AI Highlights — the whole picture — 2026-10-04 (Sun)

Investment in AI infrastructure is said to be supporting the U.S. economy, and Anthropic will invest $100 million in human resources development. The numbers are large and give the impression of continued growth. On the other hand, it has been reported that middle school students are seeking AI friends rather than human friends, and parents and teachers are protesting against the introduction of AI in schools. The speed at which adoption spreads is not the same as the speed at which its impact becomes apparent. This article sequentially traces developments in finance, government, education, creative writing, and professional fields, and shifts the perspective from the scale of the work to who is responsible for the costs and responsibilities. I would like to give readers who follow the expansion numbers a clue so that they do not overlook the changes that are occurring behind the scenes.

Massive investments supporting the economy and shaky assumptions

There is a view that investment in AI infrastructure is supporting the current U.S. economy. LinkedIn's co-founder said AI infrastructure is "the only reason we're not in a recession" (Business Insider). Funds will go towards human resources as well as equipment. Anthropic says it will invest $100 million in the Cloud Frontier Academy to train 10,000 engineers. Additionally, Anthropic's $2 trillion AI initiative has been reported (calcalistech.com). Because equipment, human resources, and financial plans are all getting bigger, if you only look at the scale of introduction, it's easy to get the impression that growth will continue.

However, an article about the same concept points out that it is built on a series of contradictions (calcalistech.com). The presence of large numbers is not in itself proof that the assumptions supporting those numbers are solid. If the economy is dependent on AI infrastructure, the impact will be greater if that premise breaks down. The Cato Institute (cato.org), a U.S. think tank, argues that even a slowdown in AI could cause severe economic disruption. A condition readers of investment plans should keep in mind is that even a deceleration, rather than a sudden decline, can have ripple effects.

There are also movements at hand that indicate a wavering of the premise. New York AI company leases 2.2 million square feet as entry-level hiring slows (citybiz). Although offices are expanding, the number of entrances is decreasing. If the plan to train 10,000 engineers is progressing, but the number of new recruits is shrinking, it is necessary to separately check the training numbers and the actual employment situation. An American man has also been arrested on suspicion of being involved in a $300 million scheme to smuggle Nvidia AI servers to China (South China Morning Post). Strong demand has also given rise to efforts to bypass supply controls.

The take-home decision for the reader is not to treat size of investment as a guarantee of growth. The funds that support the economy are piling up in the same place as uncertainties such as contradictions, slowdown risks, narrowing employment opportunities, and illicit distribution. The next question to be asked then is who is paying for this expansion outside of the numbers?

Accountability required in the field of finance and government

In China's banking industry, analysts believe that more Chinese banks are likely to adopt AI rules after Ping An Bank's move (South China Morning Post). Rules are being developed in such a way that regulations and management take the lead. Munich Re supports Manulife Carebloc after German insurance group ERGO elevates AI to board level (AD HOC NEWS). The movement to treat AI not only as a tool in the field but also as a management responsibility is spreading from banks to insurance. The UK government's draft policy is in the same vein, instructing staff to avoid thanking chatbots and instead use short prompts to save energy (The Cool Down). This means that we have entered the stage where even the etiquette for use will be defined in writing.

However, having rules in place and acting correctly in front of customers are two different things. Commonwealth Bank's AI customer bot has come under fire for cutting savings interest rates, with users calling it "pathetic" (Yahoo Finance Australia). This is an example of how automatic responses undermined trust when conveying the unfavorable news of a downgrade. If readers are to evaluate the introduction of AI by banks and insurance companies, they should look not at the number of cases introduced or the speed of response, but rather whether the system is designed to have a human take over in the event of an unfavorable notice or complaint.

In the public sphere, the cost of mistakes is heavier. A grandmother accused of bank theft after allegedly using AI facial recognition to verify her identity has filed a $10 million lawsuit (ABC News). If the verification results are incorrect, the burden is not on the operator of the system that made the wrong decision, but on the individual suspected. On the safety front, New Mexico's AG and Santa Fe legislators are touting AI safety measures after the UNM breach (Searchlight New Mexico). It shows countermeasures after a breach has occurred, and the order is reversed from the situation in which advance preparedness is evaluated.

Selection by companies is also unstable. META unbundled Virtue AI after just four months of adoption, reports say (TradingView). While reviewing in a short period of time indicates that implementation decisions can be made quickly, it also indicates that evaluation and verification can be delayed. The effectiveness of escalating the matter to the board of directors, internal documents, or announcing countermeasures after the fact is determined by who gives an explanation and who compensates when a mistake occurs. In the next section, we will look at how that burden is shifted to users and employees.

Relationship between children and AI that enters their hearts

WSJ reported that middle school students want AI friends instead of real human friends (WSJ). Globe Opinion (Substack) has a topic on ``Why you can't break up with your AI boyfriend,'' and the relationship with AI is not just a temporary play, but is starting to be talked about as an addiction that is difficult to break up with. What we should look at here is not the number of users, but the fact that situations in which AI is being used to replace human relationships with friends and lovers are expanding to the age group of middle school students. The AI partner is always responsive and does not need any adjustments to maintain the relationship. Parents and teachers are in the position of deciding how these characteristics will affect children who are at the age when they are learning how to interact with others.

On the school side, the same question has become a concrete conflict. GIGAZINE reported that as Google's AI permeates school education, there is growing opposition from parents and teachers who argue that it is having a negative impact on children's cognitive functions and grades. Meanwhile, in the US, Mike Moras' agency helped bring AI tools to public school classrooms in Texas, Houston Public Media reports. Businesses such as agencies who handle sales and public relations are involved in the implementation process. Opponents base their opposition on the impact on grades and cognition. The pros and cons of introducing the system are contested between judgments made outside the classroom and changes observed inside the classroom.

The temperature of the debate varies depending on the country and situation. Fortune India framed the debate surrounding AI in the classroom as a "pay attention, but continue to respond according to the situation." Our position is not to introduce it completely or ban it completely, but to make decisions on a case-by-case basis. Similar forums have been set up internationally. The international AI x education summit ``Raspberry Fields'' brought together more than 200 people from 45 countries, including Japan, in Cambridge, UK, to discuss ``the future of children and AI education.'' The participation report has been published (Infoseek).

When lined up in this way, AI as a friend or lover, AI as a learning tool in the classroom, and discussions among educators in various countries all lead to the same question. The question is: who will check what AI will bring to children's cognition and human relationships before it is introduced, and who will take on the responsibility when problems arise? The fact that usage continues to spread without a clear answer to this question is common to the impact it has on people working inside organizations, which we will discuss next.

Fluctuating value judgments regarding consciousness and creation

The line of what is allowed in AI is being questioned simultaneously in religion, science, creativity, and ethics. According to Diya TV, Anthropic has asked the Vatican to reconsider its position on AI consciousness. Firstpost explains why the company's Claude's "moral standing" is under serious debate. Meanwhile, Popular Mechanics reports that a scientist said, "There is a key to maintaining consciousness, and AI doesn't have it." Development companies are asking the religious community to reconsider their position, and scientists are expressing a negative view, citing the conditions of consciousness. What the reader should look at is not the conclusion itself as to whether there is consciousness or not, but who is deciding on what basis to decide "yes" or "no."

In the area of creativity, evaluations are divided. cybernews.com reported that Pope Leo's criticism of AI art targeted machine-generated images. Meanwhile, according to MusicTech, Todd Edwards was optimistic about AI in music production, saying, "It's frustrating to hear people bash something that's so great." Critics base their claims on images produced by machines, while defenders base their claims on their achievements as tools for music production. Even though the term "AI work" is the same, images and music are looking at different things. This controversy is directly connected to the question of the evaluation that creators receive and who is at a disadvantage in the field of creation.

Research on moral judgment and bias also adds material to this line. According to Bioengineer.org, AI crowds track humans' moral judgments by voting rather than by conversations. Yahoo also reported that a study found that a popular Chinese AI has a pro-China bias. The former is a result that shows how close the output of AI can be to human value judgments, and the latter is a result that shows that the position of the provider can be mixed into the output. The more situations in which decisions are entrusted to AI, the more the user is left with the task of confirming which group or position the decision reflects.

The evaluation of machines reflects the values of humans, whether they are conscious or not, the value of their work, and the fairness of their judgments. The question of who decides on the evaluation and who bears the results leads to the distribution of the burden at the workplace, which we will discuss next.

Advanced practical application in professional settings and the role of people

Business Insider's example of a business owner clearly illustrates the changes in the workplace. This person runs an AI company with 34 AI agents, and writes that his daily routine consists of Diet Coke, DoorDash, and "Claude's Walks" (Business Insider). This is an example of talking about a company in terms of the number of agents rather than the number of employees, and it shows in the person's own words that a person's job has shifted from execution to instruction and confirmation.

Applications are becoming more specific in nursing care, diagnostic imaging, livestock, and veterinary medicine. Skilled Nursing News reported on how AI can help Trilogy and Signature nursing home operations with fall detection and auditing (Skilled Nursing News). In diagnostic imaging, a review found that AI was surprisingly accurate in diagnosing chronic ankle instability (Bioengineer.org). In the pathology field, Leica Biosystems has introduced Tempus' pathology products to the Aperio AI Store (Manufacturing Chemist). In livestock farming, a small AI can weigh a cow over the phone in less than 52 milliseconds (Bioengineer.org).

In veterinary medicine, demand influences business trends. Demand for AI in veterinary medicine prompted Chicago Pacific to withdraw from CoVetAI, according to pehub.com. The same article also mentions that HealthEdge, MPK Equity, and TowerBrook are targeting skin care. Fall detection in nursing care turns monitoring into recording and auditing. In diagnosis and pathology, some of the work of reading images is transferred to software. In some cases, tasks that used to be done on-site, such as measuring the weight of cows, can now be completed over the phone.

What these examples have in common is a shift in the role of people from those who carry out the work to those who oversee and judge the results. What readers should be aware of is not so much what AI can do, but who has the final say and who takes responsibility when an error occurs. Missed falls and misdiagnosed falls have direct consequences for people. The next question is how to design the burden and responsibility of those entrusted with making decisions.

The trends discussed in each section are connected in one point: the responsibilities of trust, human relationships, recognition, and judgment have not kept up with the speed at which AI is being introduced. The scale of investment and human resource development is large, and banks, insurance, and governments are establishing regulations. However, having rules and acting correctly in the field are two different things, and there is already opposition and concern about children's relationships with others and their learning at school. It has not yet been decided who will draw the line between consciousness and creation. As the example of managers using 34 agents shows, people's roles are shifting from execution to direction and confirmation. As the number of people responsible for checking increases, it becomes necessary to clarify who is responsible for the results. The scale of implementation is an indicator of success, but not of who will pay the price.

Q1: Are you able to check the results of work entrusted to AI to the same level as before introduction? Q2: Even if rules and guidelines have been established, what kind of evidence do you use to ensure that actual usage is in accordance with them? Q3: Have you decided in advance who will bear the burden in the event of an error or disadvantage arising from the introduction of AI?

Finally, three questions

- Are we able to confirm the results of work entrusted to AI to the same level as before introduction? Q2: Even if rules and guidelines have been established, what kind of evidence do you use to ensure that actual usage is in accordance with them? Q3: Have you decided in advance who will bear the burden in the event of an error or disadvantage arising from the introduction of AI? - Even if rules and guidelines have been established, what evidence do you have to ensure that actual usage is in accordance with them? Q3: Have you decided in advance who will bear the burden in the event of an error or disadvantage arising from the introduction of AI? - Have you decided in advance who will bear the burden in the event of an error or disadvantage arising from the introduction of AI?