Questões de Concurso

        Modern societies rely upon law as the primary mechanism to control their development and manage their conflicts. Through carefully designed rights and responsibilities, institutions and procedures, law can enable humans to engage in increasingly complex social and economic activities. Therefore, law plays an important role in understanding how societies change. To explore the interplay between law and society, we need to study how both co-evolve over time. This requires a firm quantitative grasp of the changes occurring in both domains. But while quantifying societal change has been the subject of tremendous research efforts in fields such as sociology, economics, or social physics for many years, much less work has been done to quantify legal change. In fact, legal scholars have traditionally regarded the law as hardly quantifiable, and although there is no dearth of empirical legal studies, it is only recently that researchers have begun to apply data science methods to law. To date, there have been relatively few quantitative works that explicitly address legal change, and almost no scholarship exists that analyses the time-evolving outputs of the legislative and executive branches of national governments at scale. Unlocking these data sources for the interdisciplinary scientific community will be crucial for understanding how law and society interact.

            Our work takes a step towards this goal. As a starting point, we hypothesise that an increasingly diverse and interconnected society might create increasingly diverse and interconnected rules. Lawmakers create, modify, and delete legal rules to achieve particular behavioural outcomes, often in an effort to respond to perceived changes in societal needs. While earlier large-scale quantitative work focused on analysing an individual snapshot of laws enacted by national parliaments, collections of snapshots offer a window into the dynamic interaction between law and society. Such collections represent complete, time-evolving populations of statutes at the national level. Hence, no sampling is needed for their analysis, and all changes we observe are direct consequences of legislative activity. This feature makes collections of nation-level statutes particularly suitable for investigating temporal dynamics.

            To preserve the intended multidimensionality of legal document collections and explore how they change over time, legislative corpora should be modelled as dynamic document networks. In particular, since legal documents are carefully organised and interlinked, their structure provides a more direct window into their content and dynamics than their language: Networks honour the deliberate design decisions made by the document authors and circumvent some of the ambiguity problems that natural language-based approaches inherently face. In this paper, we therefore develop an informed data model for legislative corpora, capturing the richness of legislative data for exploration by social physics.

Adapted from Katz, D.M., Coupette, C., Beckedorf, J. et al. Complex societies and the growth of the law. Sci Rep 10, 18737 (2020). Available at https://www.nature.com/articles/s41598-020-73623-x

It’s been four decades since society was in awe of the quality of recordings available from a cassette recorder tape. Today we have something new to be in awe of: deepfakes. Deepfakes include hyperrealistic videos that use artificial intelligence (AI) to create fake digital content that looks and sounds real. The word is a portmanteau of “deep learning” and “fake.” Deepfakes are everywhere: from TV news to advertising, from national election campaigns to wars between states, and from cybercriminals’ phishing campaigns to insurance claims that fraudsters file. And deepfakes come in all shapes and sizes — videos, pictures, audio, text, and any other digital material that can be manipulated with AI. One estimate suggests that deepfake content online is growing at the rate of 400% annually.

There appear to be legitimate uses of deepfakes, such as in the medical industry to improve the diagnostic accuracy of AI algorithms in identifying periodontal disease or to help medical professionals create artificial patients (from real patient data) to safely test new diagnoses and treatments or help physicians make medical decisions. Deepfakes are also used to entertain, as seen recently on America’s Got Talent, and there may be future uses where deepfake could help teachers address the personal needs and preferences of specific students.

Unfortunately, there is also the obvious downside, where the most visible examples represent malicious and illegitimate uses. Examples already exist.

Deepfakes also involve voice phishing, also known as vishing, which has been among the most common techniques for cybercriminals. This technique involves using cloned voices over the phone to exploit the victim’s professional or personal relationships by impersonating trusted individuals. In March 2019, cybercriminals were able to use a deepfake to fool the CEO of a U.K.-based energy firm into making a US$234,000 wire transfer. The British CEO who was victimized thought that the person speaking on the phone was the chief executive of the firm’s German parent company. The deepfake caller asked him to transfer the funds to a Hungarian supplier within an hour, emphasizing that the matter was extremely urgent. The fraudsters used AI-based software to successfully imitate the German executive’s voice. […]

What can be done to combat deepfakes? Could we create deepfake detectors? Or create laws or a code of conduct that probably would be ignored?

There are tools that can analyze the blood flow in a subject’s face and then compare it to human blood flow activity to detect a fake. Also, the European Union is working on addressing manipulative behaviors.

There are downsides to both categories of solutions, but clearly something needs to be done to build trust in this emerging and disruptive technology. The problem isn’t going away. It is only increasing.

Nit Kshetri, Bryan School of Business and Economics, University of North Carolina at Greensboro, Greensboro, NC, USA

Joanna F. DeFranco, Software Engineering, The Pennsylvania State University, Malvern, PA, USA Jeffrey Voas, NIST, USA

Adapted from: https://www.computer.org/csdl/magazine/co/2023/07/10154234/ 1O1wTOn6ynC