Vagan Terziyan, Timo Tiihonen, Amit K. Shukla, Svitlana Gryshko, Mariia Golovianko, Oleksandr Terziyan, Oleksandra Vitko (2025), AI and Ethics, 5(5), 5165–5190 — DOI-10.1007/s43681-025-00759-9
Key takeaways
- Proposes, without empirical validation, that AI systems that design further AI systems should carry an immutable harm-prevention principle through a genetic-algorithm model of “responsibility inheritance”, with self-clones sent out and recalled (“time-travel abstractions”) to audit ethical drift.
Seeded from the abstract and a skim, before the reading in NOTE-156. What follows is what the work says about itself.
The paper addresses how to keep ethics and accountability intact when autonomous AI systems design later generations of AI. It proposes a framework of meta-responsibility built around one immutable principle: do not harm humanity or violate fundamental values, monitor and mitigate misuse, and make every derivative system inherit the principle. A genetic-algorithm model separates immutable from mutable principles so that systems can adapt while keeping the ethical core during self-cloning and design across generations. Time-travel-inspired abstractions are used for auditing across generations.
Standing in the record
Filed from the 2026-09-26 sweep of the papers-feed tracker for philosophy: 30 seconds of active reading over 1 session. It was filed Deferred, unread. NOTE-156 is the close reading of 2026-09-26, and it placed the work: Rejected — not worth a reader’s time. It is a programme statement whose one concrete proposal is contradicted by the paper’s own later sections. It demonstrates nothing, even though its conclusion says it does (p. 21). It does not engage the alignment work on preserving goals across successors that the problem calls for. A reader interested in value preservation under recursive self-design would do better starting from that literature.
It was seeded with the anthology-candidate flag (ADR-005), because the seed judged it chiefly about machine-learning practice. The close reading found no ML practice in it — no method, procedure, data or empirical claim, and only a generic penalised fitness function — so the flag was removed on 2026-09-26: it stays here.
When seeded, before the reading in NOTE-156, the entry gave a priority for a deeper reading: low — t=30s; a speculative framework without validation, and the skim captures it.
What the seeding asked a deeper reading to check (NOTE-156 is that reading):
- The problem — value preservation under recursive self-design — is a real alignment concern (goal-content integrity across successors); check whether the paper engages the alignment literature on it (it cites the IEEE and Turing guidelines, not e.g. corrigibility or reflective stability).
- There are no experiments or formal guarantees; a deeper read should confirm that nothing is demonstrated beyond the proposal.
- Home is marked anthology because its subject is how AI systems are designed rather than a philosophical question; its philosophical content (responsibility, agency) is thin.
Access when seeded: Springer’s HTML and PDF returned a bot challenge to curl; read the published CC-BY version self-archived at the University of Jyväskylä (JYX). Read the abstract, §1, all section heads, §§3.3–3.4, 5.5, 6.2–6.3, §7 including 7.4, and §8. Online date 2025-06-16 (Crossref); received 2025-01-20, accepted 2025-05-21.