Table of Contents
- What is the AI Action Plan Committee?
- Overview of the 2nd AI Action Plan Formulation Committee
- Extension of healthy period through health care
- Extension of human function
- virtual space
- Changes in social quality and values (technology and innovation required before and after COVID-19)
What is the AI Action Plan Committee?
Currently, artificial intelligence (AI) technology development is actively invested in research and development worldwide, mainly in the United States and China, and each country continues to evolve to obtain cutting-edge technological capabilities.
Japan also needs to build a system for AI technology development that can respond to these trends of the times.
It is also undeniable that Japan is lagging behind the United States and China in the utilization of AI technology through big data.
Therefore, from January 2021, NEDO (National Research and Development Agency New Energy and Industrial Technology Development Organization) will conduct a “survey for formulating a comprehensive research and development action plan and extracting business in the field of artificial intelligence (AI) technology ” . It was started.
The AI Action Plan Formulation Committee , which promotes this survey, has been held since February.
The AI Action Plan Formulation Committee will form a committee of experts and consider a clear action plan based on overseas cases and domestic and overseas institutional policies.
The committee will be held six times between February 2021 and June 2021 in the following flow.
|1st||Free discussion while reviewing the committee member’s greetings and the 2016 version of the “Vision for Social Implementation of Next-Generation Artificial Intelligence Technology”|
|2nd ~ 3rd||Decide what specific technical and social issues should be discussed|
|4th ~ 5th||Consider a specific action plan for the points to be discussed (direction of development, social issues, etc.) summarized in 2-3 sessions|
|6th||Approve the action plan formulated in 4-5 times|
In addition, the committee consists of the following members.
ASCII.jp and AINOW will serve as media partners of the AI Action Plan Formulation Committee to strengthen the utilization of AI in Japan.
This article introduces the discussion at the second meeting of the AI Action Plan Formulation Committee held on February 26 (survey to formulate a broad R&D action plan and business extraction in the field of artificial intelligence technology). I will continue.
Experts who participated in this committee will hold the “Action Plan Symposium on Overall Research and Development in the Field of Artificial Intelligence (AI) Technology” on Tuesday, June 15th.
Overview of the 2nd AI Action Plan Formulation Committee
In February 2021, the AI Action Plan Development Committee formulated 20 future societies based on the “Survey on future predictions by science and technology conducted by governments and institutions” conducted by the Institute of Future Engineering, a public interest incorporated foundation in February 2021. We will list up important events and proceed with discussions based on them.
AI technical keywords that are highly relevant to each social event are listed.
At the first meeting, the basic policy of the committee as described above was discussed and approved, and at the second meeting of the AI Action Plan Formulation Committee, the keywords that were lined up were asked, “Is that keyword really good?” Are there any keywords?” was the theme of the discussion.
This time, I would like to introduce 4 of the 20 social phenomena that were the points of particular discussion.
Extension of healthy period through health care
In the healthcare field, the focus of discussion was whether a model in which AI makes complete decisions and humans accept them.
For example, in the future, when AI discovers an abnormality in a patient through image recognition, we are concerned about whether AI will be able to make decisions such as
what kind of abnormality is it?'' andhow to treat it.”
In this regard, from the committee
It would be better if a small primary care doctor (AI) constantly checks the patient and escalates to a large primary care doctor (human) if something goes wrong.
There was an opinion.
The same opinion requires countermeasures because there are problems such as the cost and time required for small attending physicians, and the pain of patients being monitored for a long time.
Also, the final difficult part is “to what extent will AI make decisions on its own? ”
For example, when an abnormality occurs in a sick person (when an alert sounds), it is an important point to discuss whether to bring it to a human or whether the system will treat it as it is.
Currently, AI is becoming more power-saving and lightweight, and the technological innovation of processing at the edge is progressing.
At that time, the evolution of information that can be obtained on the edge side, the technological capabilities of AI, and the evolution of technology that can be integrated into general life will be important.
In addition, from another committee
When escalating, I think it would be better if humans and AI (family doctor) decide in a conference-like format.
There was an opinion.
The idea is that “AI will tell the doctor about the patient’s abnormalities, and the doctor will judge the patient’s treatment method based on that.”
In this case, AI needs to verbalize and explain the patient’s condition to the doctor. In other words, technology that can talk to humans and technology that explains the basis of opinions (technological development of natural language understanding) is necessary.
We also have to solve the “frame problem (*1)” that always comes up when we convey our requests to AI.
The key to resolving the frame problem is “how much data is available in the world”.
With regard to the world model (*2), one of the committee members said, “If the world model can be modeled appropriately, AI may be able to grasp the meaning of words to some extent.” We must also include it.”
In the healthcare field, there are many problems that need to be solved, such as ” timing of escalation, ” ” accuracy of natural language processing, ” and ” solving the frame problem (research on world models) .”
|*1 Frame problem : AI (robots), which have limited information processing capabilities, cannot deal with all problems that can occur in reality.|
|*2 World model : A framework that internally builds a model of the environment through learning based on limited information from the environment. By using the world model, it becomes possible to grasp the behavior of the past/future, counterfactual, and unobservable states that cannot be observed directly, and improve the performance of action selection according to the purpose.|
Extension of human function
In terms of extending human functions, the focus of discussion was whether AI can think in the same way as humans do .
In principle, neural networks are modeled after the human brain.
Assuming that the brain is made up of modules, the hypothesis arises that “AI can think like humans if we create and connect various modules.”
Here , discussions progressed along with the two-story brain and symbolic reasoning .
Regarding the two-story brain, the committee members said,
For physicality, sensor input and data output are connected and predictions are made with a world model, but for language, arbitrary algorithms are entered and installed on a learning basis, so basically various It makes me feel like I have the ability.
In other words, I have come to realize that the confusion is caused by the fact that the story is quite different in both physicality and language.
Learning with a large-scale self-supervised (learning) neural network such as GPT seems to contain various algorithms, and I get the impression that this is approaching the part of the two-story brain of language.
There was an opinion.
As you can see from this opinion, it seems important to advance research on the two-story brain in order to realize AI thinking.
Regarding symbolic reasoning (processing),
System 1 (3) is based on intuitive judgment, so language and symbolic processing are basically not involved, but System 2 (4) is based on knowledge of statistics and probability theory. , it is like a symbolic reasoning.
If that’s the case, System 2 has to do something symbolic or speculative.
There was an opinion.
|*3 System 1 : An intuitive and fast thinking mode. They don’t feel like they’re in control of their own consciousness, they’re good at coming up with ideas and associations, and they like consistency and coherence.|
|*4 System 2 : A logical and slow thinking mode. Logical and statistical thinking cannot be done without System 2. I usually wait with almost no effort. They require attention and cannot think well if they are distracted.|
“System 1” and “System 2” are often heard in behavioral economics, and how to connect them is likely to become an important technical issue in the future.
In the discussion of virtual space, the committee members agreed that there are many needs for the realization of virtual space .
- People with disabilities who do not have five senses
- Those who have difficulty in meeting people face-to-face
- Young people who want to become the ultimate withdrawal
- Those who want to create their ideal self
For many people, the virtual space (virtual world) has a positive impact.
As an initiative that will lead to the realization of a virtual space, ” development of a chatbot that includes individual thoughts ” was raised.
The development of this chatbot is actually progressing, but if the accuracy improves in the future, it is said that “AIs that have someone’s thoughts will connect with each other in a virtual space, and if they get along, they will actually connect with each other.” may become possible.
In addition, if you can leave clear human data (behavior, etc.) virtually, the width of the virtual space will expand.
For example, it may be possible to experience real contact with people who cannot meet in real life, such as deceased people and celebrities.
Furthermore, if there is no energy in the future or if infectious diseases continue in the future, the demand for virtual spaces is likely to increase further.
In that case, we need technology that can do various things in the virtual space.
For example, we will need AI technology that supports creative activities in virtual space, and technology that supports human creative desires.
There is a demand not only to bring your current life and yourself to the virtual space, but also to create the ideal self you want in a virtual space.
There was an opinion.
For example, if we can freely create physical parts such as appearance and voice, skills such as “fast walking” and “can draw”, and living spaces, the demand for virtual spaces will increase further.
Changes in social quality and values (technology and innovation required before and after COVID-19)
Regarding the changes before and after Corona, the first thing that came out of the committee was
Technically, telework and online 〇〇 have been around for a long time. In other words, the technology did not change before and after the corona, but only the values and needs of society changed .
This is my opinion.
Social change is accelerating due to the spread of the new coronavirus.
As a technology that will become important in this regard, the ” value network ” that can quantify and visualize changes in social values was raised.
If you have a “social values neural network “, you will be able to find out if your values are compatible with society just by comparing your values to it.
Currently, the only way to know values (society’s opinions) is through indirect methods such as voting and market research. Moreover, it is difficult for this method to respond to sudden changes and real-time situations.
For example, if you create a value network where you can’t compromise on this point, it will be easier to create authentication.
Also, there was an opinion that if AI moves to the edge side, communication between AIs will be possible.
If communication between AIs becomes possible, they will be able to share their learning results (distributed learning), which will lead to the singularity.
This time, we will discuss four topics that were particularly discussed at the 2nd AI Action Plan Development Committee .
I introduced the contents of the meeting about.
In the future, the key to the discussion is likely to be not only technical issues but also how to implement it in society, including legislation.
Next time, we will dig deeper into the new keywords that came out this time and discuss whether there are other keywords.