Understanding language is the key for accessing the vast knowledge humans have gathered for ages.
But we understand text because we also see, feel and sense the world around us.
A language possesses power not only to convey the wisdom we and our ancestors accumulated, it can be a natural and powerful mean to share and confront old knowledge with newly discovered one, also by machines.
The twenty-first century is about combining and applying methods across various fields, modalities and tasks.
Natural Language Processing has a yet underrated capability to make sense of the constantly growing enormous data which is already impossible to process by a human mind.
Machine Learning gives us ways to find patterns, but we are still the ones who have to interpret their meaning.
This is why I'm interested in more general AI (AGI) which is supposed to reason about our world on our terms, learn how to learn and explain what it learned.
Although Large Language Models showed us the power of big data and hardware, they are far from sufficient and reliable.
I believe that we need a paradigm shift in how the knowledge is accumulated and learned - as a nativist interested in human-aligned machines, I seek for novel build-in cognitive mechanisms that have to be decided before learning starts.
I'm involved in many projects and topics already, but I'm open to starting new ones, especially across various disciplines. Some of the examples of the current topics are briefly described below.
Common Sense
Common Sense Knowledge is required for an autonomous AI to reason properly about our world. Called the Holy Grail of AI, common sense is recently told to be "solved" by LLMs. But if you go into more "tacit" knowledge with us, you'll see that acquiring it beyond what can be inferred from text is still challenging.
Artificial Empathy
Understanding what a human being feels is probably the most important skill for an artificial agent. We extend emotion recognition by employing additional contextual factors as time or probability of a positive or negative state continuity. Large scale datasets are used to calculate how strong or certain are the emotional changes in humans.
Machine Ethics
From 2005 we have been developing an automatic moral decision making agent by combining common sense and affect recognition. Our original commonsense ethics approach does not rely on any particular school of thought, it observes the polarity change by processing vast numbers of cases.
AI for Well-being
Machine Intelligence has a potential not only to help us in short-term tasks, but also to influence our thinking in a long run. Scrupulously analyzed scientific data can correct errors of common sense, show us negative outcomes of cognitive biases, translate and explain what people on the other side of the world or outside your filter bubble think.
Dialog Systems
Our artificial assistants can be much more than question-answering one-utterance partners. We aim at conversational systems which can discuss with us, teach us what we are wrong about, make us motivated, help us stay positive and focused, keep us informed, entertained and emotionally touched.
Figurative Language
Artificial creativity is needed not only for coming up with new ideas helping people to discover new patterns. It is also about finding novel ways of expressing and explaining. We work on metaphor understanding and generation, machine-created poetry and irony to allow machines to improve and enrich their communication abilities.
Humor Processing
Humor is told to be a display of human intelligence but computers are not too good with recognizing and generating it. In our research we try to reach beyond the popular puns and word-plays by implementing more sophisticated language understanding algorithms.
Cognitive Architectures
Our novel approach to AGI-level holistic cognitive architectures is to integrate lifelong knowledge acquisition in KABURA by crawler-like multi-agent entities called Bacteria Lingualis. Confronting their findings from different data sources in different languages is the key for wider context understanding.
My Group Members
Current Members
Ryoma Shinto
Atsuhiro Ihara
Jin Hosono
Don Divin Anemeta
Waiyan Hein
Akihiko Obayashi
Takuto Sato
Asato Yamada
Asa Yokoyama
Previous Members
Ryūta Aoyagi
Masashi Takeshita
Kei Okada
Ryo Hashimoto
Arisa Abe
Huizhong Ji
Joanna Szwoch
Xiaodong Liu
Yiyang Zhang
Dusan Radisavljevic
Yuki Katsumata
Kaiya Sasaki
Hiromichi Kameya
Kazuki Yano
Daiki Masuda
Mizuki Yoshii
Masaki Sakata
Aitaro Yamamoto
Bartosz Ziolko
Da Li
Daiki Shirafuji
Yuki Urabe
Mateusz Babieno
Saori Hayashi
Patrycja Swieczkowska
Sho Takishita
Yoshifumi Fukuda
Yue Wang
Maria Skeppstedt
Seiya Shudo
Shuto Fukuda
Radoslaw Komuda
Keita Mitsuhashi
Yuki Konno
Daniel Harborne
Michal Mazur
Marek Krawczyk
Ikumi Inoue
Soichi Tanaka
Shohei Kitahara
Taku Nakamura
Anna Gschwendtner
Joelle Vitzikam
Tadashi Tanaka
Kohei Hara
Satsuna Denzumi
Kohei Matsumoto
Urszula Jagla
Austin Clapp
Takuya Kamiyama
Denis Kiselev
Shiho Kitajima
Michal Sylwester
Yuusuke Amaya
Mitsuru Takizawa
Krzysztof Karolczak
Svetoslav Dankov
Hajime Wakahara
Magnus Ahltorp
Dai Hasegawa
Keisuke Takagi
Ryoichi Chikamura
Takuya Emori
Koichi Muramoto
Pawel Bylica
Tyson Roberts
Motoyasu Fujita
Pawel Dybala
Michal Ptaszynski
Hugues Baudrillart
Mahfoud Bouzidi
Amrane Chabane
Kenta Imai
Wenhan Shi
Jacek Maciejewski
Shinsuke Higuchi
Yudai Kitano
Ge Yali
Research topics evolution since I joined the labジェプカの研究の三つの次元（及び次元間の相互関係）
