NTT Research launches a new physics of artificial intelligence on Harvard

When a parent teaches a small child referring to the world, teaches through associations and identification of patterns. Take the letter, for example. Parents show their child enough examples of the letter and will soon be able to identify other examples in contexts in which the guidelines are not active; School, book, billboard.

There were many constantly developing artificial intelligence technology (AI) he taught in the same way. Scientists fed systems correct examples of something they wanted to recognize, and how a small child AI began to recognize patterns and extrapola such knowledge to contexts that he had never experienced before, creating his own “neural network” for categorization. However, like human intelligence, experts lost tracking of input data that reported AI's decision.

“AI's problem” problem appears as the fact that we do not quite understand how and why the AI ​​system makes connections or variables that play decisions. This problem is particularly important when it strives to improve the credibility and security of systems and establish AI reception management.

From the AI ​​powered vehicle, which does not inhibit in time and harm pedestrians to health technology devices that help doctors diagnose patients, and Erroneousness demonstrated by the processes of employing AIThe complexity of these systems has led to an increase in a new field of research: AI physics, which aims to further establish artificial intelligence as tools for people to achieve a higher understanding.

Now the new independent research group is solving these challenges by combining the fields of physics, psychology, philosophy and neuronauks in interdisciplinary exploration of AI secrets.

Newly in love Physics of the Artificial Intelligence Group It is the separation of the NTT Research's Physics & Information (Phi) laboratory and was presented at the NTT Upgrade 2025 conference in San Francisco, California last week. It will continue to develop the physics of artificial intelligence to understand AI, to which the team has been studying for the last five years.

Dr. Hidenori Tanaka, who has a doctorate in applied physics and computer science and engineering from Harvard University, will lead a new research group, based on his previous experience in the intelligent group of NTT systems and the AI ​​CBS-NTT research program in the field of intelligence physics at Harvard.

“As a physicist, I am excited about the topic of intelligence, because mathematically, how can you think about the concept of creativity? How can you even think about goodness? These concepts would remain abstract, if not for AI. It is easy to speculate, saying” this is my definition of kindness “, which is not mathematically significant, but now with AI, it is practically important, because we would like to do AI. kindness IsFor example – Dr. Tanak told me last week on the margins of the Update conference.

At the beginning of its research, Phi Lab recognized the importance of understanding the “black box” AI and machine learning in order to develop new systems with better energy efficiency to calculate. AI's progress over the past half decade has caused more and more important considerations of security and credibility, which in this way became critical for industry applications and management decisions on AI adoption.

Through the new research group NTT Research, it concerns the similarities of biological and artificial intelligence, hoping to solve the complexity of AI mechanisms and building a more harmonious merger of man-AI's cooperation.

Although the novel in the integration of artificial intelligence, this approach is not new. For centuries, physicists have tried to reveal the precise details of technological and human relationships, from Galileo Galilei research on the movement of objects and its contribution to mechanics, after the steam engine informed the understanding of thermodynamics during the industrial revolution. However, in the 21st century, scientists try to understand how AI works in the field of training, accumulating knowledge and decision making so that in the future more consistent, safe and trustworthy AI technologies can be designed.

“AI is a neuronet, the way it is ordered is very similar to the function of the human brain; neurons connected with synapses that are represented by numbers inside the computer. And then, as we think, physics can exist … Physics consists in taking anything from the universe, formulating mathematical hypotheses on their internal actions and testing them,” said Dr. Hanak.

The new group will continue to cooperate with the Harvard University Center for Brain Science (CBS) and plan to cooperate with the professor of Stanford University Suya Ganguli, with whom Dr. Tanak was a co -author of several articles.

However, Dr. Tanaka emphasizes that the natural approach and between the industry will be fundamental. In 2017, when he was a PhD student at Harvard, the researcher realized that he wanted to do more than traditional physics and follow in the footsteps of his predecessors, from Galilee to Newton and Einstein to open a new conceptual world in physics.

“Currently AI is one of the topics that I can talk to everyone. As a researcher, it is great because everyone is always supposed to talk about AI, and I also learn from every conversation, because I realize how people see and use AI differently, even outside of academic contexts. I see NTT's mission as a catalyst to cause these conversations, regardless of the origin of people, regardless of the origin of people, regardless of the origin of people, regardless of the origin of people, regardless of the origin of people, regardless of the origin of people people.