Genesis: Who Answers for an AI Decision?
An assistant that orders a certificate and a system that decides eligibility require different kinds of trust. Genesis helps frame the difference.
Ordering a certificate through a chat and allowing a system to decide whether you qualify for assistance involve different kinds of trust. The same friendly assistant may appear on the screen in both cases. In the first, it performs a task you chose. In the second, it helps determine your opportunities.
That distinction provides a practical way into Genesis by Henry Kissinger, Eric Schmidt and Craig Mundie. The 2024 book is best approached through its question about using machine capabilities while preserving people’s authority to decide what they want, rather than as a catalogue of the best AI tools.
A system can choose the steps. Who chooses the goal?
In “Reality”, the authors consider machines capable of planning actions as well as providing answers. Their example asks a system to assess the problems it knows about in a field and choose the one where it could have the greatest impact today.
The task seems clear until you ask what “greatest” means. The most money, the fastest result, the most lives saved, the least harm? Leave that choice undefined and a system might perform very well at a task that is not quite what people intended.
The book distinguishes intermediate objectives from an ultimate mission. Knowing how to select the next step does not establish what the journey is for. That distinction remains useful without accepting the authors’ predictions about future artificial general intelligence.
In “Politics”, the stakes rise. If a leader continually relies on recommendations that seem better than their own decisions, dependence on those recommendations may begin to sustain their power. The authors ask whether human rulers might remain sovereign in name alone.
This risk does not require a conscious machine bent on world domination. An institution accustomed to saying “the system recommends it”, while ceasing to explain the criteria, is enough. That is an editorial application of the book’s question: who can challenge the outcome helps distinguish assistance from authority.
A Ukrainian example is no longer entirely hypothetical
The announcement launching Diia.AI on 1 September 2025 described more than consultations: the assistant could order an income certificate through chat. It also announced services matched to a person’s circumstances and described the service as an open beta. 1
That establishes the appearance of a particular interface. It does not establish that the assistant decides eligibility for payments, replaces officials or operates without errors today. Attributing those capabilities to Diia would be an invention.
The example nevertheless makes the distinction tangible. Finding and ordering a service more conveniently is one benefit. Handing a machine the choice of which assistance you deserve is a separate decision with different consequences. Consider a hypothetical future implementation: if a system misidentifies your circumstances, can you see why, correct the data and obtain human review? These are questions about implementation, not a report of an existing Diia error.
Specific safeguards matter more than prophecy
The European Commission explains the AI Act through the risks of particular applications. Its examples include employment tools and certain systems affecting access to essential private and public services; it discusses requirements including appropriate human oversight. 2
That does not make every chatbot high-risk, or mean every requirement applies immediately to every system. The Commission’s explanation, checked on 18 October 2026, places the relevant high-risk obligations from 2 December 2027, with a separate timeline for systems embedded in regulated products. Those distinctions matter for a particular deployment; a book essay cannot supply a legal assessment.
For evaluating Genesis, the useful shift is one of scale. The book considers humanity’s destiny. A regulatory approach asks about a defined application, documentation and oversight. General concern cannot replace a procedure through which someone can understand a decision and seek review.
European rules should not be presented as automatically applicable Ukrainian law either. They are used here as a concrete way to frame questions about systems, rather than a ready-made legal answer.
Where to keep your distance
Genesis sometimes expands the discussion through large historical analogies and predictions about future intelligence until the person responsible disappears from view. “Humanity must define itself” sounds momentous. It does not explain who checks a model in a particular institution, or who answers for its mistake.
Technical descriptions of systems the authors considered current in 2024 cannot be carried into 2026 without verification. Their scenarios do not prove that artificial general intelligence has been created or that development will follow their predicted course. The durable part of the book concerns goals and dependence. Dated forecasts should remain forecasts.
What to read
Read selectively: “Reality”, “Politics” and “Strategy”. The first establishes the difference between answering and planning; the second carries it into power; the third considers aligning machine behaviour with human rules and the limits of control. The conclusion can provide a short ending, without turning these questions into an implementation plan.
Read it for the politics and ethics of delegating decisions. Do not expect a technical textbook, current model evaluation or legal guide. Its practical core is this: when evaluating AI, check who set the goal, who can stop an action and who answers for its consequences, alongside the quality of the answer.
Sources and limits
Henry A. Kissinger, Eric Schmidt and Craig Mundie, Genesis: Artificial Intelligence, Hope, and the Human Spirit, Little, Brown and Company, 2024, English edition. This essay concerns the named chapters and conclusion, rather than validating every forecast. The Ukrainian essay’s external-source review was dated 7 October 2026. Diia’s announcement supplies a dated launch fact. The Commission’s page illustrates the European approach; the relevant implementation timeline was rechecked for this English edition on 18 October 2026.