Thursday, December 19, 2024

Can AI Scientists Replace Human Researchers? / Modelos IA de cientistas podem substituir pesquisadores humanos?


Can AI Scientists Replace Human Researchers?

I can say that I've started my professional career as a scientist/researcher… well, okay, maybe not quite. As an intern for the Sao Paulo University, Naval Architecture and Ocean Engineering Department's esteemed Prof. Kazuo Nishimoto, a leading researcher in Naval/Oceanic Offshore Engineering, I got my first taste of this amazing field. Professor Nishimoto's guidance led me to pursue a Master's Degree at the Faculty of Engineering at Yokohama National University (YNU) in Japan, where I explored applying "expert knowledge" through fuzzy control to steer scaled-down free running models during experiments in our lab's ocean Directional Spectrum Waves towing tank of YNU.


In short, I've done my fair share of research and even contributed to research papers. That's why this blog post from Sakana.Ai, a Japanese start-up research lab, caught my eye with a fascinating proposition: the AI Scientist.


This article examines AI systems that can potentially automate the entire scientific research process. Imagine a system that can brainstorm research ideas, code experiments, analyze data, and even write scientific papers! The AI Scientist supposedly even conducts peer review and iterates on its research, constantly refining its approach.


Now, as someone who spent countless hours in libraries and labs, the idea of an AI handling the entire research cycle is both intriguing and a little unsettling. On one hand, it could significantly accelerate scientific progress, freeing researchers to focus on more complex problems and interpretations.


But on the other hand, wouldn't this diminish the value of human expertise and creativity in research? My short and limited research experience working alongside and observing the work of esteemed mentors like Prof. NishimotoProf. Takezawa Seiji (竹沢 誠二)Prof. Hirayama Tsugukiyo ( 平山 次清) and other colleagues who are now leading researchers in their fields (including Carlos Hakio FucatuClaudio Mueller Prado SampaioCheng Liang-YeeKim Se-Eun/金 世殷)  instilled in me the importance of a solid scientific foundation but also intuition, critical thinking, and the thrill of discovery - those "aha!" moments that come from experimentation. Can an AI replicate these aspects of scientific exploration?


Sakana.Ai's research suggests the AI Scientist has already made novel contributions in machine learning.  This is certainly promising for the future of AI-assisted research. However, I believe the human element will remain crucial. Perhaps the future lies in a collaborative model, where AI handles the heavy lifting of data analysis and experiment execution, while human researchers provide the vision, interpretation, and that spark of ingenuity that leads to true breakthroughs.


What are your thoughts? 


Modelos IA de cientistas podem substituir pesquisadores humanos?


Poderia dizer que comecei minha carreira profissional como cientista/pesquisador... bem, ok, talvez não exatamente. Como estudante de engenharia e estagiário na Escola Politécnica da USP, especificamente no Departamento de Arquitetura Naval e Engenharia Oceânica da Universidade de São Paulo, sob a orientação do estimado Prof. Kazuo Nishimoto, um líder em Engenharia Naval e Oceânica, onde tive meu primeiro contato com este campo fascinante. A orientação do Professor Nishimoto me levou a buscar um Mestrado na Faculdade de Engenharia da Universidade Nacional de Yokohama (YNU) no Japão, onde explorei a aplicação de "conhecimentos especializados ou experiencia" através do uso de controle fuzzy para guiar modelos de navios em escala durante experimentos no tanque de reboque de Ondas de Espectro Direcional do nosso laboratório na YNU.


Em resumo, contribuí para pesquisas e até publiquei artigos científicos. É por isso que este post no blog da Sakana.Ai, uma start-up de pesquisa japonesa, chamou minha atenção com uma proposta fascinante: Modelos IA de Cientistas.

Este artigo examina sistemas de IA que podem potencialmente automatizar todo o processo de pesquisa científica. Imagine um sistema que pode gerar ideias de pesquisa, codificar experimentos, analisar dados e até escrever artigos científicos! Supostamente, o Cientista de IA até realiza revisão por pares e itera em sua pesquisa, refinando constantemente sua abordagem.


Para alguém que passou incontáveis horas em bibliotecas e laboratórios, a ideia de uma IA lidando com todo o ciclo de pesquisa é ao mesmo tempo intrigante e um pouco inquietante. Por um lado, isso poderia acelerar significativamente o progresso científico, liberando os pesquisadores para se concentrarem em problemas e interpretações mais complexos. Mas, por outro lado, isso não diminuiria o valor da expertise e criatividade humanas na pesquisa?

Minha experiência de pesquisa, embora limitada, trabalhando ao lado e observando o trabalho de mentores estimados como Prof. NishimotoProf. Takezawa Seiji (竹沢 誠二)Prof. Hirayama Tsugukiyo (平山 次清) e outros colegas que agora são líderes em seus campos (incluindo Carlos Hakio FucatuClaudio Mueller Prado SampaioCheng Liang-YeeKim Se-Eun/金 世殷) me ensinou a importância de uma base científica sólida, mas também de intuição, pensamento crítico e a emoção com a descoberta - aqueles momentos "aha!" que vêm da experimentação. Pode uma IA replicar esses aspectos da exploração científica?


A pesquisa da Sakana.Ai sugere que o Cientista de IA já fez contribuições inovadoras em aprendizado de máquina (machine learning). Isso é certamente promissor para o futuro da pesquisa assistida por IA. No entanto, acredito que o elemento humano continuará sendo crucial. Talvez o futuro resida em um modelo colaborativo, onde a IA lida com a parte pesada da análise de dados e até execução de experimentos, enquanto os pesquisadores humanos fornecem a visão, interpretação e aquela centelha de engenhosidade que leva a verdadeiros avanços.


Qual sua opinião?


=========================

I almost couldn't find any photos of the my Professors from YNU. So, I decided to post here (without any prior authorization) the only photos I could find for Professor Takezawa and Professor Hirayama. In order below.




Some references mentioned in this blog: 

  • Free Running Model Experiment on Speed Loss in Directional Spectrum Waves  (Techno marine Bulletin of the Society of Naval Architects of Japan 760 (0), 747-, 1992; The Japan Society of Naval Architects and Ocean Engineers) 
    • By Takezawa Seiji; Hirayama Tsugukiyo; Kim Se-Eun; Suzano Alberto
    • Summary: For evaluating speed loss, self-propulsion test and free running test are carried out in directional spectrum waves with a full ship model in the towing tank of Yokohama National University. Such experiments will be the first time in the world. In the free running test, the model has six degrees of freedom as in real condition, and it is controlled to run straightly by using rudder controlling system, therefore the speed loss can be derived directly by considering the relative speed against carriage by using optical position sensor. It is found that the speed loss in short-crested waves is bigger than the speed loss in long-crested waves in short wave length. The experimental results of speed loss in directional spectrum waves are compared with the results of estimation methods which are originated from the added resistance obtained by experiment and theoretical calculation.
  • YNY Systems Design for Ocean-Space Department
  • Hirayama-sensei's Research Map site page
  • 2022-12-08-Hirayama-sensein in a news article
  • Sao Paulo University - Numerical Offshore Tank TPN-USP


WIP URLs

Search results

Search results

平山 次清 (Tsugukiyo Hirayama) - MISC - researchmap

Sakana AI





[Timestamp: 2024/12/18 14:07:29]

No comments:

Post a Comment

AI Agents, a case of siblings and self reflection.

The content below was copied from this MOLTBOOK posting It is clear that AI agents are like Twins, same DNA, may look similar but are in ...