What are the problems with the interface between AI and humans?

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.

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 2nd AI Action Plan Formulation Committee meeting held on April 27 (survey to formulate an action plan for overall R&D in the field of artificial intelligence technology and extract business). I will continue. ASCII.jp and AINOW will serve as media partners of the AI ​​Action Plan Formulation Committee to strengthen the utilization of AI in Japan.

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.

4th 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.

In the 4th AI Action Plan, we examined specific action plans based on the AI ​​elemental technologies to be implemented, technical goals, and social phenomena to be implemented, which were discussed in the 2nd and 3rd sessions.

In this article, we will introduce three topics that were particularly important among the discussions in the 4th AI Action Plan.

Health/Medical

The biggest discussion this time was the topic of ” health and medical care “.

Discussions centered on the following three topics, including topics related to AI-human interfaces and ethical issues.

The first is a social phenomenon that has been ongoing since the 2nd time, ” AI makes diagnoses instead of doctors. AI is small and humans are large .”

In this regard, a member of the Committee

Instead of AI diagnosing in place of a doctor, the discussion should proceed in a way that the doctor and AI collaborate to make a diagnosis . Also, when performing collaborative diagnosis, it is necessary for AI to explain (show grounds) to humans, but I think it can be done not only in natural language but also in diagrams and graphs .

There was an opinion.

For example, when making specific decisions about how to treat a patient, ” Humans will be able to make decisions that include the opinions of AI .”

In order to achieve this, it is premised on always seeing patients as a doctor with a small AI. At this time, the attending physician must understand the patient’s background knowledge (previous diseases, patient’s constitution, etc.).

Therefore, it is necessary to give the patient’s past history as data to the small AI of the attending physician.

If that were to happen, we would also have to consider how to acquire the patient’s background data.

The second topic is, ” Isn’t it better to include stories about maintaining a healthy state (healthtech) instead of stories about people getting sick (meditech-like)? “

From the committee members

In order to maintain good health, it would be better to discuss diet, exercise, and food tech/sports tech .

In order to deliver safe, secure and delicious food to people who want to eat it as an integrated service, some start-up companies that use information technology are emerging.

It goes without saying about sports, and it is necessary to discuss how to maintain such a state of health.

There was an opinion.

For example, if we let AI learn all the results of physical examinations, it may be possible to find out, “ There is a XX% probability that a person with this number will develop this disease in how many years .”

This seems to be possible if we have enough data, but it is not easy to do because of the personal information protection law in Japan.

It is difficult to collect data from all over the country because data such as physical examination data cannot be released from local governments, and resolving such data fragmentation is an important issue.

Federated learning , which allows AI to integrate intelligence, is currently being researched, but it will be necessary to advance research, including its application to medicine.

Privacy can be protected by making statistical data within the municipality and sharing it outside.

As the research progresses, it may eventually be possible to set up a separate AI for federated learning in each prefecture, and create a state in which intelligence is connected behind the scenes.

The third topic is about the ” symbol grounding problem “.

In this regard, the committee members

Although it is possible to link image feature values ​​with language, the problem is that the feature values ​​of the sensor-actuator complex cannot be extracted.

It’s not that I don’t know how to do it, but it’s still not resolved. Since the feature value of the sensor-actuator complex is also important in robot research, I think it would be better to do it as a government research and development project.

For example, you can’t make a robot that skips right now . Putting aside the actuator output problem, such a simple, or rather basic operation has not yet been achieved.

Ultimately, this is because the features of the sensor-actuator complex have not been extracted.

There was an opinion.

In response to this, there was a question, ” Is it possible to make a robot that can skip using reinforcement learning? “

For reinforcement learning, it is necessary to set a reward function. In other words, it is necessary to judge whether the state is close to the skip state.

In the first place, learning is not possible unless the concept of the skip state is acquired. For this reason, we end up using the feature values ​​of the sensors and actuators .

It also defines the reward function, and it feels like the actions are layered accordingly.

The symbol grounding problem is a difficult AI problem like the frame problem ( see the 2nd Committee article ). If we can solve this problem, it will be an important research that can be expected to enable AI to understand the meaning of words.

Life/City

In the section on lifestyles and cities, discussion on “ research dealing with modalities that cannot be perceived by humans” was a key point.

From the committee members

It is important how to handle multimodal sensing, not limited to images, in an integrated manner.

For example, electromagnetic waves contain a huge amount of information about wavelengths, and humans can perceive only a small part of them.

There was an opinion.

High-frequency electromagnetic waves are also used for predictive detection in order to analyze structural damage and cracks.

Handling high frequencies is difficult, but if the processing power on the edge side increases, it may be possible to handle them properly. However, this is only a small part of the story, and various multimodal sensors will likely emerge.

In addition, if past earthquakes are converted into data, information inside the earth’s crust can be obtained, which may be used for predicting earthquakes.

Furthermore, regarding the frequency, the following opinions were given by the committee members.

Medical equipment (MRI, etc.) is restored to a three-dimensional structure so that it is easier for humans to see, but it must be sensing more information than that.

If you don’t drop it, more information may come out.

It is premised on doing things in line with the five senses of humans, but if analysis methods advance and that part is removed, there is a possibility that we will see something different.

Electromagnetic waves have a large spectrum, so by analyzing all of them, we may be able to find out what is happening in the city .

AI science

In the AI ​​science item, the discussion progressed on new possibilities arising from the addition of AI to science .

First , a committee member asked about the topic, ” What can be done when human modalities are removed? “

It would be great if there was a way to count how many things we have not been able to perceive or perceive. It is also a kind of frame problem.

Also, when it comes to “what is science for?” , AI cannot be planned unless it has a higher goal, such as ” I want to cure cancer ” or ” I want to predict infectious diseases and earthquakes .”

In other words, the next-higher hierarchy reflects human will as to what they want to use science for as a result .

There was an opinion.

Although it is not limited to science, ultimately clarifying ” what are human desires and ideals ” may be important for AI to be developed in the future.

Next , we entered into a discussion about how to explain to humans what AI is doing when AI becomes more intelligent .

Regarding this, there was also an opinion that ” it may not be necessary to explain “.

For example, many people today don’t understand how smartphones and cars work, yet they trust the devices themselves.

Since the current stage is in its early stages, some people may feel uncomfortable if AI does not explain it, but in the future AI may permeate without explanation or understanding of its mechanism, just like smartphones and automobiles.

In this regard, another committee member

No description is currently defined .

For example, it is possible to think that the explanation depends on whether the recipient is convinced. Shouldn’t we start with that research?

For example, when a doctor asks for an explanation from a patient, the doctor responds to it, and if we use this to collect data on the doctor’s explanation, we may be able to see ” what is the explanation? “

On the other hand, there is also a story that patients can be convinced if they are told by a doctor. For example, “Why is there a fever at 37.5 degrees Celsius? ‘ and so on, once social conventions are somehow acquired, there will be no need for further explanation.

The opinion was stated.

summary

This time, we introduced the state of the 4th AI Action Plan Formulation Committee .

The topic of “ health and medical care ” was particularly important during the discussion .

This topic is particularly important in the discussion because it focuses on AI-human interfaces and has ethical issues.

At the next 5th meeting of the committee, we will enter the stage of deepening some discussions and deciding on the contents of an action plan.

The New Energy and Industrial Technology Development Organization (NEDO), a national research and development agency, held the “ NEDO Symposium on Action Plans and Symposiums for Overall Research and Development in the Field of Artificial Intelligence (AI) – The Direction of AI Social Implementation that Japan Should Aim for – ” will be held as follows.

In this symposium, we will introduce the results of the ” Survey for the formulation of a broad R&D action plan and business extraction in the field of artificial intelligence (AI) technology “, as well as the expert members of the action plan formulation committee and Japan’s aim We will discuss the direction of social implementation of AI that should be.

If you would like to participate, please apply here.

NEDO Symposium on Action Plans for Overall R&D in the Field of Artificial Intelligence (AI) Technology -Direction of Social Implementation of AI that Japan Should Aim for-