AI Highlights — the whole picture — 2026-11-10 (Tue) Daily News

It is reported that the ballooning of the total winning bid of $5.7 billion was caused not by the actual construction of the facility, but by a single forecast that anticipated demand for AI. Around the same time, there were also reports that banks were starting to restrict AI-based connections to accounts. Replacement of funds, power, and personnel is progressing rapidly based on projected numbers. On the other hand, flaws have been found in the conditions for allowing connections, the independence of institutions that verify safety, and the assumptions made in the research. What you should decide after reading this half-day article is not how fast the implementation was, but who has verified the basis on which the decision is being made and to what extent. The question of whether the certainty of footing is given the same weight as speed is the basis for each of the following sections.

Demand forecast drives electricity and politics

In the US electricity auction, it is reported that just one AI demand forecast increased the total winning bid by $5.7 billion (CryptoSlate). The price was driven by the expected figures, not the actual facilities built. The same trend can be seen on the political side. CEO points out that AI investment will continue to expand even after Democrats take back Congress (Yahoo Finance). If the direction of investment does not change even if the power of the government or Congress changes, corporate plans are less likely to be influenced by election results. Furthermore, Lowy Institute's ``Asia Power Snapshot'' focuses on the battle for AI supremacy between the United States and China (Lowy Institute). Demand forecasts, the balance of power in Congress, and competition between countries all drive the same investment movement.

The momentum of funding is not limited to electricity, but also extends to human resources and contract conditions. Turkey announced a five-year plan for AI talent and data (Türkiye Today). Rather than leaving the use of AI to the private sector, this is a move to incorporate human development and data preparation into national plans. Among companies, Laboratoires Pierre Fabre and XtalPi announced an AI-driven collaboration focused on drug discovery (Press Release Hub). Meanwhile, the luxury fashion giant's deal reportedly seeks a wide range of AI-related rights (South China Morning Post). While collaboration progresses in the form of sharing the results, the rights to AI themselves are at issue in contracts. The more money is collected, the more difficult the agreements become over who gets to use and keep what.

What is important for the reader's judgment is that expected numbers drive actual prices, policies, and contracts first. The $5.7 billion amount came from one demand forecast. If the forecast is wrong, the burden can fall on parties other than those who made the forecast, including electricity buyers. Turkey's five-year plan, the collaboration between Pierre Fabre and XtalPi, and the terms of a contract with a major luxury fashion company are all built on pre-determined assumptions. Therefore, it is not appropriate to measure the certainty of implementation solely by the size of the investment or the number of announcements. We need to see to what extent the assumptions have been confirmed, and how and to whom rights and burdens are allocated. When a system is determined in advance with funding, the next question is whether the system will withstand the way it will actually be used.

Reduction in personnel and reorganization of roles

Disney reportedly plans to cut thousands of jobs while handing over some jobs to robots (Inside the Magic). Around the same time, a bill was proposed that would include mechanisms to protect workers from AI-induced layoffs (Philstar.com). This means that the movement for replacement and the movement for protection emerged side by side during the same period. However, from the headline of the article at hand, it is not possible to confirm the types of jobs targeted for reduction or the specific content of the bill. Since the two cases have different contexts, it is appropriate to treat them as separate materials rather than combining them into a single stream. The first thing readers should look at is how quickly the reduction figures and the institutional response to them appear in pairs.

On the other hand, work reorganization is progressing in a different direction than job reduction. Range's Ragone says that a planner with no coding experience built 24 AI agents in six months (BigGo Finance). In the same week as the article about reductions, there was an example of someone who had not previously been on the creator side becoming a creator. The figures are for one company, and cannot necessarily be extended to other companies. Still, it is clear that these two articles cannot be explained at the same time by choosing either ``jobs disappear'' or ``jobs increase.'' Replacement and role change can occur simultaneously within the same organization.

Articles have also been published on how to deal with employee anxiety about these changes. Google Business Profile discusses the idea of turning concerns about AI into successful AI implementation. From a different angle, The Newark Advocate asks whether our reliance on AI is undermining human connection. The former is about the operation of those who are promoting the introduction, and the latter is about the lives of the users, so their positions are different. However, they both overlap in that concerns should not be ignored and should be treated as a prerequisite for introduction.

When deciding whether to introduce this service at your company or how you want to work, you need to check not only the number of people you have to reduce or the number of agents you have built, but also the steps you can take to address concerns and whether you still have contact with people. So, how reliable are the parties supporting its introduction, that is, the conditions of connection and the independence of the provider? We will look at this in the next section.

Failure that occurs at the boundary between connection and trust

Yahoo Finance named three payment stocks to watch as banks restrict AI-based connection to accounts (Yahoo Finance). The bank that holds the account, rather than the user, is starting to draw the line as to whether or not to allow AI to touch the account. Meanwhile, officials from the U.S. Center for AI Safety have reported to the White House, raising concerns about the center's independence (ABC News). If the role of ensuring safety is shaken, questions will be asked on what basis other decisions on introduction will be made. There are question marks attached to both the side that allows the connection and the side that verifies the connection.

A flaw was also found in the search chain. RAG, which runs on devices without radio waves, updates its knowledge with signed knowledge packs. However, what the signature guarantees is who published it, not whether the embedding vector in the pack was calculated according to the text. arXiv's pre-peer-review preprint measured how many questions a method of matching vectors with a tolerance yields to an attacker, and showed a countermeasure to match quantized codes (on its own site). It is pointed out that whether the signature is correct and whether the contents are reliable are two different issues. However, it is necessary to read the authors' claims and conditions that cannot be confirmed in the abstract separately.

The scope of responsibility continues to expand. CUA-Master proposed a double cycle in which the operating agent, which clicks and inputs while looking at the screen, digs out the work flow from the instructional video and synthesizes the tasks, and then decides the next data according to the model's weaknesses. The authors report scores that are competitive with others on multiple benchmarks, but they do not list in the abstract who they were compared to and the size of the differences (on their own website). DR Pixel is a deep research agent that searches and infers from page images without transcribing papers using OCR, and has reported that it has outperformed methods using OCR without training. The authors conclude that ``OCR is not only unnecessary, but also not optimal,'' but you need to read the abstract to see how far that claim extends (on their website). Both are preprints that have not yet been peer-reviewed.

As the number of agents responsible for operations and searches expands, the question of who should be entrusted with what extent becomes a point of contention. Restrictions on banks, questions about safety centers, and blind spots in signatures all show that the line is not clearly drawn. Next, we will look at who is responsible for drawing this line from the perspective of systems and organizations.

On-site measurement progressing in medical care, agriculture, and nursing care

On the medical side, how the output of AI is explained and how it is applied to clinical practice determines how the results are interpreted. A systematic review of Alzheimer's diagnosis featured by Bioengineer.org examines the combination of explainable AI and RDoC (Research Domain Criteria). The analysis of pancancer classification published in the SUST Journal of Science and Technology (SUST JST) is a comprehensive comparison of machine learning models with an eye toward clinical application. Cancer Cell deals with the spatial heterogeneity of neutrophils in breast cancer and its clinical implications. In all three cases, the topics include not only accuracy but also the reasons for the results and how to use them in clinical practice. Rather than reporting that ``AI guessed it'', it is better for readers to check first whether the explanation is explained and the situations in which it is used are written.

The same perspective can be used in the area of prognosis prediction and measurement. Discover Oncology's hepatocellular carcinoma research developed and validated HDAC-related prognostic signatures using single-cell transcriptomics and machine learning. The rice research reported by Newswise uses AI and 3D images to measure structures that are difficult to see from the outside. Regarding the elderly, The Malaysian Reserve reported on the Senior Industry Expo in Chongqing, China, and reported that there was an exhibition of new services for the elderly using AI. A signature consists of two stages: ``development'' and ``verification,'' so the way it is read differs depending on which group conducted the verification. Measuring rice is a result of converting the invisible into numerical values. Exhibits at industrial expos only show the stage at which the services were presented, and the actual results of use need to be verified separately.

The more you get into the professional field, the more you can no longer measure the results of AI with a single number of precision, and you will need to consider how it is explained, who is verifying it, and how it is used. This perspective comes down to the question of how solid the scaffolding is that supports the premises of research and implementation.

Re-examining questions about fairness and awareness

Regarding how to measure the educational effectiveness of universities, Bioengineer.org introduces a framework for fair evaluation that combines multiple methods such as deep learning (Bioengineer.org). What I would like to focus on here is that the issue is less about the novelty of the method and more about the condition of ``fairness.'' Evaluation by AI does not necessarily become objective once it is introduced. If we don't confirm that it's fair for which students and universities, even if the numbers come out, they won't serve as a basis for making decisions. When looking at this type of framework, it is better to first look at what is considered fair when designing it, rather than focusing on its high accuracy.

Even in the debate over consciousness, the way we make assumptions is being questioned. The Decoder reports that viewing the brain as a computer may mislead the debate about AI consciousness (The Decoder). If we consider the brain as a computer, the question naturally arises as to whether computers become conscious if they become sufficiently complex. However, if that view itself is not valid, the discussion will deviate from its starting point. When I come across an article or statement that talks about whether AI has consciousness, I want to check how the brain and machine are related before coming to a conclusion. Analogies help explain, but they are not a substitute for evidence.

On the receiving side, AI has also entered the works and stories. An AI art exhibition (Tipperary Live) was held at Killoskehan Castle in Tipperary, Ireland. The project to show AI works in a historical place such as a castle can be said to be an attempt to bring AI to the public as an object to be appreciated, rather than just a technology. On the other hand, Alternative Blog is dissecting the story and voice actors for episode 10 of ``Satsu no Good News'' (Alternative Blog). The title suggests that the story discusses loneliness and connection through the work of voice acting. Although the AI art exhibition and the analysis of voice actors' performances are separate topics, they can be read side by side in terms of what people see and receive from the work.

The evaluation of education, how we talk about consciousness, and the appreciation of works are all related to the premise of how to measure and perceive AI. In addition to looking at the speed and scale of introduction, confirming that the yardstick is appropriate will be the basis for determining the reliability of the next step.

The movements covered in each section come together at one point. Whether it's electricity, personnel, medical care, education, or consciousness, the output and forecast numbers move first, and the assumptions that support those numbers have to be confirmed later. Demand forecasts drive prices, plans for reductions go hand in hand with legislation seeking protection, and accuracy reports can be read differently depending on how they are explained and whether they are written in the context of use. Restrictions on connectivity, questions about independence, and limits on the scope guaranteed by signatures are also examples where unverified assumptions can affect decisions. Therefore, before counting the speed of implementation as an achievement, the current situation requires us to reconsider what that judgment was based on and who verified it.

Q1: Can you explain in your own words the basis and scope of application of the output of the AI you are using? Q2: When the assumptions at the time of approval for introduction change, are the procedures for detecting the change and redoing the decision functioning while maintaining the independence of the confirming party? Q3: Of the numbers supporting investment and personnel plans, how many are based on estimates rather than actual results, and who verified those estimates?

Finally, three questions

- Can you explain in your own words the basis and scope of application of the output of the AI you are using? Q2: When the assumptions at the time of approval for introduction change, are the procedures for detecting the change and redoing the decision functioning while maintaining the independence of the confirming party? Q3: Of the numbers supporting investment and personnel plans, how many are based on estimates rather than actual results, and who verified those estimates? - When the assumptions at the time of approval for introduction change, are the procedures for detecting the change and redoing the decision functioning while maintaining the independence of the confirming party? Q3: Of the numbers supporting investment and personnel plans, how many are based on estimates rather than actual results, and who verified those estimates? - How many of the numbers supporting investment and personnel plans are based on estimates rather than actual results, and who verified those estimates?