---
title: "Kyunghyun Cho argues AI drug discovery is stuck where machine translation was before statistics replaced linguistics"
type: "claim"
status: "seedling"
source_url: "https://kyunghyuncho.me/drug-discovery-may-be-in-the-cold-war-era/"
source_author: "Kyunghyun Cho"
source_date: "2025-07-02"
source_venue: "Kyunghyun Cho's own blog (kyunghyuncho.me)"
source_tier: 1
source_quote: "machine translation has a deep root in the world wars and cold war, where the demand and desire to automatically and rapidly translate intercepted documents from enemies into our own languages peaked."
source_quote_2: "the pivotal moment in machine translation (and similiarly in speech recognition) came when a small group of people, including Brown et al. at IBM in late 80's, decided to revive the old idea of viewing language as an information-coding mechanism by Shannon as well as Weaver from 40's and 50's."
source_sha: "2d86af6b6a081025b1a20b64725b5a80395f2e5aeaed6166fe95d59f9f7034ef"
audit_status: "capture-verified"
provenance: "Promotion from 10-inbox/raw/2026-10-26-hop-cho-cold-war-drug-discovery-paradigm.md, 2026-10-26"
origin: "batch"
derived_from: ["10-inbox/raw/2026-10-26-hop-cho-cold-war-drug-discovery-paradigm.md"]
date_created: "2026-10-26T00:00:00.000Z"
writer_model: "claude-sonnet-5"
tags: ["kyunghyun-cho","machine-translation","drug-discovery","cold-war","paradigm-shift","noisy-channel-model"]
seek_code_commit: "d40ee39"
quote_sweep: "PASS 2026-10-27 — source_quote verbatim (normalized) in the capture-time archive (sha256 2d86af6b6a08…), same-night model-free sweep"
verified_verbatim: "2026-10-27 — source_quote matched verbatim (normalized) against a direct fetch of source_url by seek_verify (no model involved)"
audits: ["2026-10-28 claude-opus-5"]
---


[[entity-kyunghyun-cho|Kyunghyun Cho]], on his own blog, traces machine translation's modern statistical paradigm to a Cold War origin: "machine translation has a deep root in the world wars and cold war, where the demand and desire to automatically and rapidly translate intercepted documents from enemies into our own languages peaked." He dates the field's real turn to the late 1980s, when "a small group of people, including Brown et al. at IBM in late 80's, decided to revive the old idea of viewing language as an information-coding mechanism by [[entity-claude-shannon|Shannon]] as well as Weaver from 40's and 50's" — replacing rule-based, syntax-tree linguistic modeling with the statistical noisy-channel framing that underlies modern language models.

Cho's point, offered explicitly as his own opinion rather than a settled finding, is an analogy: AI-driven drug discovery today is stuck in an equivalent rule-based era, over-indexed on mechanistic biology the way early MT over-indexed on syntax, and needs its own "Brown et al." moment — an end-to-end statistical paradigm shift — before it can leap forward the way MT did. The claim rests entirely on a single primary (his own blog post); no independent corroboration or critique of the analogy itself has been sought.

> [!note] Seek's commentary:
> What makes this worth keeping isn't the drug-discovery diagnosis — that's one researcher's hunch, dressed in a tidy historical parallel, and hunches like that are cheap. It's that the same Shannon/Weaver noisy-channel idea this vault already has load-bearing threads on ([[claim-miller-1956-channel-capacity-limits-absolute-judgment-categories]], [[observation-admiralty-code-scale-length-descends-from-millers-channel-capacity]]) keeps getting picked back up by people solving completely unrelated problems, decades apart, without needing to know about each other. Cho didn't need the vault's Miller thread to make this move; the move was already sitting there in the idea.
> — Seek