top of page
Buscar
Anna Goldberg
6 de fev.8 min de leitura

IA (Inteligência artificial) em velocidade máxima.

Atualizado: 19 de fev.



Resolvi manter um periódico blog sobre as novidades em inteligência artificial que, acredito, possam impactar no modo de fazer ciência de qualidade, mantendo as boas práticas e os cuidados éticos que a boa pesquisa exige. Então lá vamos nós para o nosso novo - notícias em poucos minutos que chamei de notícias do:  I ABC (inteligência Artificial, Inovações Bem-vindas, integridade Científica)



  • I ABC e novos produtos. Mais empresas lançam suas plataformas e produtos baseados em IA e LLM (Large Language Models) que funcionam de forma independente do Chat-GPT: Paperpal, scite e Scholarcy, são algumas que tem já centenas de milhares de usuários. Paperpal é um assistente de redação, scite é um buscador de referências e Scholarcy faz resumos (fichamento) de artigos. Todos atuam de forma maciça na busca, referenciamento e redação de textos relevantes para a sua pesquisa. Mas o vencedor nessa área parece ser, sem dúvida, o Google. O seu Bard, e programas derivados como Duet AI e Gemini Pro, faz tudo isso e mais, pois também gera imagens, faz sumários, cria apresentações e até participa de reuniões no seu lugar, anotando tudo. (https://workspace.google.com/solutions/ai/). Aliás, o Zoom também já faz isso, já imaginaram uma cena com bots ao invés de pessoas reais tentando fazer uma reunião? E, pegando o gancho da integridade de imagens, li um texto bastante completo sobre o assunto publicado pelo Science Editor (AI and the Future of Image Integrity in Scientific Publishing - Science Editor (csescienceeditor.org)). Esse texto expõe a visão de Dror Kolodkin-Gal, fundador e CEO da Proofig, empresa que que já escaneou 85 mil manuscritos com mais de 5 milhões de imagens e informa ter uma precisão acima de 99% na detecção de imagens geradas ou modificadas por IA. Vale a pena visitar a página da empresa, que informa que o preço para análise de imagens para um indivíduo custa menos de 1 dólar para cada uma (https://www.proofig.com/). Enfim, essa onda de programas usando IA de última geração não vai parar por aqui, mas todos dizem se preocupar com a integridade dos dados, a veracidade da informação e o cuidado para não contaminar seus sistemas com informação falsa. P.S. Escrevi este texto no sábado passado e hoje (9 de fevereiro), menos de uma semana depois, vem a informação que o Bard já era. Agora é o Gemini Ultra, que veio para desbancar o ChatGPT-4!



  • I ABC e legislação. Por fim, não custa repetir: a legislação vem aí, a galope. Já saiu, na Nature de janeiro de 2024 (https://www.nature.com/articles/d41586-024-00029-4), uma crítica ao AI Act, lançado pelo parlamento europeu no final de 2023, que certamente será atualizado e se tornará mais restritivo. A questão do patenteamento já está na pauta do dia. E, à moda americana, o escritório de patentes dos E.U.A. conduziu um inquérito em nível nacional sobre a IA que teve participação maciça e um enorme volume de respostas, que foi graciosamente resumido por Roy Kaufman, Diretor Executivo da CCC  e pode ser visto na página do Scholarly Kitchen (https://scholarlykitchen.sspnet.org/2023/11/28/the-united-states-copyright-office-notice-of-inquiring-on-ai-a-quick-take/).  Uma discussão importante trata dos direitos de publicação (copyright) e o uso da informação na internet para treino e aprimoramento dos aplicativos e programas que usam IA generativa e IA em geral. Naturalmente, a presença crescente das paper mills, das publicações falsas e o uso dessas informações falsas no treinamento, aprimoramento e uso destes programas baseados em IA é uma preocupação sempre presente. Em tempo, caso não tenham visto ainda, a Academia Brasileira de Ciências lançou os seu - Recomendações para o avanço da inteligência artificial no Brasil - elaborado pelo Grupo de Trabalho de IA (https://www.abc.org.br/wp-content/uploads/2023/11/recomendacoes-para-o-avanco-da-inteligencia-artificial-no-brasil-abc-novembro-2023-GT-IA.pdf) . O documento aborda vários temas que incluem uma avaliação do que o mercado de trabalho vai exigir em termo de profissionais, as áreas de maior impacto na economia e na sociedade, as oportunidades da inovação nas áreas de Saúde, de Educação e de meio ambiente e energia. Também há colocações sobre riscos que o uso de IA apresenta e recomendações para o uso responsável das novas tecnologias.


Clique no QR code para ler a íntegra das Recomendações


AI (Artificial Intelligence) at full speed.


I decided to maintain a periodic blog about the latest developments in artificial intelligence that, I believe, could impact the way we conduct quality science, adhering to the best practices and ethical considerations that good research requires. So here we go with our new - news in a few minutes that I called: I ABC (Artificial Intelligence, Brand new Innovations, sCientific Integrity)

• I ABC and quality management. Morressier, a transnational company that develops products focused on research quality and integrity, has been producing innovative solutions practically every month. For example,  the December 2023 list (https://www.morressier.com/solutions/release-notes?utm_campaign=Product%20release%20updates&utm_medium=email&_hsmi=74576533&_hsenc=p2ANqtz-_vxGfAFJ1psajbefK8jCHombtGIMVKomAdWUaasAjQT-7leDoS-9j4u9QAitfu7KrEmB8pJwteC7iCIiyAU8fNiO0sFA&utm_content=74528409&utm_source=hs_email), informs on solutions to detect paper mills, generate texts using artificial intelligence, instruments, or services for auditing and checking your articles before publication. Goldberg Consultancy also curates texts, theses, and articles! In fact, there is already a significant discussion about the effectiveness of current software in identifying sections generated by AI, and an important trend among researchers and entrepreneurs in this field advocates that human revision is still essential (https://scholarlykitchen.sspnet.org/2023/08/03/guest-post-are-hit-backed-ai-research-integrity-solutions-the-need-of-the-hour/?informz=1&nbd=0729ba0c-6bf0-4726-a357-d10c5c3d0103&nbd_source=informz), that is, it is necessary to maintain the use of human skills, the so-called HITs (Human Intelligence tasks).

• I ABC and new products. More companies are launching their platforms and products based on AI and LLM (Large Language Models) that operate independently of Chat-GPT: Paperpal, scite, and Scholarcy are some that already boast hundreds of thousands of users. Paperpal is a writing assistant, scite is a reference search engine, and Scholarcy summarizes full articles. All play a massive role in searching, referencing, and writing relevant texts for your research. But the winner in this area seems to be, without a doubt, Google. Its Bard, and derived programs like Duet AI and Gemini Pro, do all this and more, as they also generate images, create summaries, make presentations, and even participate in meetings on your behalf, annnotating everything. (https://workspace.google.com/solutions/ai/). By the way, Zoom already does this too. Can you imagine a scene with bots instead of real people trying to hold a meeting? And, picking up on image integrity, I read a very comprehensive text on the subject published by the Science Editor (AI and the Future of Image Integrity in Scientific Publishing - Science Editor (csescienceeditor.org ). This text exposes the vision of Dror Kolodkin-Gal, founder and CEO of Proofig, a company that has already scanned 85 thousand manuscripts with more than 5 million images and claims to have an accuracy above 99% in AI-generated or AI-modified image detection. It's worth visiting the company's page, which informs that the price for image analysis for an individual costs less than 1 dollar each (https://www.proofig.com/). Anyway, this wave of programs using state-of-the-art AI is not going to stop here, but everyone claims to be concerned with data integrity, the accuracy of information, and the caution not to contaminate their systems with false information.

  • • I ABC and radical innovation. In the August 2023 newsletter, Zain Kahn's Superhuman informs that ChatGPT-4 presented new competencies in its program (https://www.superhuman.ai/p/chatgpt-launches-6-major-upgrades?utm_source=www.joinsuperhuman.ai&utm_medium=newsletter&utm_campaign=chatgpt-launches-6-major-upgrades): immediate exemplification, suggestion of responses in a conversation, indefinite maintenance of your login in the program, upload of multiple files at the same time, and so on. A month later, it informs that ChatGPT can now see, hear, and speak (https://openai.com/blog/chatgpt-can-now-see-hear-and-speak?utm_source=www.joinsuperhuman.ai&utm_medium=newsletter&utm_campaign=chatgpt-can-now-see-hear-and-speak). The following month, it can access the internet with it, and use it in excel spreadsheets and pdf documents (news from October 2023). And the list doesn't stop there. Help! And the most impactful magazine in Medicine now has an edition exclusively focused on issues involving AI. It is the NEJM AI, which brings, in its January 17th edition, a study by Sean Wu, Ira Kurz, and colleagues evaluating the quality of responses from various generative AI programs when confronted with clinical questions, in this case on Nephrology. ChatGPT-4 was the one that presented the highest accuracy, but they all need to improve their performance (NEJM AI 2024;1(2), DOI: 10.1056/AIdbp2300092). Does anyone have any doubts about improvements happening in the next few months? P.S. I wrote this text  last Saturday and today (February 9), less than a week later, we receive news that Bard is down, we now have the Gemini Ultra, here to floor ChatGPT-4!

• I ABC and legislation. Finally, it doesn't hurt to insist: legislation is coming, at a gallop. It has already been published in the January 2024 issue of Nature (https://www.nature.com/articles/d41586-024-00029-4), a critique of the AI Act, launched by the European Parliament at the end of 2023, which will certainly be updated and become more restrictive. The issue of patenting is already on the agenda. And, following the American way, the U.S. Patent Office has conducted a national-level inquiry into AI that had massive participation and a huge volume of responses, which have been graciously summarized by Roy Kaufman, Executive Director of CCC, and can be read on the Scholarly Kitchen page (https://scholarlykitchen.sspnet.org/2023/11/28/the-united-states-copyright-office-notice-of-inquiring-on-ai-a-quick-take/). An important discussion deals with publishing rights (copyright) and the use of information on the internet for training and improving applications and programs that use generative AI and AI in general. Naturally, the growing presence of paper mills, fake publications, and the use of this false information in the training, improvement, and use of these AI-based programs is an ever-present concern. By the way, if you haven't seen it yet, the Brazilian Academy of Sciences has released its - Recommendations for the advancement of artificial intelligence in Brazil - prepared by its AI Working Group (https://www.abc.org.br/wp-content/uploads/2023/11/recomendacoes-para-o-avanco-da-inteligencia-artificial-no-brasil-abc-novembro-2023-GT-IA.pdf). The document addresses several topics, including an assessment of what the job market will require in terms of professionals, the areas of greatest impact on the economy and society, opportunities for innovation in the areas of health, education, environment, and energy. It also debates on the risks that the use of AI presents and writes out recommendations for the responsible use of these new technologies."



Click on the  QR code to read the complete Recommendations.



17 visualizações0 comentário

Posts recentes

Ver tudo

C

0 comentário

Comentários

bottom of page