AI Highlights — the whole picture — 2026-10-20 (Tue)

Amazon's restructuring, layoffs in Canada and the UK, and interpersonal skills training at Goldman Sachs. At the same time, a review of people and organizations against the backdrop of AI is emerging in multiple places. Countries and companies are spending huge sums on GPUs and partnerships, and their use is expanding to suit specific decisions in diagnostics and agriculture. Meanwhile, opposition to data centers, responses to isolation, and conflicts over the delineation of safety regulations have surfaced just as quickly as the investments. What this article focuses on is not how quickly AI can be introduced, but rather how the design is prepared to support people once it is introduced. We will examine the examples in each section to see whether the success or failure of implementation depends on the design.

Between personnel reductions and organizational restructuring

Amazon is restructuring its business amid a combination of job cuts, regulatory compliance and investments in AI growth (HR Katha). Canadian food tech companies are also cutting jobs, raising questions about what AI and automation will bring to Canada (Digital Journal). The UK financial industry is also focused on how AI will affect employment (Bloomberg.com). Although these three cases are in different regions and industries, they have one thing in common: personnel and organizational forms are being reconsidered in the context of AI. When deciding whether to downsize or reorganize your company, you need to separate the jobs that will be replaced by AI from other factors such as regulations and the business environment. As the Amazon case shows, AI is not the only reason for restructuring.

Meanwhile, Goldman Sachs is training more advisors than ever before and thoroughly teaching them "how to build relationships" in the age of AI (Business Insider Japan). At the same time, Amazon, Canadian food tech, and the British financial industry are reducing their workforce, while Goldman Sachs is expanding its workforce through training. The question for readers here is where to draw the line between tasks that should be entrusted to AI and tasks that should be handled by humans. Drawing the line will not only affect the number of people hired, but also what will be included in the training.

Another point of contention is operations after AI is incorporated into operations. An article in the HRTech Series argues that when AI enters the organizational chart, autonomous AI operations will require new operational structures. Even if AI is introduced, if it is not decided who will supervise and who will be responsible, it cannot be said that AI will be able to fill the gap in personnel lost due to reductions. It would be appropriate to consider reduction decisions and operational structure design at the same time, rather than proceeding separately.

Looking at it this way, what determines the speed of reorganization is not only the ability of AI, but also the design of where people will remain and how they will be supported. This difference in design is beginning to manifest itself in workers' anxiety, bias, and isolation.

Investment in AI infrastructure where countries and companies compete

The movement of funds and equipment related to AI infrastructure is spreading to the government, finance, and pharmaceuticals. CNN reported that AI exports are driving growth in China's economy, and sluggish domestic demand is weighing on it. The situation is that while AI will become a source of foreign demand, domestic consumption will remain weak. South Korea's Newspaper News reported that GPUs are at the forefront of competition in sovereign AI (AI infrastructure that a country owns). Whether a country can establish its own infrastructure depends on how much computational resources it can secure. In finance, Santander is teaming up with Nvidia to build AI-powered banking capabilities, according to PYMNTS.com. Investors range from the government to private banks.

In pharmaceuticals, funds were mobilized in the form of partnerships. According to Fierce Biotech, Eli Lilly is teaming up with Hong Kong's Metis to enhance the search for nanoparticles using AI. marketscreener.com reports that METiS TechBio and Eli Lilly have signed an AI-powered drug delivery deal worth up to $1.1 billion, including milestone payments. As the expression "maximum" suggests, this amount is not a lump sum payment. The system is such that payments accumulate each time development reaches a milestone, and payments do not increase unless results are achieved.

The introduction of AI by major pharmaceutical companies is also accelerating on a large scale. Fudu Niu Niu's ``AI Tracking in Biopharma'' summarized that major pharmaceutical companies are rushing to introduce AI on a large scale, and pointed out that a review of corporate value depends on improving the productivity of research and development (R&D). The view is that the evaluation is not based on the scale of the introduction, but on whether productivity has increased.

The deciding factor for readers is not so much the size of the numbers, but what the spending is tied to. China's AI exports are linked to external demand, Sovereign AI's GPUs are linked to securing computing resources, the partnership between Santander and Nvidia is linked to banking business functions, and the Lilly and Metis contracts are linked to development milestones. If you cannot see this connection, you cannot measure the results of your investments. However, whether equipment and funds produce results depends on how the people who use them work. The next thing to look at is the impact on people, which is reflected in the speed of investment.

Expansion of practical use in diagnosis, research, and agriculture

A Boston University research team received a $3.2 million grant for AI ultrasound for neonatal care (The Heights). In the field of diagnostic imaging, verification is also progressing for cancer in adults. For hepatocellular carcinoma, a study was published that quantitatively evaluated abdominal organs using AI and investigated the relationship with the results of immunotherapy in a cohort from multiple institutions (NPJ Precision Oncology). In renal cancer, the performance of a radiomics model to distinguish between Xp11.2/TFE3 translocation renal cell carcinoma and clear cell carcinoma was reported (BMC cancer). What all three have in common is that they are specifically targeted to ``ultrasound of newborns,'' ``organ evaluation related to the effectiveness of immunotherapy,'' and ``discrimination between two similar types of kidney cancer.'' The value of AI is beginning to be measured not by its general-purpose intelligence, but by how well it matches specific judgments in the field.

The same trend exists in basic research and agriculture. A paper in Nature Biomedical Engineering describes an interpretable AI model that deciphers the rules of how cells flock to specific destinations. The design allows people to check not only the results but also the evidence, making it easy for researchers to use when reconsidering their hypotheses. In agriculture, it has been reported that an AI predictive model has predicted rice yield with high accuracy (Bioengineer.org). Yield prediction directly leads to planting and shipping plans. Whether in medicine, basic research, or agriculture, what the adopter needs to confirm is not only whether the predictions are correct, but also whether they can trace why they made the decisions they made.

Another thing that cannot be overlooked is fairness. A scoping review that sorted out bias in large-scale AI models for medical care was published in Nature. This shows that in addition to the high performance, it is necessary to systematically check where bias may be introduced. Diagnostic support is used by different patients and groups. Even if accuracy is achieved in one group, it may not be the same in another group. Conditions for trust include not only accuracy numbers but also the fairness of who the verification was performed against.

As results in specialized fields become more specific, the question of whether or not to introduce the technology shifts from ``can it be used?'' to ``Is it safe for whom and how can it be explained?'' This question is not limited to medicine and agriculture, but also connects to issues of employment and isolation, which will be discussed in the next section, as a design issue that supports working people and organizations.

The side of life that deals with anxiety and isolation

The movement against building data centers is driven by concerns about AI itself, according to the New England News Collaborative. The reasons for opposition are not limited to problems with facilities such as land and electricity. Anxiety has also reached the church. In Ireland, a priest asked an AI expert to speak to Mass attendees, the Irish Independent reported. A place of faith provides an opportunity to learn and talk about AI. There is also movement in response to families and isolation. Figurobot has released an 11-inch AI companion robot “FR-NIA” (StreetInsider). In an era when 22,222 people died in isolation in Japan, LINE's AI monitoring service ``Fureri'' was launched to connect families (Nico Nico News). The protests, lectures, robots, and monitoring are all attempts to find out how far society should go in dealing with AI.

In places of creation and learning, acceptance takes a concrete form. AI art museum Dataland has installed Riedel's wireless intercom (Inavate). It can be seen that the exhibition that uses AI is now in the stage of preparing on-site equipment to welcome visitors and operate it. In education, the College of DuPage brought OpenAI Academy to Illinois and held its first AI summit (Crain's Chicago Business). Museums are places where you can come into contact with works of art, and universities are places where you can re-learn. In both cases, we chose not to push AI away, but to provide a place where people can actually touch it and check it out.

Reactions to anxiety and movements toward acceptance proceed side by side within the same society. Readers should not only look at the performance of the technology when deciding whether to introduce it. The key is whether a place where people can talk about their concerns, a system to support their isolation, and an entry point for learning are provided along with the introduction. The next question is whether the design that supports people can keep up with the speed at which AI is spreading.

Tug of war over safety regulations

It has been reported that there is a backlash against the $1 billion AI safety bill supported by Anthropic (shattered.io). The article's headline suggests that AI development companies are now on the side of demanding safety regulations themselves, and that there is some disagreement over where the line should be drawn. The focus is reportedly on the $1 billion dollar threshold. The question of what scale of development and business should be subject to regulation is causing a backlash. This article has only been able to confirm the scope of the headline, and will not discuss the provisions of the bill or the specific arguments of those opposing it.

This standard is attracting attention because it overlaps with the investment and implementation trends we have seen in the previous section. The government and companies are investing huge sums of money into infrastructure development, and its use in the field is expanding to include diagnostics and agriculture. In this situation, if the amount is used as the standard, regulations would apply to those making large-scale investments, and would not apply to smaller-scale investments. Depending on how standards are set, there are companies that are subject to regulations and companies that are not. The fact that there is such a backlash can be interpreted as a sign that where the line is drawn has a direct impact on each company's business.

What readers can use to make decisions is not so much whether there are regulations, but rather where the lines are drawn and who is asking for them. Even if companies that place importance on safety support regulations, if the regulations impose a burden on other businesses, the pros and cons will differ depending on the company's position. It is not possible to evaluate the pros and cons of monetary standards until the details of the bill are available. For those who are proceeding with the introduction, it is worth checking early on which line the company's size and purpose fall within.

If the introduction of the system continues without clear regulations, it will become unclear who will be responsible for checking the impact on workers and users. Drawing the line for safety is not only a question of technology, but also a question of how far to incorporate systems that support people into the system.

The movements we have seen this time are in different fields: investment, restructuring, practical use, anxiety, and regulation, but they have in common that the introduction of AI is simultaneously having an impact on people. Funding for infrastructure has spread to countries and companies, and targeted use has progressed in diagnostics and agriculture. Meanwhile, backlash over employment reviews, opposition to data centres, responses to isolation and regulatory standards emerged alongside. The scale of investment and technology alone cannot determine whether the implementation was successful or not. Identifying the reason for the reorganization, choosing a method of use that matches the judgment of the site, providing a point of contact for responding to concerns and isolation, and clarifying the lines of regulation. Introductions that involve these things will become established in a way that will be accepted by people. The difference between success or failure is whether a design that supports people is prepared at the time of introduction.

Q1: On what basis do you differentiate between parts of your work that should be left to AI and parts that need to be judged by humans? Q2: At what scale and use have you decided to apply the standards for how to use AI, and who can explain the line? Q3: When reviewing your personnel and organization, how do you distinguish between AI-related reasons and other reasons?

Finally, three questions

- On what basis do you decide which parts of your work should be left to AI and which parts should be judged by humans? Q2: At what scale and use have you decided to apply the standards for how to use AI, and who can explain the line? Q3: When reviewing your personnel and organization, how do you distinguish between AI-related reasons and other reasons? - At what scale and use have you decided to apply the standards for how to use AI, and who can explain the line? Q3: When reviewing your personnel and organization, how do you distinguish between AI-related reasons and other reasons? - When reviewing your personnel and organization, how do you distinguish between AI-related reasons and other reasons?

It was reported that Oracle will reduce its workforce by 13% in fiscal 2026 by increasing efficiency through AI, and the stock price hit its lowest level in about two months. Data centers and AI emerged as issues of contention in Virginia's midterm elections. In the former case, the burden of reductions is felt on employment, and in the latter case, the burden of the reduction appears on the local area as a place to store equipment. The introduction of AI is often talked about in terms of the magnitude of its results. However, when we look at the trends discussed this time, it becomes clear that the benefits of efficiency and growth and the associated burdens are not reaching the same people. The burden will vary depending on employment, prices, region, and discipline. The criterion for measuring the pros and cons of introducing a system is not only what is gained, but also who bears the burden. In this text, we will examine the destination of this burden section by section.

The fruits of efficiency and the burden of personnel and prices

Oracle's stock price hit its lowest level in about two months, and it was reported that the company will cut its workforce by 13% in fiscal 2026 by increasing efficiency through AI (Stocktwits). The results of efficiency improvements are first seen in the numbers of personnel. AI-related layoffs are tracked by Layoffs.fyi, where you can see a list of companies and how widespread the cuts are (Layoffs.fyi). Employment is not the only place where the burden will shift. The Fed recognizes that the AI boom will increase inflationary pressures, and it has been pointed out that other sectors and entities may bear the costs (Reuters). Efficiency improvements in one company will result in a burden on prices and employment elsewhere. When evaluating implementation, you need to look not just at internal cost savings, but also at what's being transferred externally.

On the other hand, people's roles have not been uniformly reduced. Fifth Third Bank's CEO says he doesn't see AI agent Muse as a threat (Banking Dive). In airline flight management, AI co-pilots integrate pre-arrival information and prepare for return (Bioengineer.org). Here, AI is responsible for gathering information and making preliminary preparations, and the final decision can be interpreted as leaving a lasting impression on humans. The advertising industry is also discussing why soft skills will become important in the age of AI (Campaign Middle East). The idea is that the more routine tasks taken on by AI, the more work that will be required of humans to rely on interpersonal coordination and judgment.

Oracle's layoffs and the cases of flight management and banking show that the same AI implementation can have different results. The difference lies in whether to use the free time and manpower to reduce time or to redefine work that only humans can do. However, redefinition is only possible if the people and regions available for the role are secured. The next question is the distribution of who bears the burden and where.

The power of judgment and discovery that is being proven in specialized fields

In the development of antiviral therapy, researchers reportedly used AI to identify early-stage inhibitors (News-Medical). AI was used in the process of determining which molecules could become inhibitors from a large number of candidates, and this is the result of this process. On the diagnostic side, AI-based analysis of pathology slides is said to be promising for lung cancer biomarker testing (Clinical Lab Products). These two cases are not about tidying up the text or shortening the work. This is a process of judgment and discovery, such as ``Is this molecule effective?'' and ``What findings are on this slide?'' It can be seen that the yardstick for measuring the value of AI is expanding from the reduction of work time to the quality of judgments.

A similar story can be found in agriculture. EIN News published a sectoral analysis and strategic outlook on the AI market for crop monitoring. The work of looking at field conditions and finding abnormalities is similar in structure to the work of picking up findings on slides. In both medicine and agriculture, AI is responsible for seeing, finding, and sorting, and has reached the point where it has become the subject of market analysis. The field of creative writing is no exception. Cristina Ardelean's AI art was featured on freeyork as a work that combines shadows and surrealism. Here, AI is not used to replace human ideas, but rather as an expressive tool for artists to give form to their themes.

What all three areas have in common is that the output of AI is not the final conclusion, but rather serves as a starting point for humans to confirm. Whether it's inhibitor candidate, biomarker determination, or field abnormality detection, we move on to the next step only after someone confirms it. For readers who are deciding whether to introduce a product, high accuracy is only a prerequisite for entry. Only by looking at who will check the output and who will be responsible in the event of an error, can we begin to understand the details of the implementation. The more decisions and discoveries are made by machines, the more new tasks of confirmation and responsibility will be born, and the next question is where these tasks will be concentrated.

Drawing the line of governance in elections, military arms, and data protection

Data centers and AI have emerged as issues of contention in Virginia's midterm elections (Virginia Mercury). This means that the question of where in their area they would like to locate equipment that uses electricity and land has begun to be sent directly to voters in the form of a ballot. The imbalance in the burden that we have seen in the previous sections is manifested not only in employment and prices, but also in the physical location of equipment storage. For companies deciding to introduce the system, deciding who will agree on the location is becoming a prerequisite, along with estimating the results.

Equipment isn't the only thing to draw the line on. An editorial in The Washington Post argues that the US and China need an arms control agreement on AI. The editorial is just one medium's assertion, and does not indicate the content of the agreement or the outlook for its realization. However, it can be seen that we have entered the stage where AI safety is no longer limited to corporate self-discipline or one country's regulations, but is discussed as an agreement between countries. In Mexico, Mexico Business News argues that privacy issues could undermine AI strategies. The view is that unless the public trusts in the handling of data, the country's plans to advance AI will be undermined. Equipment in Virginia, military armaments in the US and China, and data protection in Mexico are all different in terms of their scope, but all are similar in that if the side that is rushing to introduce them doesn't draw the line first, the outsiders will draw the line later.

In order to draw a line, we need verification to provide a basis for drawing it. According to citybiz, Zedge's DataSeeds.AI has completed its first AI model evaluation project with Bria. This shows that the movement to contract out specialized evaluations is beginning to take shape as a business. Additionally, Inside UNC Charlotte reports that UNC Charlotte shared AI literacy for STEM education with Malmö University. The infrastructure is being built on both the side of those who evaluate the model from the outside and the side of the users who interpret the results.

Looking at it this way, the same questions are repeated in different situations: equipment storage, armaments, personal data, model evaluation, and education. The question is who will confirm whether or not it can be introduced, and at what level? The next point to check is the extent to which systems are actually in place to support this decision, which is not determined solely by the size of the results.

Relationship between people and AI affected by spouse, will, and rights

Should we use AI as a tool, or should we face it as an opponent? This distinction is becoming blurred both in everyday usage and in the text of the system. Spain's Universidad de Córdoba (UCO) provides an explanation of what "AI girlfriends" and "AI partners" are, who respond via voice or chat. The fact that the university is issuing such an explanation shows that AI companions are not just a topic for a few users, but are something that should be properly explained. (Universidad de Córdoba (UCO))

The treatment of relationships goes beyond just relationships during life. La Voce di New York looks at cases where AI companions appear in wills and asks what happens to their presence after death. (La Voce di New York) A will is a document that specifies to whom property and important things are entrusted. When AI enters the picture, the people and businesses left behind will have to decide whether to dispose of AI as property or take it over as a relationship that should continue. Those considering the introduction will also need to think in advance about how it will end when users build deep relationships.

The same question arises on the institutional side. Forbes reported that AI rights were included in Anthropic's terms of service, prompting pushback from Microsoft. (Forbes) The two major companies are divided on whether or not to include a description granting rights to AI. Terms and conditions are the basis of contracts with users, so how AI is positioned there will also affect the design of the service and where responsibility lies. What this conflict shows is that the treatment of AI has passed the stage where it is determined solely by the specifications of the technology, and is now being expressed as a value judgment for each company.

Meer (English edition) captures this trend with the title ``The human question behind AI.'' (Meer | English edition) When it comes to spouses, wills, and contracts, the issue is not so much the nature of the AI itself, but rather the position that people give to the AI. If the use of technology continues without determining its role, the burden of ambiguity will likely fall on those who are forced to make decisions in individual situations first, such as users who have been deeply involved and the families left behind. How the relationship between humans and AI is defined is also connected to the distributional issue of who will bear the costs.

Although the movements seen in each section are in different fields, they are connected by the same points. Reductions in personnel, the impact on prices, the location of data centers, arms control between nations, and the handling of wills and rights regarding AI companions all emerged as questions about who would benefit from the introduction and who would bear the burden. On the other hand, AI has also been shown to improve the quality of decisions and discoveries in drug discovery, diagnostics, and agricultural monitoring. There is a balance between ensuring that results are achieved and that the burden is uneven. That is why the success or failure of implementation is not determined solely by the size of the results. We are now at the stage where the legitimacy of continuing to use AI will be determined by whether we can show efficiency and growth figures, as well as the destination of the burden and the consent and preparedness of those who will take it on.

Q1: When you incorporate AI into your work, apart from the work time saved, are you able to understand whose work and burden has increased? Q2: Do the standards for checking the use of AI include not only the quality of results but also the impact on those receiving the burden? Q3: Do the indicators used to evaluate the effectiveness of AI implementation include the burden of shifting jobs, regions, business partners, etc.?

Finally, three questions

- When you incorporate AI into your work, apart from the work time saved, are you able to understand whose work and burden has increased? Q2: Do the standards for checking the use of AI include not only the quality of results but also the impact on those receiving the burden? Q3: Do the indicators used to evaluate the effectiveness of AI implementation include the burden of shifting jobs, regions, business partners, etc.? - Do the standards for checking the use of AI include not only the quality of results but also the impact on those receiving the burden? Q3: Do the indicators used to evaluate the effectiveness of AI implementation include the burden of shifting jobs, regions, business partners, etc.? - Do the indicators used to evaluate the effectiveness of AI implementation include the burden of shifting jobs, regions, business partners, etc.?