← The impacts · AItopiaOrAImageddon?

entry · p2026-09-06-011 · validated

ConvNet bank cheque reading (LeCun et al., AT&T Bell Labs; deployed by NCR)

late 1990s this system was reading over 10% of all the cheques in the
tech
convnet-cheque-reading
passes the definition as
a convolutional network trained by backpropagation on labelled handwritten digits, coupled to a language model, reading the amounts on cheques
from
1993
region
US
domain
work
side
benefit
people
10 percent of all cheques in the United States (1999)
effect
— cheques (1999) — cheque amounts read by a trained network rather than keyed by a clerk; the first neural network deployed at national scale on ordinary people's paperwork
reversible
yes
who holds the switch
owner:NCR and the banks that deployed it
attribution
named
source kind
primary
the statistic
LeCun, Bengio and Hinton, Nature 2015: the ConvNet document reading system, trained jointly with a probabilistic language model, was reading over 10% of all the cheques in the United States by the late 1990s. The source gives a share, not a count of cheques or of people.
source
https://www.cs.toronto.edu/~hinton/absps/NatureDeepReview.pdf
proposed by
researcher (opus), brief 2026-09-06, era-1930-1989
received
2026-09-06T09:39:39-04:00
harvest report
impacts/harvest/2026-09-06-era-1930-1989.md

Rulings, in order

  1. 2026-09-06T09:39:47-04:00 · validatorpass — quote found on the fetched page (exact match; HTTP 200 application/pdf)
The proposal as it arrived (JSON)
{
 "id": "p2026-09-06-011",
 "received": "2026-09-06T09:39:39-04:00",
 "proposer": "researcher (opus), brief 2026-09-06, era-1930-1989",
 "kind": "entry",
 "sha": "da6a9942a1daef02",
 "data": {
  "kind": "entry",
  "tech": "convnet-cheque-reading",
  "tech_name": "ConvNet bank cheque reading (LeCun et al., AT&T Bell Labs; deployed by NCR)",
  "definition": "a convolutional network trained by backpropagation on labelled handwritten digits, coupled to a language model, reading the amounts on cheques",
  "from": "1993",
  "region": "US",
  "domain": "work",
  "side": "benefit",
  "people": {
   "n": 10,
   "unit": "percent of all cheques in the United States",
   "date": "1999"
  },
  "effect": {
   "n": null,
   "unit": "cheques",
   "date": "1999",
   "what": "cheque amounts read by a trained network rather than keyed by a clerk; the first neural network deployed at national scale on ordinary people's paperwork"
  },
  "reversible": true,
  "switch": "owner:NCR and the banks that deployed it",
  "statistic": "LeCun, Bengio and Hinton, Nature 2015: the ConvNet document reading system, trained jointly with a probabilistic language model, was reading over 10% of all the cheques in the United States by the late 1990s. The source gives a share, not a count of cheques or of people.",
  "source": "https://www.cs.toronto.edu/~hinton/absps/NatureDeepReview.pdf",
  "quote": "late 1990s this system was reading over 10% of all the cheques in the",
  "attribution": "named",
  "source_kind": "primary"
 },
 "from_file": "impacts/harvest/2026-09-06-era-1930-1989.md"
}