AI Highlights — the whole picture — 2026-10-18 (Sun)

While AI is being used as a means to reduce the number of people, it is also being talked about as a reason to rehire younger workers. The same technology is also used as a person to confide in, with experts pointing out its limitations. Funds continue to flow even under high interest rates, and the supply is unlikely to stop any time soon. At first glance, CBRE's office demand forecast and reports that seafarers are consulting AI seem to be separate stories. However, both scenes involve people's locations being rearranged by AI. What I want to find out in this half-day is what remains as a basis for decision-making when we look at the benefits and limitations of introduction in the same situation.

Jobs will be cut and replaced at the same time

CBRE predicted that office demand growth in the Asia-Pacific region will slow down due to the impact of AI-driven layoffs (Startup Fortune). If the number of people working decreases, the outlook for the necessary floor space will also decrease. Changes in employment are beginning to extend beyond hiring and pay to include real estate planning. Entrepreneur reports that AI has already become the biggest factor in layoffs. Retrenchment has moved from being a side effect of the introduction to being the primary reason for layoffs.

However, the same medium also conveys a different side. According to Entrepreneur, layoffs due to AI are also a stepping stone to starting a business. There are also moves in the opposite direction on the corporate side. Even with advances in AI, companies are likely to become more active in hiring young talent again (Mezha). A move to cut and a move to rebuild are occurring at the same time and in the same market.

Investment in places of learning continues. Microsoft will build a €1 million education hub on land adjacent to its data center in Ireland (Microsoft Source). The idea is to create a place to train people next to the land where computing resources will be placed. Investments are being made to develop people in places where technological equipment to reduce people is being built.

From the perspective of those making decisions about introduction, these are not issues that cancel each other out. If you look only at the number of personnel reductions, the introduction of AI appears to be a means of downsizing. If you only look at the recruitment of young people and the educational base, it appears to be a means of expansion. In reality, jobs that are being reduced are lined up with jobs that are newly required in the same company and in the same region. The department that reviews the size of offices and the department that designs ways to accommodate young workers cannot be operated separately.

The changes brought about by AI cannot be measured only by the amount of work done. One article (Northeast Mississippi Daily Journal) recalled the conversation with ChatGPT as ``interesting and disturbing.'' People may have already begun entrusting not only their work but also their thoughts and feelings to AI. In the next section, we will discuss AI as a person to consult with regarding your worries.

Funds do not decline, forcing a choice of camp

Even with continued high interest rates, the flow of funds into AI has not stopped. Inquirer.com reported that the AI boom continues unabated despite high interest rates, posing a challenge for the US Federal Reserve (Fed). Raising interest rates typically acts as a means to dampen investment enthusiasm. The Fed is making it difficult to make decisions in the face of the exception of the AI boom. The same trend exists outside the United States. Citi sees Chinese AI supply chain IPOs as solid despite lower valuations, according to Reuters. Capital continues to flow toward AI, both in the US interest rate environment and in the Chinese IPO market.

This point changes the outlook for those considering introduction. It is difficult to expect that the spread of AI will naturally slow down if the funding environment deteriorates. Suppliers should continue to provide products and services for the time being. In that case, reshuffles such as the reduction in personnel and rehiring of younger employees, as discussed in the previous section, could proceed without waiting for changes in market conditions. Whether to rush the introduction or postpone it will be decided based on the circumstances of the company, not on the availability of funds.

As long as the funds are not exhausted, the debate will shift to whose system should be chosen. Nikkei Asia reported that Southeast Asia is caught between the US and China's respective AI camps. The situation is such that the choice of procurement source can become the choice of the camp. There is also movement in rule-making. Anthropic has privately proposed its own AI safety bill to lawmakers, Reuters reported, citing Politico. Companies that create AI are beginning to bring their ideas for regulation into the legislative arena.

When reading these two factors together, when a company is choosing a supplier, it will not only consider the performance and price of the technology, but also what camp the supplier belongs to and what kind of regulations it wants to follow. The impact is even greater in regions like Southeast Asia, where there is less room for choice. If the system is introduced before the rules are in place, the person implementing the system will bear the burden of re-engineering its operations to match the regulations that will be decided later.

If funding continues and the lines between parties and regulations are drawn at the same time, the pros and cons of introducing the system will be inseparable from the work of verifying its effectiveness and limitations on the ground. This is because a reorganization is occurring not only in the way people work, but also in the people they rely on. In the next section, we will look at AI as the partner to rely on.

AI that listens to your worries and its limits

The use of AI is increasing in the field of mental health, KFYR-TV reports. In the same report, therapists point out the limitations of AI. TradeWinds News reported that seafarers are turning to AI to discuss their deepest concerns, and experts have warned of its limitations. General users and seafarers have different positions and environments. Still, the use of talking to AI about your worries is expanding, and experts are issuing similar warnings. What we can understand from this is that the increase in the number of users is not a good reason to introduce it. It is necessary to compare the person's track record as a consultant and the limitations cited by experts in the same situation.

The people you worry about are not just friends or professionals. An article published under the name Universidad de Córdoba (UCO) entitled ``Create an AI girlfriend for free'' treats the idea that you can create an AI to play the role of a lover for free. The roles that people want machines to play range from listening for advice to being partners in intimate relationships. In this area, it is also necessary to confirm whether the warnings of therapists and experts mentioned above still apply.

Reexamining relationships is not only occurring on the user side. According to The Independent, AI consciousness researchers say machine intelligence is shaping human intelligence. The view is that not only people use AI, but also the way people think is influenced by AI. The same question appears in the creative field as well. WLNS 6 News reports that Michigan State University's "Reality Check" exhibit features short films created by AI. The university exhibit provides an opportunity for people to look at works created by AI and think about what is real and what is not.

Counseling about worries, playing the role of a lover, discussing consciousness, and creating. In every situation, AI goes beyond being a tool to help people do their jobs and enters into the very relationships between people. If this is the case, the question of whether or not to introduce AI will not only include what role AI will play, but also how people will perceive that role.

Reports of violence and election fraud that shake trust

When a chatbot receives a consultation regarding a murder, which should be prioritized: the user's privacy or reporting violence? This question is now being posed in a concrete form. Al Jazeera highlighted the case of chatbots being used to guide murders, and how AI is caught between protecting privacy and reporting violence. As we saw in the previous section, an increasing number of people are using AI as a partner to confide their concerns. There is still no fixed answer as to how the other person will behave when they hear of the intention of serious harm.

If you prioritize reporting, you have a better chance of saving lives. On the other hand, users may stop talking in the first place because they fear that what they confide in may end up being passed on to the authorities. Prioritizing privacy will maintain trust as a forum for consultation, but it will also allow AI to know about dangerous plans and do nothing. What is being tested here is whether users are given advance notice of what they can say to the AI and who they will tell if the AI learns anything. Organizations considering implementation cannot put AI at the point of contact for employees or customers without confirming this commitment in writing.

The same trust issues are present in elections. According to Westlaw Today, the US government has asked election authorities to be wary of AI-based deception ahead of the midterm elections. The reason why the government went out of its way to call election authorities can be seen as being concerned that the output of AI would appear to be genuine, making it difficult for recipients to distinguish between authenticity and falsehood. However, the title of the article only reveals the existence of the request, and we have not been able to confirm the intended method.

Reporting violence and fraudulent elections are two different things at first glance. However, both questions ask, ``To what extent should people trust the answers given by AI and the information learned by AI, and to what extent should they trust it?'' Can the person using the AI trust how their input is handled? Can the person receiving the AI trust the authenticity of the output? These two directions of trust need to be checked separately before implementation. In the next section, we will consider how to divide the inspection on-site.

Evaluation of diagnostic AI depends on the quality of public data

Imaging AI in the field of cancer has passed the stage where its performance is discussed in a single paper, and has entered the stage where it is being verified all at once. The performance of machine learning and deep learning in predicting distant breast cancer metastases from images was examined in a systematic review and meta-analysis (Langenbeck's Archives of Surgery). The diagnostic accuracy of deep learning for predicting microsatellite instability in gastric cancer is also organized in the same format (medRxiv). A radiomics/clinical model using GCN was investigated in a multicenter study to predict response to primary chemoimmunotherapy for advanced squamous cell carcinoma of the lung (Journal of Thoracic Oncology, P3.062). MetaPaCS, a meta-learning framework, was proposed for pancreatic cancer subtype discrimination (Advanced Intelligent Systems).

Target organs and issues are expanding. Automatic classification of bone, lymph node, and local lesions was investigated in CT images of prostate cancer patients (Journal of imaging informatics in medicine). In glioma, a pathological image-guided multimodal transformer predicts survival benefit from adjuvant chemotherapy (Frontiers in Immunology). The histopathological type of small-diameter renal cell carcinoma was differentiated using deep learning based on contrast-enhanced CT and habitat radiomics (BMC Urology). On the treatment side, the electric field generated by transcranial magnetic stimulation is rapidly predicted using a physical information neural operator and investigated in healthy subjects and tumor patients (Scientific Reports). However, what is listed here is a verification framework and a wide range of targets, and does not guarantee performance that can be used directly in the field.

The reason for the difference in evaluation lies not in the algorithm but in the data used for learning and evaluation. Regarding thyroid AI, it has been pointed out that there are hidden flaws in the publicly available data and there is a risk of overestimating diagnostic performance (Bioengineer.org). Even if public data shows a high value, there is a possibility that the value has been inherited along with the defects. Meanwhile, in agriculture, self-supervised learning AI is said to predict crop yields without labels (Bioengineer.org). This raises the question of whether methods that reduce the burden of labeling can also reduce the effort required to confirm the quality of data. A weekend read (Substack, October 17, 2026) that covers AI and diagnostics alongside Alzheimer's disease and generic drugs also shows that diagnostic support has become a daily topic.

The first thing a person considering introducing the technology should check is not the size of the numbers in the paper, but the source of the data that produced those numbers and whether they have been reproduced at another facility. In both medicine and agriculture, the accuracy of the answers provided by AI is no greater than the accuracy of the input. This confirmation is also a matter of drawing the line between what should be entrusted to AI and where should humans take over.

The sections on employment, funding, counseling on concerns, reporting violence, and evaluation of diagnosis all have the same structure. AI is changing the allocation of personnel, changing the people we consult with, and changing the basis for decisions, but its benefits and limitations have tended to be discussed in different places. Reduction and re-evaluation are progressing at the same time, and even as usage increases, experts cite its limits, and the performance of diagnostics is divided depending on the quality of public data. The spread of use and the momentum of funding are not good reasons to introduce it. The pros and cons can only be determined by comparing the effects achieved and the range not reached in the same situation and using the same criteria. The decision to introduce it can only be made after completing that comparison.

Q1: When incorporating AI into your work, have you ever written down and compared the effects that can be obtained and the areas that cannot be delegated to the same situation? Q2: Who determines the scope of promises made to users regarding how the AI will treat the content revealed and the basis for its judgments, and based on what criteria? Q3: Even if funding and market conditions are favorable, are you able to state in advance the reasons for not implementing the system and the conditions for returning to the system once it has been implemented?

Finally, three questions

- When incorporating AI into your work, have you ever written down and compared the effects that can be obtained and the areas that cannot be delegated to the same situation? Q2: Who determines the scope of promises made to users regarding how the AI will treat the content revealed and the basis for its judgments, and based on what criteria? Q3: Even if funding and market conditions are favorable, are you able to state in advance the reasons for not implementing the system and the conditions for returning to the system once it has been implemented? - Who determines the scope of promises made to users regarding how the AI will treat the content revealed and the basis for its judgments, and based on what criteria? Q3: Even if funding and market conditions are favorable, are you able to state in advance the reasons for not implementing the system and the conditions for returning to the system once it has been implemented? - Even if funding and market conditions are favorable, are you able to state in advance the reasons for not implementing the system and the conditions for reversing the system once it has been implemented?