AI Highlights — the whole picture — 2026-10-27 (Tue) Daily News

Comcast is reportedly moving its counter services to AI agents and reducing its workforce due to a decline in broadband subscriptions. However, Apollo points out that AI is not producing financial returns for U.S. companies. When the decision to reduce the number of people is made before confirming results, work, judgment, words, and trust quietly shift toward AI. The articles lined up in this half-day span different situations, such as administrative reviews, police, elections, education, and specialized research, but they all ask readers to make judgments on the same point. Before deciding whether or not to include AI, it is important to consider who will shoulder the burden and responsibility that will arise after the introduction of AI.

Reductions are taking precedence and there is no return in sight

Comcast is reportedly moving its counter operations to AI agents and reducing its workforce in response to a 144,000 drop in broadband subscriptions (TechSpot). TechStory also reported that the company is expected to make large-scale staff cuts, citing contract reductions and the spread of AI as reasons. There are two reasons for the reduction. The number of subscribers has decreased and the point of contact can be left to AI. However, it cannot be gleaned from these articles how costs and customer service will change after relying on AI. However, Apollo points out that AI is not generating financial returns for U.S. companies (Business Insider). This means that the decision to cut staff may be underway before the results are confirmed.

Reduction is not the only option. JPMorgan's private bank uses AI to streamline advisors' work (Bloomberg.com). This method is used to reduce the amount of work required of advisors while still maintaining their role in dealing with customers. In Singapore, NTUC and KPMG have launched a human resources transformation program (NTUC UPortal) that connects the introduction of AI to improving the quality of employment. What these two cases deal with is not the number of people, but the content of their roles.

If we compare the example of Comcast with the examples of JPMorgan, NTUC, and KPMG, we can see that even though they have introduced the same AI, the assumptions are different. In the former case, due to the deterioration in business performance due to a decrease in contracts, the entire function of the counter will be transferred to AI. The latter is trying to leave work to be done by people and change its quality. However, the JPMorgan article only shows operational efficiency, and the magnitude of the results is not clear from this point. The NTUC and KPMG programs are also at the announcement stage of their efforts.

When readers read introduction articles, they want to see not only what AI will change, but also what they will reduce first and what they will get in return. If Apollo's point is correct, Comcast is not the only example where cuts are made without seeing any return. The question then becomes who will shoulder the burden that remains after the cuts. In the next section, we will look at the process in which tasks and decisions are increasingly carried out by AI, and to whom the burden will shift.

Automated examination and use of police fix bias

Regarding Medicaid's automatic payment review, HomeCare has cited five countermeasures, citing the risks of AI-based review. Leaving the examination to a machine could put the recipient of the payment at a disadvantage. This means that it is necessary to prepare measures in advance to counter this. The BBC reported that researchers have warned that the use of AI by police could deepen discrimination. In Berlin, the question is whether AI-based psychotherapy should be used (The Berliner). What all three cases have in common is that the recipients of AI's decisions and support are those receiving benefits, those subject to police control, and those suffering from mental health problems.

What influences the reader's judgment here is not so much whether the introduction will improve efficiency, but rather who will object when an error or bias occurs and who will pay the cost of correction. HomeCare emphasizes the need for countermeasures because, even if screening automation is talked about as an efficiency for the implementation side, it can end up being a disadvantage to on-site businesses and users. The warnings from researchers introduced by the BBC also indicate that the use of AI in police departments raises concerns about fairness. The question of whether or not a technology can be used for psychotherapy indicates that it is necessary to first decide whether the technology can be used in situations rather than whether it can be used. Those considering the introduction should consider not only the expected effectiveness but also the mechanisms for receiving and verifying complaints.

The same structure appears between countries. According to Al Jazeera, a senior United Nations official has warned that disparities in AI use threaten to widen global inequality. Seeking Alpha reports that in the U.S.-China AI race, each country is being forced to make difficult choices between competing technological spheres. Countries that have the capacity to make full use of technology have a choice, but countries that do not have the capacity to utilize technology themselves may be constrained in terms of opportunities to use it or in determining which technological sphere they belong to. Again, the benefits and burdens of implementation are not necessarily distributed on the same side.

At both the individual and national level, the introduction of AI can make the difference between those who are able to shoulder the burden and those who are forced to shoulder it. In this case, the question that should be asked before deciding whether to include AI is who will be responsible for any disadvantages that may arise. In the next section, we will look at how this responsibility is handled in practice.

Truth and humanity waver

Two days before the election, Prime Minister Netanyahu posted an AI-generated image depicting a nuclear explosion in Tel Aviv, The Jerusalem Post reported. In the run-up to voting, the task of determining whether the image is genuine falls to each recipient. The situation is the same in the educational field. Shinano Mainichi Shimbun Digital reported that misinformation is increasing due to the rapid expansion of AI, and that educational institutions are also struggling with fact-checking. Наша Ніва, with the headline ``AI puts doubt on the author of the world's most expensive painting,'' reports on a case where even the assumption of authorship is being questioned. What they all have in common is that it takes more time and effort to verify things in different settings such as elections, classes, and art. Even if the person implementing AI can reduce the amount of time it takes to create something, the burden of verifying authenticity remains on the recipient or the school. When considering the pros and cons of introducing it, it is also necessary to consider who will pay for the verification.

This fluctuation extends not only to the information side but also to the human side. The Mainichi Shimbun reported that AI bot-like speaking styles are spreading among humans. Not only do we question whether the words on the other side of the screen are machine-like, but the way we ourselves speak is becoming more machine-like. Unisba Media reports that Professor Brian Green spoke about the threat of AI, consciousness and the end of humanity, and the debate over whether machines are conscious has entered the professional arena. On the lifestyle side, The Sun reported that a person who spent his stipend money on a Kim Kardashian-like AI girlfriend said: "She's going to have a baby." There are people who form relationships even if the other person is not human. The book introduces the life and thoughts of Turing, the ``father of artificial intelligence,'' as an eccentric and lonely genius. The question of whether machines can think has long been inseparable from human life.

Checking whether something is real or not, and drawing the line between humans and AI, are now left up to the individual's personal sense of life rather than institutions or companies. The question to be addressed in the next section is to what extent that burden should be borne by people other than the parties concerned.

Verification determines results in specialized fields

In the diagnosis of oral cancer, the role of AI is expanding from image diagnosis to precision treatment, reports Bioengineer.org. In agriculture, AI and robots will improve precision farming and crop monitoring, reports Analytics Insight. In drug discovery, Pathos AI has secured $237 million in Series E funding to expand its agent-based AI drug discovery platform and advance its oncology pipeline, BioSpace reports. What all three topics have in common is that the aim is not to reduce the number of people, but to improve the quality of the work itself: diagnosis, cultivation, and search for therapeutic drugs. However, the huge amount of funding only indicates the level of market expectations, not proof of results. What is different from the situation seen in the previous sections where ``reductions occur first and there is no visible return'' is that the targets for measuring results are specific from the beginning.

It is the verification conducted on the research side that builds up the evidence for these results. The Scientific Reports paper deals with how to suppress batch effects in Raman spectroscopy using deep learning. The content deals squarely with variations due to measurement conditions and timing, as a separate issue from model accuracy. The International Journal of Oncology Research reports the classification of breast cancer subtypes using GEO expression data, along with external validation in an independent cohort. The Journal of Radiation Research and Applications deals with the prediction of radiation therapy effects for gastric cancer using contrast-enhanced CT radiomics, both in terms of model development and validation. The Kashmir Journal of Academic Research and Development has a paper that discusses a pharmaceutical framework that supports lung cancer treatment using machine learning, a joint research project between Pakistan's Faculty of Pharmacy and China Pharmaceutical University. Across multiple institutions and multiple cancer types, there is a common procedure for validating developed models with other data.

This contrast shows what the reader should ask when listening to the introduction. The only way to estimate the potential for cost reduction is to check the accounting records to see if the savings have actually been achieved. AI in specialized fields must first answer questions such as whether it can be applied to independent cohorts, whether the results will hold even if the measurement conditions change, and whether it can be incorporated into on-site procedures. If the system is to be used for diagnosis or treatment decisions, the cost of verifying the answers must be included in the cost of implementation.

In areas where there are people in charge of verification, there will be more materials to judge the value of AI. So where does the responsibility for errors fall in areas where products are used without a person or organization responsible for verification? The next section begins with this question.

The examples seen in each section are from different industries and situations. What remains true, though, is that while there is talk of efficiency and promise with the introduction of AI, the burden that remains afterward is being passed on to someone else. In the case of personnel reductions, decisions are made without confirming the results, and in the case of automated examinations and use by the police, those who are disadvantaged bear the costs of objections and corrections. Verification of authenticity remains with the recipient and the school. What differentiated the results by area of expertise was whether the aim was specific and whether it had been verified. Therefore, the factors that determine whether or not to introduce AI are not only AI performance and market expectations. When an error or bias occurs, it is important to decide in advance who will check it, who will respond to objections, and who will bear the costs.

Q1: When an error or bias is discovered in work using AI, who has the burden of checking and correcting it? Q2: Are the channels for appealing AI judgments and who will bear the costs of corrections determined in writing before implementation? Q3: Have you made a decision to change personnel or authority before confirming the results of the implementation?

Finally, three questions

- When an error or bias is discovered in work using AI, who should be responsible for checking and correcting it? Q2: Are the channels for appealing AI judgments and who will bear the costs of corrections determined in writing before implementation? Q3: Have you made a decision to change personnel or authority before confirming the results of the implementation? - Are the channels for appealing AI decisions and who will bear the costs of remediation established in writing before implementation? Q3: Have you made a decision to change personnel or authority before confirming the results of the implementation? - Are decisions made to move personnel or authority before confirming the results of the implementation?