Bell Labs

What it is

Bell Labs is a research laboratory that was organised on 1 January 1925 as Bell Telephone Laboratories, to consolidate the research of the Bell System, the telephone network then run by AT&T. It was owned equally by AT&T and Western Electric, and its headquarters moved to Murray Hill, New Jersey, in 1967.^[1] The laboratory's name and owners changed several times over the decades.

Origins

The laboratory was formed to bring the Bell System's research into one organisation. Its ownership then changed over time. AT&T and Western Electric owned it from 1925 to 1996, Lucent Technologies from 1996 to 2006, Alcatel-Lucent from 2006 to 2016, and Nokia from 2016 onward. When Lucent was spun off in 1996, AT&T kept part of the labs as AT&T Laboratories.^[1]

Known for

Bell Labs is credited with 11 Nobel Prizes and five Turing Awards for work done there.^[1] Its computing landmarks include the transistor, which John Bardeen, Walter Brattain and William Shockley produced in 1947. They also include Claude Shannon's paper “A Mathematical Theory of Communication” from 1948, Unix, created by Dennis Ritchie and Ken Thompson in 1969, and the C language, which Ritchie developed in 1972.^[1]

AI has a direct link to the laboratory too. Claude Shannon of Bell Telephone Laboratories was one of the four signatories of the 1955 Dartmouth proposal, the document that introduced the term “artificial intelligence”.^[5]

Yann LeCun joined the Adaptive Systems Research Department at AT&T Bell Laboratories in Holmdel, New Jersey, in 1988. In 1996 he moved to AT&T Labs-Research as head of its Image Processing Research Department.^[2]

The 1989 paper “Backpropagation Applied to Handwritten Zip Code Recognition” was written by LeCun, B. Boser, J. S. Denker, D. Henderson, R. E. Howard, W. Hubbard and L. D. Jackel of AT&T Bell Laboratories in Holmdel. It trained an early convolutional network on 9,298 handwritten zip-code digits supplied by the U.S. Postal Service.^[3] Backpropagation is the method that works out how much each connection in a network contributed to its error, so the connections can be adjusted. A convolutional network scans an image with small learned filters, which lets it pick up local patterns such as strokes and curves wherever they appear.

Yoshua Bengio did a postdoctoral stay at AT&T Bell Laboratories in 1992 and 1993, in the group of Larry Jackel and Yann LeCun.^[4] The 1998 LeNet-5 paper, which introduced the MNIST dataset and described a check-reading system that read several million checks per day, came from the same line of work at AT&T Labs-Research.^[6]

Place in AI history

Bell Labs shows that AI history is also a history of hardware and communications research. The transistor and Unix were not AI projects, but they supplied the machines and tools that later AI work depended on. The handwritten-digit work of 1989 and the LeNet-5 line that followed showed that convolutional networks could handle real postal and check data. The Holmdel complex, where the 1989 paper's authors worked, closed in 2007 and was redeveloped as Bell Works.^[1]