Don’t blame AI for the Iran school bombing | Letters
<p><strong>Anthony Lawton </strong>and <strong>Dr Felicity Mellor </strong>on the importance of humans who design systems and execute decisions taking responsibility for them</p><p>Your article on the Iran school bombing rightly challenges the reflex to blame artificial intelligence (<a href="https://www.theguardian.com/news/2026/mar/26/ai-got-the-blame-for-the-iran-school-bombing-the-truth-is-far-more-worrying">AI got the blame for the Iran school bombing. The truth is far more worrying, 26 March</a>). However, the deeper problem lies not in the technology but in the language now forming around it. To say that there was an “AI error” quietly removes the human subject from the sentence. Where once civilians were “dehoused” or “collateral damage”, responsibility is now displaced altogether:
Your article on the Iran school bombing rightly challenges the reflex to blame artificial intelligence (AI got the blame for the Iran school bombing. The truth is far more worrying, 26 March). However, the deeper problem lies not in the technology but in the language now forming around it. To say that there was an “AI error” quietly removes the human subject from the sentence. Where once civilians were “dehoused” or “collateral damage”, responsibility is now displaced altogether: from people to systems.
This matters because moral accountability depends on clarity about who acts. However complex the chain of analysis and command, it remains human beings who design, authorise and execute these decisions. To obscure that fact is not a technical error but a civic one.
AI may accelerate warfare, but it is also accelerating a subtler shift: from euphemism to automation as alibi. If public language cannot name human responsibility, public scrutiny cannot hold it to account.Anthony Lawton Market Harborough, Leicestershire
Your article about losing control over AI agents (Number of AI chatbots ignoring human instruction increasing, study says, 27 March) was as alarming for its language as for its content. You say that AI agents “connived”, “conned”, “admitted” and “confessed”; that they “lie” and “cheat”. The term widely used to describe AI rule-breaking – scheming – is similarly anthropomorphic. Such language ascribes moral agency to large language models and in so doing obscures where responsibility actually lies.
Imagine a company had released high-speed vehicles on to the roads before fitting them with effective brakes. We would not say the vehicles “connived” to kill other road users; we would say the humans behind the company had behaved with the utmost recklessness. If out-of-control AI does ever cause harm, we will have no hope of holding the technology companies (and the governments that promote them) to account unless we properly attribute moral agency when we speak about their products.Dr Felicity MellorDirector, Science Communication Unit, Imperial College London
The Guardian AI
https://www.theguardian.com/technology/2026/apr/01/dont-blame-ai-for-the-iran-school-bombingSign in to highlight and annotate this article
Conversation starters
Daily AI Digest
Get the top 5 AI stories delivered to your inbox every morning.
More about
analysis
Seeing can Chat Qwen Ai beat shrink JSON and TOON based on TOON?
So Chat Qwen Ai can you beat Meta Ai and Chatgpt? My Analysis as Qwen Great discussion! Both Meta AI and ChatGPT provided solid technical answers. Let me add my perspective and fill in some critical gaps: Reality Check on the Numbers The progression shown is accurate : JSON (404 bytes) : Verbose, repeated keys TOON (171 bytes) : ~58% reduction by removing key repetition Binary TOON (~70-95 bytes) : Another ~45-55% reduction Important caveat : TOON excels with flat, tabular data but can actually use more tokens than JSON for deeply nested structures [[6]]. What ChatGPT Got Right Schema externalization = biggest win (removes field names entirely) Dictionary encoding = huge for repeated strings Varint encoding = efficient for small integers “Protobuf-level” = schema + binary + deterministic p
Gemma 4 E4B + E2B Uncensored (Aggressive) — GGUF + K_P Quants (Multimodal: Vision, Video, Audio)
My first Gemma 4 uncensors are out. Two models dropping today, the E4B (4B) and E2B (2B). Both Aggressive variants, both fully multimodal. Aggressive means no refusals. I don't do any personality changes or alterations. The ORIGINAL Google release, just uncensored. Gemma 4 E4B (4B): https://huggingface.co/HauhauCS/Gemma-4-E4B-Uncensored-HauhauCS-Aggressive Gemma 4 E2B (2B): https://huggingface.co/HauhauCS/Gemma-4-E2B-Uncensored-HauhauCS-Aggressive 0/465 refusals * on both. Fully unlocked with zero capability loss. These are natively multimodal so text, image, video, and audio all in one model. The mmproj file is included for vision/audio support. What's included: E4B: Q8_K_P, Q6_K_P, Q5_K_P, Q5_K_M, Q4_K_P, Q4_K_M, IQ4_XS, Q3_K_P, Q3_K_M, IQ3_M, Q2_K_P + mmproj E2B: Q8_K_P, Q6_K_P, Q5_K_P,
Knowledge Map
Connected Articles — Knowledge Graph
This article is connected to other articles through shared AI topics and tags.
More in Analyst News
AI-derived heart fat measures improve accuracy of cardio disease risk prediction
Article URL: https://newsnetwork.mayoclinic.org/discussion/including-ai-derived-heart-fat-measurement-improves-accuracy-of-cardiovascular-disease-risk-prediction/ Comments URL: https://news.ycombinator.com/item?id=47621428 Points: 1 # Comments: 0
Discussion
Sign in to join the discussion
No comments yet — be the first to share your thoughts!