AI Highlights — the whole picture — 2026-10-11 (Sun) Morning News

U.S. economic growth, supported by AI-related investment and demand, coincided with Oracle's layoffs. Strong demand and fewer people are the result of a single investment. On the other hand, in imaging diagnostics, research is competing for accuracy in areas such as the breast, liver, and brain, and the basis for AI decisions in operating rooms and nursing care settings is beginning to be directly linked to human safety. The introduction of the system is progressing, and the numbers showing the results are increasing, but it is not clear from the headline of the article to what extent these numbers are being verified or who is responsible for checking them. The reports gathered over the past half day indicate that we have entered a phase where decisions about how far to entrust AI are determined not by the size of the results, but by how much verification has been done.

Single-legged growth and reorganization of personnel and training

A US bank's investment manager has pointed out that the US economy is dependent on AI (The Dark Side Of The Boom). The view is that while AI-related investment and demand are supporting growth, support from other fields is correspondingly weak. Around the same time, an article appeared (Times Now) discussing the pros and cons of Oracle's personnel cuts from the perspective of the responsibilities of the "landlord" who lends infrastructure in the age of AI. Strong demand and a decline in the number of people are not separate issues, but are two sides of the same coin. On the field side, Gal Limon, founder of Centrical, talks about managing field personnel using AI (Pulse 2.0). If AI is responsible for managing deployment and operation, introducing AI will not only reduce the number of people, but also reorganize the management method itself.

There are also moves to reorganize training and operations. It was reported that Job Corps in the United States is making a comeback with a focus on vocational training, which is said to be difficult to replace by AI (Wyoming Public Media). This is an example of reselecting the content of training based on the premise of jobs that will be replaced by AI. In marketing, industry media has introduced how marketers are using AI today (Precision Farming Dealer). Rather than jobs disappearing, it seems that the distribution of work within the same occupation is changing. As for industry-specific developments, Beijing Haizhi and Jointown have entered into an AI partnership to advance the sophistication of pharmaceutical distribution (The Globe and Mail). Even in fields such as distribution, which involve a large number of goods and procedures, a movement has begun to incorporate AI into operations through partnerships.

When you put these things side by side, there are two points to look at. One is that if the returns to growth are biased towards AI, employment and investment will depend on whether that demand continues. Another issue is that personnel reductions, on-site management, training, and reassignment of work are proceeding at the same time in different locations. The materials presented here are mainly statements and announcements of alliances, and are not numbers that measure the actual status of replacement. Therefore, before making decisions about introduction or hiring based on the strength of demand, it is necessary to check on a site-by-site basis to see which jobs will be replaced and how.

As the reorganization of people and jobs becomes more widespread, the questions of who will check whether the judgments and proposals made by AI are correct and who will be responsible will become more important. The next focus will be on how funds and politics support and constrain this movement.

The cost of competition in elections and data centers

Regarding the AI competition between the US and China, China-US Focus reports that the composition is changing completely. One indication of this is a report that Tencent's Hunyuan has ranked first among Chinese companies in Arena's alignment index (BigGo Finance). This is said to be Yao Shunyu's first performance since taking office. Alignment is a measure of how well a model conforms to human intentions and safety constraints. Following the competition in performance, the US and China have entered a stage where they are also competing for rankings in indicators of safety and controllability. However, ranking first in one index does not guarantee safety in actual operation. Implementers need to separately check the ranking table position and the verification results for their own use.

The cost of competition has surfaced as an electoral issue in the United States. CNBC reported that the midterm elections are becoming AI elections, with Wall Street bracing for growing opposition to data centers. A data center is a physical facility that requires electricity and land, and it is also something that residents of the construction site can vote for or against. Investors may be wary of growing opposition because they see capital investment plans as potentially dependent on election results and local agreements. For companies considering the introduction of AI, the possibility that the supply of computing resources to be used may be influenced by politics is a condition that should be factored into contracts and procurement.

In terms of security, Korea JoongAng Daily reports that a Chinese-made AI tool that was implicated in the hacking of a South Korean bank has stopped public updates. The involvement is merely an indication, and the reason for the suspension of updates cannot be determined from this report alone. Still, it clearly demonstrated that AI tools can be used to attack critical sites such as financial institutions, and that it is difficult to verify when incorporating tools of different origins. Tools that have stopped being updated may not be fixed even if vulnerabilities are discovered. When adopting a product, it is necessary to check the provider's update policy and continuity of supply in addition to performance and price.

Three events demonstrate that competition is not just a matter of speed, but also comes with costs such as public consent, financial security, and model control. The next question is who will bear the cost and how will it be verified on-site?

Accuracy of image diagnosis and distance to reach the scene

Research aimed at prediction and differentiation has been reported in three areas: ultrasound, MRI, and pathological imaging. Regarding ultrasound, a study was published that uses two-step deep learning using grayscale images to automatically isolate subpleural lung lesions and differentiate between benign and malignant lesions (BMC Medical Imaging). The same medium also includes ultrasound deep learning radiomics for preoperative prediction of sentinel lymph node metastasis in invasive breast cancer, a two-institution retrospective study (BMC Medical Imaging). Regarding MRI, an attempt was reported to predict TACE refractoriness of hepatocellular carcinoma by combining habitat radiomics of multiparametric MRI and temporal changes in clinical indicators (Scientific Reports). In pathology, there is research using AI to stratify the prognostic risk of adult-type diffuse glioma using H&E-stained slides (Cancers).

Although the targets and types of images differ in these cases, the ranges confirmed in each case are different. As the breast cancer research clearly states, it was ``retrospective at two institutions,'' most of the studies were conducted using past data, and were not prospectively applied to future patients to measure results. When readers see the numbers, what they should be looking at is not so much the accuracy, but the design of the verification, including how many facilities and whether it is retrospective or forward-looking.

Research is also expanding into the search for explainability and therapeutic targets. A study using explainable machine learning to characterize radiation esophagitis in lung cancer patients using data from the NRG/RTOG 0617 clinical trial has been published (Artificial Intelligence in Health). For prostate cancer, we are stratifying the prognosis of highly malignant cases using E2F-G2M signatures derived using machine learning, and are also searching for therapeutic targets (Cell Communication and Signaling). Here, the role of AI is expanding from judging images to indicating risk factors. However, whether the factors shown can be used for clinical decision-making requires separate verification.

On the product side, Scanvio Assist received the world's first CE mark for AI ultrasound assistance for endometriosis (Startupticker). The appearance of cases that have been approved within the framework of regulations indicates that a path has opened for research results to be put into practical use. On the other hand, the CRB's investigation found delays in AI implementation in regulated biopharmaceutical operations (Lab Manager). Although biopharmaceutical work and image diagnosis are not the same work, they have in common that in regulated fields, the performance shown in research does not directly serve as the basis for introduction.

As the number of studies increases, those deciding whether to introduce the technology will be less concerned with the actual results and more concerned with what kind of verification the numbers have undergone. How deep should the verification go in order for the company to take on responsibility in the field? In the next section, we will look at how this question appears in settings other than diagnosis.

Beware of bias and “pretending to care”

In operating rooms and nursing care settings, the basis for AI decisions is directly connected to people's safety and daily lives. Bioengineer.org reports that bias in surgical instruments can mislead AI in the operating room. It has been pointed out that if there is a bias related to the equipment, AI could interpret what is happening in the operating room in a different way than it actually is. Business Insider reports on a case in which AI introduced a caregiver to a man who became paralyzed. AI is starting to become involved in decisions that are central to daily life, such as choosing a caregiver. However, what can be gleaned from both articles is ``on what basis are decisions made?'' rather than ``whether or not it can be used.'' Unless we can confirm what information is included in the introduction and recognition criteria and whether there is any bias, a successful case cannot be considered as a basis for introduction.

In situations related to mental health, the questions asked change. IOL points out that AI only pretends to care, but doesn't feel anything. Meanwhile, The Business Journals reports that Ai2 (Allen Institute for AI) has won the Wellbeing Excellence Award 2026. While some efforts have been praised by outsiders, there are also those who believe that AI's concern is just a façade. These two are not contradictory, but have different targets for evaluation. The award is an evaluation of the organization's efforts, not the result of verifying that AI behaves appropriately in front of individual users. Just because you sound concerned does not mean that your concern has been confirmed.

In the case of companions, the issue is whether the users themselves understand this difference. mos.ru explains the difference between the relationship between AI companion apps and reality. If the response feels warm but the other person feels nothing, what the user should expect from the relationship depends on the provider's explanation. What those implementing the system should check is not how comfortable the response is, but whether users are aware of what can and cannot be done.

What all three situations have in common is that there is a difference between the results looking good and having a reason to trust them. Who checks the basis of trust, how does it work, and who takes responsibility when a mistake is made? The next section deals with the confirmation procedure and where the responsibility lies.

Drawing the line that leaves intentions and souls in humans

The Economic Times juxtaposes Pope Leo Religious leaders and AI developers are facing the same question. It can be seen that the word "soul" has begun to relate not only to the realm of faith but also to policies on how to design and operate AI. However, the only thing that can be ascertained from the title is that the question is posed, not that the answer has been determined. Those considering the introduction need to decide in their own words the role that will remain for humans before waiting for an answer.

In the field of creation, there are concrete instructions on how to draw the line. According to Indie Boulevard Magazine, Ute Lemper defended the AI-based work ``MOTHERS'' and said, ``It is the artist, not the technology, that creates the intention.'' From this standpoint, whether or not AI is used is not a criterion for determining the value of a work. The standard is where the intention behind what to create comes from. Even if the tools change, the intention and responsibility remain with the person. However, this is Lemper's own defense, and his views may not be shared by those on the receiving end. When using AI, the first thing to check is whether or not the person who will be responsible for the work can be expressed in words.

There are similar lines in everyday choices. Kenta Tokui's Bodhisattva Perspective No. 292, published on Yahoo! News, uses the theme of a family Netflix competition before going to bed, and writes that there is no better recommendation than the word of mouth of acquaintances and friends than AI. Even in the small scene of choosing a movie to watch, people are choosing whose recommendations to believe. This is one writer's personal experience, not statistical support. Still, the form of trust in which someone you can see face-to-face makes a recommendation in their own words, and you are responsible for the results, shows a different kind of performance than performance numbers.

The question of soul, intent, and word-of-mouth is not whether AI can do it, but who will take care of it. The next question is how to apply this line to on-site operations.

The topics covered in each section are different from each other, such as employment, elections, diagnostic imaging, nursing care, and creativity, but they all end up in the same place. Even if the results of AI are shown, if it is not known in what subjects and under what conditions the results were confirmed, there is no basis for its introduction. Research on diagnostic imaging differs in the extent to which it has been confirmed, cases of operating rooms and nursing care show that the criteria for judgment are not visible, and cases of creative writing concretely illustrate the line that places the subject with the intention on the human side. When deciding to introduce a system, what should be considered is the depth of verification and who is responsible for the results, rather than ranking or accuracy. Results numbers are only the starting point, and the focus of judgment is on confirmation from there.

Q1: To what extent have the results of the AI you are planning to introduce been verified based on your own targets and conditions? Q2: If there is an error or bias in the AI's judgment, is it determined before operation who will take responsibility and at what point? Q3: Are you making decisions to cut down on the time and cost of verification just because the results are good?

Finally, three questions

- To what extent have the results of the AI you are planning to introduce been verified based on your own targets and conditions? Q2: If there is an error or bias in the AI's judgment, is it determined before operation who will take responsibility and at what point? Q3: Are you making decisions to cut down on the time and cost of verification just because the results are good? - If there is an error or bias in the AI's judgment, is it determined before operation who will take responsibility and at what point? Q3: Are you making decisions to cut down on the time and cost of verification just because the results are good? - Are you making decisions to reduce the time and cost of verification just because the results are good?