . Hence the importance of implementing compliance focused on AI. In the last decade we have seen the concern with data protection : the decade of Big Data. This decade that we have just entered promises to be the decade of the Big Something , that is, the decade in which algorithmic decision making will be a central issue in terms of compliance . [3] Hence the need for a regulatory effort in which the whole of society must be a part.

The European Union has announced a coordinated approach to the human and ethical implications of artificial intelligence. This approach proposes that investment in AI be oriented towards, at the same time, promoting its adoption and guaranteeing a regulation capable of dealing with the risks that will arise as this technology becomes adopted on a large scale. After all, reliable AI is a necessary condition for people to start using it. For this reason, an adequate regulation of AI is the best way to obtain confidence from citizens and companies and, in this way, make its use really widespread. The European Union's proposal is to create an “ecosystem of trust”, based on a human-centered approach. [4]

From a European point of view, the development of AI must be based on the fundamental values ​​and rights from which that political community was built , such as human dignity and the protection of privacy. This also seems to be the Brazilian path, given the similarity between our legal model and the European one. In Brazil, there are already some initiatives in the National Congress, both in the Chamber and in the Senate, for the elaboration of norms that deal with the theme of AI. [5]

The European Union has an Expert Group of High Level proposed seven essential requirements for reliable IA: i niciativa and control and by human ; r obesity and safety ; p rivacidade and governs NCA data ; t transparency ; d iversity ; non-discrimination and equity ; b in being soci al and environmental and r esponsabilização. In addition, the Group presented a checklist so that companies can use, in order to app air each of the requirements in practice. These checklists will become more and more common in the future and will be part of an AI audit process that will take place in organizations that use this technology .

One point key n the s recommendation s of the European Commission in relation to legislation s aimed m regular AI is that they should m be effective es to achieve their goals, but can regulate so excessive , creating burden s disproportionate is . To achieve this balance, the Commission recommends that a risk-based approach should be followed. [6]

Thus, AI applications could be classified as high risk and low risk . In principle, only high-risk applications would be required to comply with certain mandatory legal requirements proposed by the Expert Group : adequacy of the taset ; record and data conservation; provision of information; robustness and accuracy; human supervision; specific requirements for certain AI applications, such as those used for the purpose of distance biometric identification .

But what would a high-risk AI application be ? It should be considered if well - fill the two requirements cumulative frei ously :

1) the application is used in a sector where the risks are greater and the probability of them becoming a reality is also greater . This occurs in areas such as health, transport and energy. It is recommended that high-risk sectors be listed in a specific and exhaustive manner so that there is no legal uncertainty in this regard . In addition, this list should be periodically reviewed and changed, whenever necessary, as new developments in AI are always taking place ;

2) it is also necessary to demonstrate that the application of AI , in addition to being in a high-risk sector , is used in a way that makes the appearance of significant risks likely . Thus, not all uses of AI, even within a high-risk sector , imply the existence of significant risks. In the health sector, for example , a fault in the marking system of consultation hours in a hospital usually does not generate significant risks sufficient to justifi car legislative action. The risk, therefore, must be assessed according to the impact on the affected parties. The complications of AI that affect individual rights are natural or legal ; that is ejam linked to risk s injury or death must be treated differently than applications that do not are able to produce consequences of this type.

The European proposal for an ethical and reliable AI is, therefore, an important model for Brazil. This is a necessary path if AI is to achieve its enormous potential to generate well-being for all of society, leading to economic and social development in our country.

References:

BUOLAMWINI, Joy. Response: Racial and Gender bias in Amazon Rekognition - Commercial AI System for Analyzing Faces. Available at: https://medium.com/@Joy.Buolamwini/response-racial-and-gender-bias-in-amazon-rekognition-commercial-ai-system-for-analyzing-faces-a289222eeced .

EUROPEAN COMMISSION. White Paper on Artificial Intelligence: a European approach to excellence and trust . Available at: https://ec.europa.eu/info/publications/white-paper-artificial-intelligence-european-approach-excellence-and-trust_en .

EUROPEAN COMMISSION. Ethics guidelines for trustworthy AI. Available at: https://ec.europa.eu/digital-single-market/en/news/ethics-guidelines-trustworthy-ai .

KOSHIYAMA , Adriano et. al, 2021. Towards Algorithm Auditing : A Survey on Managing Legal, Ethical and Technological Risks of AI, ML and Associated Algorithms. SSRN ( Jan, 2021), 1-31. DOI: https://dx.doi.org/10.2139/ssrn.3778998 .

[1] EUROPEAN COMMISSION . White Paper on Artificial Intelligence: a European approach to excellence and trust . Available at: https://ec.europa.eu/info/publications/white-paper-artificial-intelligence-european-approach-excellence-and-trust_en .

[2] BUOLAMWINI, Joy. Response: Racial and Gender bias in Amazon Rekognition - Commercial AI System for Analyzing Faces. Available at: https://medium.com/@Joy.Buolamwini/response-racial-and-gender-bias-in-amazon-rekognition-commercial-ai-system-for-analyzing-faces-a289222eeced .

[3] KOSHIYAMA , Adriano et. al, 2021. Towards Algorithm Auditing : A Survey on Managing Legal, Ethical and Technological Risks of AI, ML and Associated Algorithms. SSRN ( Jan, 2021), 1-31. DOI: https://dx.doi.org/10.2139/ssrn.3778998 .

[4] EUROPEAN COMMISSION. Ethics guidelines for trustworthy AI . Available at: https://ec.europa.eu/digital-single-market/en/news/ethics-guidelines-trustworthy-ai .

[5] In the Chamber of Deputies, the PL 21/2020 ( https://www.camara.leg.br/propostas-legislativas/2236340 ) and the PL 240/2020 ( https://www.camara.leg.br /proposicoesWeb/prop_mostrarintegra;jsessionid=node0kbt38p317wg4qp3xc7ada5974668009.node0?codteor=1859803&filename=Avulso+-PL+240/2020 ) . In the Federal Senate, PL 5051/2019 ( https://www25.senado.leg.br/web/ividade/materias/-/materia/138790 ); PL 5691/2019 ( https://www25.senado.leg.br/web/ividade/materias/-/materia/139586 ) and PL 872/2021 ( https://www25.senado.leg.br/web / activity / materials / - / material / 147434 ).

[6] EUROPEAN COMMISSION. White Paper on Artificial Intelligence: a European approach to excellence and trust . Available at: https://ec.europa.eu/info/publications/white-paper-artificial-intelligence-european-approach-excellence-and-trust_en .