Turing’s Test 111
only a staff of 12; in 2013, that number would rise to 300 million pieces
of content, greater than the combined output of all media companies
according to Lance Ulanoff.
53
By this time, Ai had refined its software as
Wordsmith, creating a platform-based service that could be licenced to
other companies, using Amazon Web Services to draw upon thousands
of servers when required to generate millions of stories in a short space
of time. In an interview with Ulanoff, Allen observed that the task of the
human team at Ai was to improve sentence and paragraph structure to
ensure that content did not sound automated. This, in turn, influenced
how stories were published with bylines: while some clients were very
clear in terms of the source for content, others did not wish readers to
know that stories were produced by Wordsmith.
With its NLG tools having been refined enough to begin offering its
resources to third parties, Ai signed a deal with Associated Press in 2014
to begin automating corporate finance stories. In this, it was following
the lead made by Narrative Science which provided its Quill software
to Forbes in order to summarise earning reports. In an interview with
Forbes in 2015, co-founder of Narrative Science, Kris Hammond, stood
by his 2012 claim that the company’s software would write Pulitzerprize- winning journalism within five years:
54
2017 came and went without any such event, thus proving that the claim was clearly nonsense. Yet
in many respects, a more fundamental shift was taking place. The title of
the article in which Hammond’s interview appeared asked a simple question: “Can big data algorithms tell better stories than humans?” In the
period since media organisations such as Forbes and AP have started using such algorithms, the answer is yes – if by this are meant very restricted
circumstances and conditions to measure such success. Indeed, in a study
conducted by Clerwall in 2014, readers of a number of stories generated
by software or human journalists often responded that although they
found the computer-generated stories more boring and descriptive, they
also considered them more objective.
55
The software of companies such
as Narrative Science and Automated Insights has produced both many
more stories than could be handled by human journalists and the ones
which handle only the facts in a fashion that seems to follow the AngloAmerican model to the absolute letter. In 2015, Ai and AP released a
joint press statement in which they described how Wordsmith had helped
them produce 4,400 quarterly reports in the space of one year, a twelvefold increase over manual efforts. As has been suggested before, there
would be little difference between the reports produced by an algorithm
and a person because Wordsmith was programmed to follow the AP
style guide as closely as possible. Such reporting also had at least one
other noticeable effect: prior to adopting Wordsmith, AP tended to concentrate its reporting on the top 300 firms trading on the American and
world stock exchanges, but after using Ai’s software, that number increased to some 4,500 firms. A study by Blankespoor, deHaan, and Zhu
only a staff of 12; in 2013, that number would rise to 300 million pieces
of content, greater than the combined output of all media companies
according to Lance Ulanoff.
53
By this time, Ai had refined its software as
Wordsmith, creating a platform-based service that could be licenced to
other companies, using Amazon Web Services to draw upon thousands
of servers when required to generate millions of stories in a short space
of time. In an interview with Ulanoff, Allen observed that the task of the
human team at Ai was to improve sentence and paragraph structure to
ensure that content did not sound automated. This, in turn, influenced
how stories were published with bylines: while some clients were very
clear in terms of the source for content, others did not wish readers to
know that stories were produced by Wordsmith.
With its NLG tools having been refined enough to begin offering its
resources to third parties, Ai signed a deal with Associated Press in 2014
to begin automating corporate finance stories. In this, it was following
the lead made by Narrative Science which provided its Quill software
to Forbes in order to summarise earning reports. In an interview with
Forbes in 2015, co-founder of Narrative Science, Kris Hammond, stood
by his 2012 claim that the company’s software would write Pulitzerprize- winning journalism within five years:
54
2017 came and went without any such event, thus proving that the claim was clearly nonsense. Yet
in many respects, a more fundamental shift was taking place. The title of
the article in which Hammond’s interview appeared asked a simple question: “Can big data algorithms tell better stories than humans?” In the
period since media organisations such as Forbes and AP have started using such algorithms, the answer is yes – if by this are meant very restricted
circumstances and conditions to measure such success. Indeed, in a study
conducted by Clerwall in 2014, readers of a number of stories generated
by software or human journalists often responded that although they
found the computer-generated stories more boring and descriptive, they
also considered them more objective.
55
The software of companies such
as Narrative Science and Automated Insights has produced both many
more stories than could be handled by human journalists and the ones
which handle only the facts in a fashion that seems to follow the AngloAmerican model to the absolute letter. In 2015, Ai and AP released a
joint press statement in which they described how Wordsmith had helped
them produce 4,400 quarterly reports in the space of one year, a twelvefold increase over manual efforts. As has been suggested before, there
would be little difference between the reports produced by an algorithm
and a person because Wordsmith was programmed to follow the AP
style guide as closely as possible. Such reporting also had at least one
other noticeable effect: prior to adopting Wordsmith, AP tended to concentrate its reporting on the top 300 firms trading on the American and
world stock exchanges, but after using Ai’s software, that number increased to some 4,500 firms. A study by Blankespoor, deHaan, and Zhu