AI Highlights — the whole picture — 2026-10-12 (Mon) Evening News

Nokia has announced 253 job cuts as it expands its AI strategy, and HSBC is using AI-related job cuts as the basis for its undervalued stock price argument. Companies have begun to incorporate AI-driven efficiencies directly into headcount and revenue calculations. However, as I read through this half-day's material, what draws attention is not so much the reduction figures themselves, but the lines that lie before them. An AI that makes phone calls to elderly people every day, an AI that exceeds the accuracy of doctors by reading videos of breast ultrasounds, and an AI that says "I don't know" on its own. How much of human work, intimate relationships, diagnosis, and judgment will be entrusted to AI, and who will measure its suitability and by what criteria? Each of the following passages shows, from a different perspective, that there is no fixed way to measure it.

An abacus of growth expectations and personnel reductions

Nokia plans to cut 253 jobs as it expands its AI strategy, and the stock price is valued at 9.33 euros (ad-hoc-news.de). Regarding HSBC, there is talk that the company's stock price may be 39% undervalued due to AI-driven workforce reductions (Yahoo Finance). What they all have in common is that reductions in personnel are listed directly as a material for constructing evaluations for investors. The results of introducing AI are beginning to be seen primarily in the numbers of reductions in labor costs, rather than the performance of the technology.

On the other hand, the scale of reductions varies. Nokia's 253 employees is a small number in isolation, but when combined with the company's expansion of its AI strategy, it comes into context when discussing stock prices. At HSBC, the reduction itself is the basis for the argument that it is undervalued. In both cases, the fact that the company will reduce headcount is treated as an assumption supporting the earnings outlook. When readers read a company's announcement, they need to look not at the number of people being cut, but how it relates to their evaluations.

Opposite pressures have also been reported from the profit side. The Korea Daily reported that smart agents will put 100 trillion won in bank profits at risk. The view is that AI will not only make banks' operations more efficient, but will also take away the very profits that banks have earned. While the article on HSBC uses cost reduction as a factor in its evaluation, this article is wary of a decline in profits. Even within the same industry, AI is sometimes seen as a tool to lower costs, and sometimes as a threat to revenue sources.

Political language is also leaning towards growth. The Guardian argued that Trump is dreaming of AI-driven growth, a return to Reagan-era economic promises. Furthermore, Bloomberg.com reports that AI is rewriting the picture of the US-China conflict. When juxtaposed with discussions of corporate financial results and stock prices, it becomes clear that AI is being talked about not only in terms of improving the efficiency of individual companies, but also in terms of both the promise of growth and competition between nations. However, the materials cited here do not indicate the extent to which growth will be achieved, but rather indicate that a framework for talking about growth has already been prepared.

In this way, while AI is talked about as a growth story, it is actually also emerging as a pressure to rearrange the allocation of costs and revenues for companies. The next question is who will measure how that pressure is applied and by what standards.

Net increase in employment and people left behind

McKinsey has expressed the view that AI will create more jobs than will be lost (Fortune). However, the same analysis also states that 11 million U.S. workers will need help with the transition (Fortune). The forecast for total employment growth and the forecast that 11 million people will be unable to move on their own are juxtaposed in the same report. Net gains do not tell us who is losing jobs and who is gaining them. Readers of companies' human resources plans need to look not just at the overall income and expenditure, but also whether the plan is designed to actually reach these 11 million people.

Movements over where to place that burden have already begun elsewhere. KQED reported that AI and politics are a catalyst for change for tech workers, who have long distanced themselves from unions. The question of whether the cost of the transition will be borne by employers or individual workers is now being asked by those who have avoided organizing until now.

On the education side, the same question is being asked for younger people. World Economic Forum experts said child safety should be prioritized in AI education in India (Rediff). In the debate about speeding up the acquisition of skills, the order of putting safety first introduces a different standard from the logic of corporate efficiency in terms of who decides the speed of introduction.

Another question is what remains and what has value. Contemporary artist Olga Hilgers talked about how AI becomes more valuable as it gets smarter (Democrat and Chronicle). This is a question outside of employment statistics, but when considering what kind of jobs to provide for the 11 million people who need support, it becomes a prerequisite for how to define the parts that only people can do.

Calculations about whether jobs will increase or decrease will not be a practical basis for judgment unless we first answer the question of who will shoulder the burden of the transition. This question leads to a debate about how much to entrust to AI in specific areas such as people, intimacy, and diagnosis.

AI that enters into intimacy and the line drawn

To coincide with World Mental Health Day, a report focused on the risks of AI, social media, and loneliness (The Media Line). Around the same time, Kindable announced Loved, an AI care app that makes daily phone calls to seniors (Yahoo Finance). As a response to loneliness, a product designed to have an AI at the telephone has entered the market. Meanwhile, The Morning has a story titled ``The love of my life is AI.'' The function of talking to someone every day and the relationship of calling AI the love of one's life are very different in what they are looking for even though they are the same "talking partner." The former can be evaluated as a tool to compensate for the frequency of monitoring, but the latter asks what the relationship itself means to the person in question. What readers should look at is what users expect from AI, and how those expectations shift as they use it.

In response to this expansion of intimacy, Gordon-Levitt warned that the threat posed by AI lies in its ability to take advantage of human intimacy (Hollywood Reporter Japan). From a different perspective, the place where products that fill loneliness come in can also be used as an entry point for exploitation. Fast Company asks, "Can AI be moral without being sentient?" If you are entrusting your intimacy to someone who has no sense of feeling, the first thing that comes to mind is how to demand morality and responsibility from that other person. On the legal side, Rep. Ted Lieu opposed giving legal rights to AI and expressed doubts about prosecuting AI companies (Benzinga). If we deny rights to AI and are cautious about prosecuting companies, it is difficult to see from these statements alone where responsibility should be placed when damage occurs due to dependence or exploitation.

It remains to be determined whether product designers, users, or the law will draw the line between the power to fill loneliness and the dangers of addiction. If the subject of entrustment after intimacy expands to heavy areas such as diagnosis and judgment, the same question appears in a more severe form.

Towards AI that can reduce overconfidence and verify verification

Bioengineer.org reports that an AI that reads the entire video outperformed doctors in diagnostic accuracy during breast ultrasound examinations. Ultrasound examinations involve not only looking at images one by one, but also reading successive images while moving a probe. The report that AI exceeded the accuracy of doctors for the entire video indicates that AI may be used more widely in the field of diagnosis. However, what this headline indicates is a comparison of accuracy, and the question of where the AI makes mistakes and how to distinguish between them remains another question.

To address this question, Bioengineer.org has also introduced a method to make AI say "I don't know." It uses neutrosophic geometry to reduce overconfidence in AI. If an AI with higher accuracy responds with more confidence even when it misses, the user will not be able to distinguish between a hit and a miss. The system of self-indicating that it cannot answer the question replaces the choice between accepting AI's output at face value or doubting it completely with the decision of ``To what extent can I believe this answer?''

The movement to measure reliability next to performance can also be seen on the industry side. The Tech Buzz reports that third-party AI safety reviewers are increasing their presence in the industry. In addition to the numbers provided by the developer, companies and hospitals considering introducing the system will have a basis for comparison by adding the measuring stick from an external evaluation organization. On the user side, The Fulcrum introduces an attempt by high school researchers to teach AI how to read the news. It is also beginning to be treated as part of verification that the people using AI have the ability to judge the accuracy of information themselves.

Accuracy that exceeds that of doctors, design that allows the user to say "I don't understand," external evaluation, and training on the part of the reader are all attempts to format the answer into a form that can be verified rather than accepting the AI's answer as is. While performance numbers are reported first, the future points of contention will be the system of how those numbers are verified and who is responsible for measuring reliability.

Numbers of layoffs, the prospect of net job gains, phone calls, love of life relationships, and diagnostic accuracy that outperforms doctors are all evidence of AI's achievements. However, what each section had in common was that, rather than the size of the results themselves, it was not clear who was measuring them and by what standards. The reduction in the number of people is read as a premise for profit forecasts, and the total number of net additions does not tell us whether or not support will be provided to the 11 million people. Only when a system is in place to measure it will it be possible to determine in which direction users' expectations lean and how to identify errors in diagnosis. The extent to which AI is entrusted will be determined not by its high performance, but by whether it can show how to measure.

Q1: For tasks that use AI, to what extent do you have a system in place that allows you to notice when the output is out of line? Q2: Apart from the size of the results, who sets the criteria for determining whether the use of AI is appropriate or not? Q3: When factoring the effects of AI into human resources and revenue plans, to what extent can you explain how to measure those assumptions and where to place the burden on those affected?

Finally, three questions

- To what extent do we have a system in place for our work to be able to notice when the output is out of whack in the work that uses AI? Q2: Apart from the size of the results, who sets the criteria for determining whether the use of AI is appropriate or not? Q3: When factoring the effects of AI into human resources and revenue plans, to what extent can you explain how to measure those assumptions and where to place the burden on those affected? - Apart from the size of the results, who is setting the criteria for determining whether the use of AI is appropriate or not? Q3: When factoring the effects of AI into human resources and revenue plans, to what extent can you explain how to measure those assumptions and where to place the burden on those affected? - When factoring the effects of AI into workforce and revenue plans, to what extent can we explain how to measure those assumptions and where to place the burden on those affected?