AI Highlights — the whole picture — 2026-10-16 (Fri)

Investments in AI are currently being funded from two sources. The first is a reduction in headcount, with Oracle reportedly planning to make large-scale reductions that will be used to invest in AI. The other is a huge partnership, with the contract between AstraZeneca and insitro reportedly worth ``$1.86 billion'' or ``$40 million.'' The reason for multiple amounts in the same contract is that fixed contributions and conditional limits are listed without distinction under the headings. When reading announcements from companies and governments, the size of the amount and the speed of introduction are difficult to judge. We will look at the examples in each section to see who will pay for the funds and who will receive the results.

Companies that seek employment and financing as sources of investment

AI investments such as OpenAI and Oracle are first derived from the number of employees and financial surplus. Oracle plans large-scale layoffs to free up funds for AI investments, Mezha.net reported. The plan is to reduce the number of people and use equipment and computing resources, and the first recipients of the burden will be the employees targeted for reduction. Meanwhile, Zoho founder Bembu announced that he will not lay off his employees even though SaaS companies are facing difficulties due to AI (The Economic Times). In the same age of AI investment, there are companies that seek capital from employment and companies that do not. What readers should look at is not who is right, but what companies are citing as reasons for making cuts.

At the same time, discussions regarding the distribution of results are progressing. The ILO has stated that the benefits of productivity gains from AI should be shared fairly among workers (Fibre2Fashion). This argument is based on the premise that even if productivity increases, profits may remain only for capital. For companies like Oracle that are seeking to reduce investment resources, the question of who will receive both the costs saved through reductions and the future results from AI will be the subject of a specific examination of allocation theory. Zoho's selection will be one of the examples that can be compared in this test.

Views on employment are not monolithic. In response to a KPMG survey, US CEOs responded that AI will create jobs (Axios). While some companies are currently planning to cut back, managers are showing prospects for job creation. These two may seem contradictory, but the survey responses are based on management's perspective, and the timeline differs from the reductions that have already occurred. Judging the impact on employment based solely on responses will obscure the burden on those currently targeted for layoffs.

The burden does not only fall on employees. A Citi executive said governments are turning to non-dollar-denominated bonds due to AI-driven liquidity crunch (Reuters). It has been pointed out that huge investments are putting pressure on the capital market, and that the impact extends to the country's financing choices. There is a possibility that the burden and the benefit fall on different parties in three areas: job cuts, distribution of results, and funding. In order to measure the claims of companies and policies, it is necessary to ascertain whose burden each claim is based on.

Pharmaceuticals incorporate drug discovery AI through alliances and acquisitions

The amount of money involved in the partnership between AstraZeneca and insitro is not uniformly reported depending on the media. Bloomberg reports the deal is a "$1.86 billion AI drug discovery deal," while ForkLog and BioSpace say it's worth "up to $1.5 billion." Konsulteer headlined the deal with a one-time payment of $40 million, while Mezha.net also reported it as a $40 million contract. The figures of $1.86 billion and $1.5 billion can be interpreted as indicating the upper limit of the total amount paid based on performance. On the other hand, the $40 million is a lump-sum payment determined at the time of the contract, and its nature is different. AstraZeneca itself has announced a strategic partnership to use advanced AI systems to develop transformative treatments for intractable cardiometabolic diseases (AstraZeneca).

What readers should look at is which numbers are actually moving amounts. A large sum is only the maximum value if the conditions are met. The confirmed burden is on the lump sum side. BioSpace sees this partnership as an expansion of its efforts in the MASH and cardiovascular fields. The company's biggest bets are on two areas: liver disease and heart disease. The larger the headline amount, the more it becomes necessary to check the terms of the contract to determine who will pay how much if results are not achieved.

There are other forms of collaboration. Recursion Pharmaceuticals announced that Sanofi has exercised its option on the first development candidate from an AI-driven collaboration (Quiver Quantitative). In the early stages of a partnership, a format is used in which you can later choose whether or not to acquire the rights. Exercising an option is an example of a choice resulting in a concrete outcome, a candidate product. However, the amount involved in the exercise cannot be determined from this material.

There are also movements on the supply chain side. ParkourSC acquires Quartic and launches an integrated life sciences supply chain AI platform (Pulse 2.0). From the beginning of drug discovery to the manufacturing and distribution processes, entities responsible for AI are being incorporated into the periphery of pharmaceuticals through alliances and acquisitions.

In the examples so far, the substance of the transaction is determined by who pays and who receives the results under what conditions, rather than the size of the amount. The AI that pharmaceuticals incorporate through partnerships and acquisitions will be measured by the same criteria in the hiring and allocation discussions we will see next.

Determining which AI to enter the field and its introduction

In the field of radiation therapy, AI is beginning to improve the accuracy of irradiation. AI-assisted adaptive radiation therapy refines the treatment of breast cancer, and doctors describe it as "GPS-like" (CBS News). On the nursing care side, AI-based security tools are also being used to prevent nursing home residents from leaving their homes without permission (McKnight's Long-Term Care News). AI is moving beyond the testing stage and into actual use, both in the professional field of diagnosis and treatment planning, and in the daily work field of monitoring.

The more it is introduced, the more important it becomes for buyers to confirm its effectiveness. Radiology Business shares a three-step buying guide for evaluating radiology AI implementation. This means that a step-by-step process has emerged from within the industry that does not accept vendor explanations as is. In agricultural detection technology, Bioengineer.org reports that lightweight AI detects late blight on potatoes with 99% accuracy and also provides a basis for decision. In addition to the accuracy figures, showing why a decision was made is a condition for whether or not the field can trust the results.

These developments on the ground have begun to be linked to macro numbers. The British economy has outperformed expectations, supported by AI, and the Chancellor of the Exchequer has received a tailwind ahead of the budget, according to Reuters. Here, the contribution of AI is treated as a basis for determining fiscal management. However, it is unclear from the material presented how this contribution can be correlated with individual implementations in the field.

What readers should look at is the extent to which the evidence is provided, rather than the announced accuracy or contribution to growth itself. Providing purchasing guides and decision-making grounds is a mechanism that allows claims of effectiveness to be verified, and is a prerequisite for questioning who will receive the results and who will bear the costs and uncertainties of implementation.

Institutional movements surrounding safety measures and initiative

The United States and China are competing for leadership in the field of AI ahead of the meeting between President Xi Jinping and President Trump. The event is expected to be held in Shenzhen, the South China Morning Post reports. Just before the summit, technological leadership is coming to the fore. Companies' investment and partnership decisions are easily influenced by this tug-of-war between nations. Readers need to read each company's announcement not only in the context of the market, but also in the context of competition between nations.

At the same time, improvements were also being made to the system on the brake side. Singapore has introduced measures to make AI safer and more resilient (Smart Cities World). The content calls for a national framework that not only allows AI to be used quickly and widely, but also has a system that can survive when it stops working or makes a mistake. For those implementing the system, in addition to comparing performance and cost, this means that standards have been publicly shown to ask how the system will withstand failures.

The same question becomes more concrete when the subject is moved to the field of social welfare. A new study published by EurekAlert! recommends safeguards for social care agencies' use of AI. Users of welfare institutions often have no means of objecting to the output of AI. The person receiving the support bears the burden of the error, not the person who decided to introduce it. The research's recommendations can be read as the position that rules should be placed first in situations where the decision to introduce the system and the attribution of damage do not match.

The rules also extend to corporate management. Stephenson Harwood presents seven rules for directors to use AI safely (Stephenson Harwood). Literacy itself will become a requirement for governance, as boards will be both the users and overseers of AI. The more a company incorporates AI into its business decisions, the more it is necessary to first decide on internal rules about who will check and who will be responsible.

While the competition for leadership between nations is accelerating, national security measures, recommendations from welfare workers, and board rules are providing restraints at each level. As competition accelerates, the question becomes who will bear the cost and responsibility of stopping it. In the next section, we will look at what happens to that burden.

What do human emotions and creativity remain?

The question of consciousness has come to be discussed as a matter of safety policy. With the headline "Anthropic wants to protect AI from you. Welcome to the crisis of consciousness," the BBC juxtaposed the question of whether AI is conscious with the way companies protect their users. On the other hand, Cosmopolitan says that it has become easier to "dating" AI, but that the reality is that it is different from love (Cosmopolitan). Mimi Jiang's ``A Day in the Age of AI'', published in the London Review of Books, is a one-day account of daily life in which AI has entered (London Review of Books). What all three have in common is that when AI gets closer to people's inner lives and lives, it raises the question of what to call it.

Emotion and imagination are also talked about from the industrial side. According to Chosunbiz, Samsung's Porcini urged South Korea to fuse AI with human emotions and imagination (Chosunbiz). Here, emotions and imagination are positioned as elements that can be combined with AI, rather than as objects that can be replaced by AI. The same line can be seen in the creative field. LBB - Little Black Book reported that Tokyo-based director and AI visual artist Ikuo Yamamoto has joined Smoke&Oakum (LBB - Little Black Book). The combination of ``director'' and ``AI visual artist'' in their titles indicates that the person using AI is calling themselves the subject of production.

The deciding factor for the reader is who decides where humanity ends. The discussion of consciousness is told by the companies that make AI, the articles about relationships show the reality of usage, and the discussion of emotions and creativity is told by corporate executives and production companies. Even with the same ``humanity,'' what is being protected and what is being sold differs depending on who is talking about it. The standard for measuring the claim, as in the previous section, comes down to who takes on the responsibility and who receives it.

The examples in each section return to the same point. In Oracle's layoffs, the burden fell first on the employees being cut. In the AstraZeneca partnership, behind the large sum was another burden: a fixed lump sum and conditional caps. The radiology AI purchasing guide and Singapore's safety measures require that adopters take steps to verify effectiveness and resistance for themselves. The debate over emotion and creativity also overlapped with the question of who should attribute the achievements of AI. Announcements related to AI tend to be talked about in terms of performance and scale. However, the standard of measurement is who bears the costs and risks, and who receives the benefits and results. Companies and policy statements can be distinguished by whether or not they clearly state this allocation.

Q1: When introducing AI, at what stage do you decide who will be responsible for confirming its effectiveness and shouldering the burden if it does not go well? Q2: Do you record the confirmation procedures and responsibility for the results and errors of the introduced AI in a way that a third party can trace them? Q3: How do you explain how the benefits and burdens of AI investments will be divided among employees, customers, and shareholders?

Finally, three questions

- When introducing AI, at what stage do you decide who will be responsible for confirming its effectiveness and shouldering the burden if it does not go well? Q2: Do you record the confirmation procedures and responsibility for the results and errors of the introduced AI in a way that a third party can trace them? Q3: How do you explain how the benefits and burdens of AI investments will be divided among employees, customers, and shareholders? - Do you record the confirmation procedures and responsibility for the results and errors of the introduced AI in a way that a third party can trace them? Q3: How do you explain how the benefits and burdens of AI investments will be divided among employees, customers, and shareholders? - How do you explain how the benefits and burdens of AI investments will be divided among employees, customers, and shareholders?