AI Highlights — the whole picture — 2026-10-22 (Thu)

OpenAI's acquisition of Orchestra Bio, Astellas' $400 million deal with AI Proteins, Qualcomm's layoffs and pivot to AI data centers. A series of reports over the past half day indicate that AI is deciding where funds go for research, equipment, and staffing. Meanwhile, in an article listing Americans' fears, anxiety about AI ranked in the top 10, and Duke University is progressing with the clinical implementation of AI for dementia. This sequence cannot be explained solely by measuring how accurate the technology has become. The success or failure of AI implementation now depends on how organizations position AI and how people perceive it. In each of the following sections, we will examine these changes based on individual reports.

Drug discovery has entered the stage of competing for leadership through acquisitions and alliances

OpenAI has acquired AI drug discovery startup Orchestra Bio. CHOSUNBIZ reported on the acquisition as a move to "accelerate AI-driven drug discovery," and another article headlined accelerating new drug development. Business+IT reports that the acquisition comes amid increasing competition from Anthropic and Google. AI companies chose to acquire startups rather than conduct their own R&D as a way to enter drug discovery. When evaluating AI for drug discovery, it is no longer sufficient to simply compare model performance. Which companies have a strong drug discovery team and technology on hand will also factor into the decision.

On the pharmaceutical company side, there is also progress in retaining customers through alliances. Astellas has signed a $400 million computational drug discovery agreement with AI Proteins, according to Fierce Biotech. In a Business Wire announcement from AI Proteins, the two companies signed a research collaboration and licensing agreement. The target is the search and development of new therapeutic drugs based on miniproteins, and the total cost is said to be over $400 million. While acquisitions are primarily moves by AI companies to enter drug discovery, this agreement is a move by a major pharmaceutical company to partner with a company specializing in AI. Since the amount shown is the total amount, readers need to check the terms of the contract in the announcement to determine the actual burden and reward. Still, we can see that funding is shifting from individual experiments to long-term collaborations with partners that have specific technology infrastructure.

The form of contracts themselves is also changing. GeneOnline reported that hourly billing is coming to an end as AI revolutionizes the $400 billion pharmaceutical backend business. If the system of accumulating work hours and billing becomes untenable, contractors will have to demonstrate their value through results and the mechanisms they provide. Ordering parties will also increasingly choose subcontractors based on how they incorporate AI, rather than on the amount of time they can put in the work. The media portrays this change as a structural shift for the entire industry.

The three moves - acquisitions, partnerships, and outsourced contracts - show that the adoption of AI is not just about technology, but extends to the allocation of funds and the design of transactions. The next question is how the people in the field will perceive and utilize the system that has been put in place.

Capital investment and personnel reallocation will change the shape of the organization

Qualcomm is reportedly cutting staff and shifting the focus of its business to AI data centers in line with its focus on results (news.lavx.hu). From the materials available here, it is not possible to confirm the scale of the reductions or the types of jobs targeted. What is certain is that the decision to reduce personnel and the decision to shift focus to AI equipment are discussed in the same article. Around the same time, Jamie Dimon talked about the economy, "sticky" inflation, and AI spending (WSJ). The fact that AI spending is being talked about alongside inflation indicates that investment in AI is no longer limited to the technology choices of individual companies, but is beginning to be measured in terms of costs and economic outlook. When readers consider their own AI investments, they will not only look at the merits of the technology, but also how the level of spending will look in the external environment.

Distribution on the human side is also changing. isolved places AI agents alongside human resources on the organization chart to promote workforce capital management (Yahoo Finance). AIが担う仕事を人の仕事と同じ図の上で管理するということで、導入の単位がツールから組織上の役割に変わる。 Changes are also occurring in the recruitment process. The CMIC Group, which is close to the pharmaceutical industry, has abolished the Entry Sheet (ES) for new graduate recruitment and introduced AI dialogue. Students are asked to ``speak in their own words'' (Nikkan Yakugyo).書類を読む手間を減らすことが目的なのか、応募者の人柄を見極める手段を変えることが目的なのかは、材料からは読み取れない。 However, the point remains that the introduction of AI is changing the very process of selection.

The four movements so far differ in scale and industry. Still, they are consistent in that funding for equipment and allocation of personnel and roles are being reshuffled at the same time. Those evaluating the introduction of AI will overlook the overall movement if they treat the amount of capital investment, increase or decrease in personnel, positioning in the organizational chart, and how to use it in recruitment as separate issues. Where funds go and where people are placed need to be seen as two sides of the same coin. In that case, the success or failure of implementation will depend on how people within the organization perceive AI. What we will see next is a scene in which AI is penetrating the way we work and even the realm of our minds.

Images and analytical models take root in the field of diagnosis and prediction

AI that uses images as input is expanding from its use in assisting diagnosis to its use in estimating risks before surgery. Medical Economics' morning medical news reported that AI has improved the accuracy of echocardiographic diagnosis. The same issue also featured the conviction of a pharmaceutical executive for black-marketing HIV drugs. International urology and nephrology has published a study in which a multicenter model for identifying non-functioning kidneys using plain CT radiomics was developed and externally verified. Breast cancer research: BCR research uses an ensemble model based on DCE-MRI to preoperatively predict positive surgical margins in breast-conserving surgery. This is a combination of multicenter retrospective and prospective studies.

Examples of preoperative prediction can be found in other organs as well. Frontiers in Artificial Intelligence reported a machine learning model that extracts and integrates two-dimensional and three-dimensional intratumoral heterogeneity features from multi-sequence MRI to predict perineural invasion in pancreatic ductal cancer with high accuracy. The Asian journal of andrology published a study using T2-weighted MRI radiomics to predict preoperative apical margin risk for prostate cancer. The target is narrowed down to cases where the tumor is visualized at the top. Abdominal Radiology has a model that uses machine learning and CT radiomics to identify esophageal varices, which have a high risk of bleeding in patients with hepatocellular carcinoma.

When lined up, there are three things in common. The input is an image taken during daily medical treatment. The targets to be predicted are events related to the selection of surgical methods and treatment strategies, such as margins, infiltration, and varicose veins. Evaluations are not limited to internal verification at a single facility, but also include multi-center development, external verification, and prospective studies. There are also designs that first determine the conditions under which the technology can be used, such as prostate research that focuses on cases where the tumor is visible at the top. Readers should look at more than just the model's accuracy numbers. It is necessary to confirm which facilities, under what imaging conditions, and to which patient groups it can be applied.

These studies indicate that validation designs are beginning to question whether image AI can be used clinically. However, whether a model that passes verification continues to be used in the field is determined separately from its accuracy. It all depends on how the doctors and surgeons reading the images perceive it and how they incorporate it into the flow of medical treatment.

AI that approaches daily life and emotions causes anxiety and expectations at the same time

In an article listing Americans' fears, anxiety about AI ranked in the top 10 (La Voce di New York). At the same time, in the medical field, it was reported that Duke University was promoting efforts to promote the clinical introduction of AI for dementia (News-Medical). While AI is talked about as something to be feared, it is also entering the delicate field of dementia care. These two things are happening in parallel in the same society. Therefore, it is difficult to judge how AI will be introduced just by looking at the level of anxiety about AI. We need to see whether people with concerns will accept AI in situations related to caring for themselves and their families.

Applications that are close to everyday life are not limited to care settings. An announcement from Spain's University of Córdoba (UCO) discussed why the free AI girlfriend is so popular, along with the real thoughts of its users. In South Korea, it was reported that Apple's Siri AI will support Korean and will also refer to travel memories (조선일보). The former positions AI as a partner for emotional exchange, and the latter as an assistant who uses personal records. The more you use both, the more you record your private feelings and actions. Free entry and memory references create convenience. At the same time, it leaves the user with questions about what can be remembered and how it can be used.

Technology to read emotions itself is also beginning to be incorporated into products. It has been reported that Emotional Technologies' emotional analysis AI will be installed in Robutler (Dellows News). If the conversation partner or assistant can analyze the user's emotions, it becomes easier to tailor responses to the individual's situation. On the other hand, some people may find it a burden to be read. All of the articles listed here are limited to the information contained in the headline. Accuracy and safety records cannot be confirmed in individual reports. Therefore, rather than focusing on the number of functions, those implementing the system need to first show how much they want to inform users about and how much they can choose.

Whether it's dementia care, companions, voice assistants, or emotional analysis, user trust is more likely to be affected by the way information is handled and explained than by performance numbers. As technology comes closer to everyday life, the success or failure of its introduction depends on how organizations and people perceive it. This point is also common when looking at changes in workplaces and work styles.

The base of implementation extends to crops, learning, and creation

It was reported that efforts are being made to use AI to develop potatoes and rice that are resistant to climate change. That's the EU-funded BIOCRAFT, SeedWorld reports. The project targets two staple food crops and is funded by the European Union's public framework. Within the scope of the news, it is not possible to confirm the development method or numerical results. Still, it can be seen that AI is being introduced not only for summarizing materials and automating tasks, but also for cultivating crop varieties, which take a long time to grow. What readers should look at also changes. In addition to the performance of the model, clues for evaluation include under what conditions and for how long public funds are being used for variety development.

The wave of introduction has also reached individual learning. Mashable featured a $35 set of courses that will help you hone your AI skills and even monetize them. The scale of an EU-funded research project and a $35 course are completely different. The fact that these articles are placed in the same group shows that the use of AI is simultaneously expanding from organizational investment to individual learning. What readers want to see here is not the low price. The point is that monetization is included in the purpose of the course. Being able to use AI is now being talked about as part of one's job skills, and is being linked directly to an individual's earning power. Both those taking the course and those hiring them need to separate the content of the skills from the results of actual work.

There are also movements in the cultural field. According to AzerNews, the National Museum of Art has participated in an international dialogue on AI and culture. A museum is an institution that collects, preserves, and displays works of art. The reason why this is happening in a forum for dialogue that transcends national borders is probably because AI is seen as an issue that cultural leaders should address in their own words, rather than an issue that is limited to the technical sector. All we know from the reports is the fact of participation, and it is not possible to confirm whether an agreement or specific policy has been announced. As more institutions sit at the table of dialogue, the treatment of AI in creation and exhibition will be questioned not only from a technological perspective but also from a cultural perspective.

The three examples, crops, courses, and museums, are different in field and scale. However, the common thread is that decisions about whether to include AI now depend not only on the performance of the technology, but also on who is funding it, who is learning it, and who is accepting it. As the scope of implementation expands, the factors that determine success or failure shift from technology to people and organizations.

Although the cases covered in each section are in different areas, they all point in the same direction. Acquisitions and alliances were determining the position of AI in drug discovery, and investment and reassignment in terms of equipment and personnel. For image analysis, multi-facility external verification was the key to evaluation, and for crops, learning, and creation, the sources and ways of funding were used as clues for evaluation. In areas close to daily life and emotions, anxiety and expectations were talked about side by side in the same society. In all cases, model accuracy is a prerequisite, not a deciding factor. The success or failure of implementation will depend on who provides the funds, who takes responsibility, and how users perceive it. Evaluation of AI is shifting its focus from measuring the performance of the technology to measuring how it is perceived by organizations and people.

Q1: When incorporating AI into your work, in addition to the accuracy numbers, at what point and how do you check whether your colleagues and users will accept it? Q2: As the person responsible for the use of AI, have you decided before the introduction who will check the scope and to what extent and who will explain when a problem occurs? Q3: When deciding whether to invest in AI or allocate personnel, what criteria do you use to factor in the external environment and how users perceive it, in addition to the performance of the technology?

Finally, three questions

- When incorporating AI into your work, beyond the accuracy numbers, how and when do you know if your colleagues and users will accept it? Q2: As the person responsible for the use of AI, have you decided before the introduction who will check the scope and to what extent and who will explain when a problem occurs? Q3: When deciding whether to invest in AI or allocate personnel, what criteria do you use to factor in the external environment and how users perceive it, in addition to the performance of the technology? - As the person responsible for the use of AI, have you decided before implementation who will check the scope and to what extent, and who will explain when a problem occurs? Q3: When deciding whether to invest in AI or allocate personnel, what criteria do you use to factor in the external environment and how users perceive it, in addition to the performance of the technology? - When deciding whether to invest in AI or allocate personnel, what criteria do you use to factor in the external environment and how users perceive it, in addition to the performance of the technology?