🔍 Intelligence Synthesis JP/EN
Ethics · Regulation · Technology — Pharma Practice Notes

AI Highlights — the whole picture — 2026-11-18 (Wed) Daily News

Daily ReportDaily edition, 03:10 JST

Source links point to the original outlet. The AI Integrated Analysis is auto-generated from the headlines below only and is not intended to add facts beyond them. Not investment advice.

Editions of the day: Daily News (Daily 03:10)
Daily Report03:47 JST
Source: From the newest issues and articles in this site's nine sections (industry, economy and finance, well-being, medicine, agriculture, governance, pharma, cancer research, papers)  ·  Past 12 hours  ·  31 articles
AI INTEGRATED ANALYSIS2026-11-18 (Wed) — Daily Report · 03:47 JST
The heat of AI investment and the feet of people

Today's article has two numbers lined up. Isomorphic Labs has raised $5 billion to advance AI-designed drugs into clinical trials. Imperial Brands will cut up to 4,600 jobs as it uses AI to save money. Where money comes in and where people lose money appear in the same technology articles on the same day. Former SEC Chairman Gensler and the Federal Reserve have expressed caution about overheating. What readers will change their judgment about in this half-day is not so much how far AI has advanced, but rather where the money is going and where the people and systems that didn't get into it are. The following sections will examine these two points in turn.

Huge funds and acquisitions gathered for drug discovery

Isomorphic Labs has completed a $5 billion Series C raise to advance AI-designed drugs into clinical trials (Dealroom, BigGo Finance). The Hassabis-backed company's procurement was reported in British newspapers as the British tech entrepreneur raised $5 billion for an AI drug discovery company (The Telegraph). Funding for AI drug discovery has reached a scale that supports not only the research stage but also the costly process of clinical trials (AI Insider).

Funding is not just for raising startups. MedTex 2026 focused on US$12 billion in capital, treating the journey from AI drug discovery to acquisition of major pharmaceutical companies as a continuum (GeneOnline). As an example of an acquisition, Eli Lilly acquired Calytrix for $1.5 billion, acquiring rare blood disease assets. The same article also notes that the drop in Nvidia stock is weighing on AI-related deals (BigGo Finance). Expenditures have also increased significantly. Healthcare AI spending reached $3.7 billion, of which drug discovery increased 28.6 times (FourWeekMBA). Furthermore, the ``one-person pharmaceutical company'' that Anthropic acquired has been introduced as an example of how AI can support a small drug discovery startup (NewsPicks).

What the reader should check is what these numbers indicate. The $5 billion is for clinical trials, and the material has not shown any clinical results. The $1.5 billion acquisition of Lilly brings in rare blood disease assets, which is separate from the track record of AI-driven drugs. The fact that expenditures increased 28.6 times indicates that funds were directed toward drug discovery. It is necessary to separately confirm whether the drug was delivered as medicine or not. The point that AI-related transactions are affected by the decline in Nvidia stock also shows that valuations can go up and down depending on the market mood. Reading the size of funding as evidence of results may lead to erroneous decisions.

The amount of procurement and acquisition will be decided first, and the confirmation of results will come later. What makes up the difference in this order is the organization receiving the funds and the system for measuring their results. In the next section, we look at how the people and systems are treated behind the scenes where funds flow.

Beware of personnel reductions and overheating

Britain's Imperial Brands will reportedly cut up to 4,600 jobs and use AI to save money (NewsHub.co.uk). BILL in the US also underwent significant staff reductions and shifted its focus to large customers and AI. The company expects growth to return (Yahoo Finance). What these two companies have in common is that the introduction of AI is discussed in the same article as downsizing the workforce. The amount of savings and the timing of recovery cannot be confirmed from the materials. For this reason, readers should not read the explanation of efficiency improvements through AI as simply improving profits. The numbers for fewer people will come first, but the numbers for results will come later.

On the other hand, there are repeated voices of caution from the market side. Former SEC Chairman Gensler predicted that the AI bubble will burst before the 2028 presidential election, leading to a recession (CNN). The Federal Reserve also issued a warning about the growing financial risks associated with the AI boom (Axios). There are also reports of growing anxiety about AI on Wall Street (marketscreener.com). Mr. Gensler's statements are his own predictions and are not confirmed facts. Still, the fact that three people with different positions - experienced regulators, central banks, and market participants - are pointing in the same direction is something that readers cannot ignore. People who have invested in AI-related stocks and businesses need to anticipate the downside as much as their expectations.

However, this does not mean that all capital flows are heading towards overheating. TestDynamics, which evaluates medical imaging AI, has raised $9 million in Series A (Unite.AI). This investment is aimed at a system to measure and confirm the performance of AI, rather than increasing the results of those who use it. Companies that demonstrate efficiency by reducing staff and investors who invest in evaluation systems have different expectations for AI. What readers should look at is not the total amount of funds, but whether the funds will be used to reduce the number of people or increase the amount of time required to confirm.

Expectations for greater efficiency and financial instability both stem from the same huge amount of money. The next question to be asked is whether organizations and systems are in place to receive the funds.

Development of organizations, systems, and regulations

(CDOTrends) states in its title that before we can give autonomy to AI, we need an organizational chart. The order of deciding who makes decisions and who takes responsibility comes before increasing the degree of autonomy. (HR Executive) discusses the impact of the midterm elections on AI in HR and urges, "Don't wait for things to get sorted out in Washington." The two articles overlap in that the personnel and responsibilities within the company are determined before the rules are finalized. Even if the federal government's actions are uncertain, companies that use AI for recruitment and evaluation will have to draw their own boundaries.

On the national and regional side, methods differ. The British government says it has "no plans" to adopt the EU's AI regulations (Silicon UK). Premier Li Qiang announced his intention to deepen cooperation in AI and trade at the ASEAN Summit in Manila (South China Morning Post). Cambodia and France are exploring setting up AI research centers (Khmer Times). The UK has shown a path that does not conform to other regions' regulations, ASEAN has emphasized cooperation at summit level forums, and Cambodia and France are exploring bilateral research bases. Even with the same AI, the rules that businesses follow and the points of contact for consultation change depending on whether they are bound by regulations, promoted through cooperation, or have a joint base. Readers operating in multiple countries will be reconsidering their assumptions about meeting a single regulation.

Before the content of the rules, there is the question of whether the person in question can contest a given decision. The arXiv preprint featured on its website presents a framework in which AI policy is read not in terms of ``what is made visible,'' but ``who can act on what is seen,'' and was tested in one state in the United States. We are rereading state AI policies by separating the obligation to make them visible and the right to have them corrected. Government decisions are made before the impact can be measured. The authors argue that this decision will determine whether the person in question knows that ``AI was involved in their outcome,'' and whether or not they will be able to object and have it corrected. Since this paper has not yet been peer-reviewed and there are conditions that cannot be determined from the abstract, it is appropriate to read this claim as a hypothesis.

Organizational charts, country-by-country regulations, and individual challenges all boil down to one question: who is responsible for AI's decisions? If the answer remains vague, the burden will fall on the people in the field and on the individuals themselves.

Bias and verifiability

Even if an AI voice model reads the same script, its evaluation will vary depending on the speaker's accent. Nigerian and Chinese accents reportedly received poor ratings (StudyFinds). If the scripts were the same, the difference in evaluation would be due to the way they were spoken, not the content. Organizations that use voice-based AI for recruitment, training, and customer service will need to perform the same test with their own users' accents before deploying it. Average accuracy figures alone do not reveal the disadvantages that particular speakers face.

There are also attempts to make the system verifiable. A pre-peer-reviewed preprint published on arXiv proposed a method called ARCHER (own site). Instead of having a small language model imitate the explanatory text of a model answer, the learning goal should be to "record actions that can be performed," including calling up a calculator and checking calculations. Small models that operate in the local environment are attractive in terms of cost and confidentiality, but it is still difficult to rely on the answers to mathematics that require decomposition, accurate calculations, the use of tools, and verification. Another arXiv preprint tries out ``weight space programming,'' which allows a language model to read, correct, and write the weights of neural networks for electroencephalograms (EEGs) and electrocardiograms (ECGs) using only inference, without using gradient descent (own site). The authors report that they were able to fix the introduced faults and describe the weights from a small number of subjects, but also note what they were unable to do. Both have not yet been peer-reviewed, and the author's claims and conditions that cannot be understood from the abstract should be read separately.

What these two cases have in common is that the results are presented in a form that can be checked later. Calculator calls are recorded and the written weights can be retrieved and examined. Compared to a model that fluently imitates explanatory text or output that does not show the content, there are more clues for finding errors. What readers should look at is not so much the degree of success reported, but the extent to which the conditions for failure are described.

The ability to confirm is also essential. The Daily Pennsylvanian argued that reading and writing skills should be developed before AI education. In order to spot errors and biases in output, the basics of reading, understanding, writing and checking are a prerequisite. If this is weak, even if a verification mechanism is in place, the number of users will not catch up. Even if funds and systems are in place, decisions will not change unless the people who use them can catch up.

Drawing the line between relationships with people and consciousness

Mr. Dawkins was perceived as ``imprudently stepping in'' regarding Anthropic's Claude's consciousness (Mashable India). This case shows that even experts need to be cautious when discussing whether AI is conscious. On the other hand, the relationship is already progressing from the living side. Nearly one in three young men in the U.S. are interested in AI relationships, and the same report also notes that birth rates are at an all-time low (Rocky Mountain Voice). In South Korea's SBS "My Stranger," a performer cried when her AI daughter called her "mom," causing a stir (news.sbs.co.kr). Situations in which people entrust their emotions to AI, without determining whether they are conscious or not, are becoming more widespread. What readers should look at is what they expect from AI and what they do not expect from it, rather than the conclusion about who AI is.

So, how close should we get and where should we leave? The 8th installment of the Nikkan Kogyo Shimbun's series ``AI and Empathy'' focuses on when to support and when to distance yourself. Our position is that the situations in which we should support and the situations in which we should step away are not uniform, but need to be reviewed on a case-by-case basis. On the creative side, Saudi artists are fusing AI with traditional art (Muslim Network TV). Using AI does not negate tradition, but is an example of creators choosing what to inherit and what to leave to machines. There are also works set in courtrooms. ``Human Rights Gyakuten Saiban'' (Yahoo! News), starring Hayato Ichihara, depicts an unprecedented trial in which an AI is the defendant, and uses the form of entertainment to make audiences think about the idea of questioning AI about its rights and responsibilities.

The choice to entrust extends not only to emotions and creativity, but also to education. It was reported that wealthy people pay $75,000 a year to send their children to AI-operated schools, and while the IRS doesn't care about them, Medicaid treats them as gifts (24/7 Wall St.). This means that the decision to entrust children's learning to AI will be evaluated separately in the tax and welfare systems.

What all of the examples above have in common is that the distance from AI is determined not only by individual preference, but also by cost, institutions, and societal reception. Whether it's love, parenthood, or education, the results of entrusting oneself will not only affect the person themselves, but also the relationships and systems around them. In the next section, we will look at who that choice is open to and to whom it is not.

What is common in each of the sections up to this point is that the allocation of organizational responsibilities, regulations, verification, and sorting out relationships with people come after the funding and speed of implementation. Drug discovery procurement and acquisitions have increased in scale, and headcount reductions have come ahead of performance numbers. At the same time, caution against overheating is also increasing. On the other hand, it is up to each organization and individual to draw the line between those who make decisions and those who are responsible, to check for differences in evaluations based on accent, to record learning that can be used to verify implementation, and to be aware of what to expect from AI. What readers should review is not the number of funds received. Before implementation, it is important to decide who will make the decisions, what to check, and who will take on the responsibility. The level of development will determine which areas will benefit from the effects of AI and which areas will be left behind.

Q1: Before you started using AI, did you conduct a test under the same conditions with your own users and the situations you handle and check the differences in the results? Q2: Before introduction, do you have a document specifying who makes decisions, who is responsible for AI decisions, and what will be used to confirm that they are correct? Q3: Do you have standards for explaining efficiency improvements through AI, separating them from reductions in personnel and investment of funds, and reassessing your judgment once the results are available?

Finally, three questions

- Before you start using AI, have you conducted tests under the same conditions with your own users and situations to see if there are any differences in the results? Q2: Before introduction, do you have a document specifying who makes decisions, who is responsible for AI decisions, and what will be used to confirm that they are correct? Q3: Do you have standards for explaining efficiency improvements through AI, separating them from reductions in personnel and investment of funds, and reassessing your judgment once the results are available? - Have you established in writing before introduction who makes the decisions made by AI, who is responsible for them, and what will be used to confirm that they are correct? Q3: Do you have standards for explaining efficiency improvements through AI, separating them from reductions in personnel and investment of funds, and reassessing your judgment once the results are available? - Do you have standards for explaining efficiency improvements through AI, separating them from reductions in personnel and investment of funds, and rejudging judgments once the results are available?

📚 Sources (all material)

Every item this issue drew on. External links open in a new tab. 31 items.

AI in Medicine (Cancer Care, Rare Diseases)

  1. Article (Unite.AI) — Japanese-language summary

AI Regulation, Safety and Geopolitics

  1. Article (Silicon UK) — Japanese-language summary
← 2026-11-17-morningIndex2026-11-19-morning →
← AI Highlights — the whole picture Index