530 V O L . 7 5 P A R T 4 J U L Y 2 0 1 7 THE ADVOCATE
negate a copyright claim by the programmer even if there was no further
input or selection (e.g., computer-generated stories, music or art).
In the majority of cases, where the program is acting as a tool or agent of
the user who is the one ultimately exercising skill and judgment in the creation
of the work, the copyright would belong to the user of the program
(e.g., works created through word processors, graphics programs, etc.). The
situation would be similar to that involving the more complicated program
considered in the Express Newspapers decision noted above.
With the modern development of more advanced computer programs,
the line between programmer and the output of the program becomes
blurred, and there is less need for user guidance.
In particular, it is worth considering those AI programs that are capable
of machine learning (i.e., programs having the capability to shape the final
course of their own programming). Programs of this nature may generate
an outcome based on an initially programmed goal or objective, but with a
final resolution unknown to the programmer. Machine learning programs
have the ability to change or adapt their own programming based on new
data as it is presented to the program, or discovered by the program itself
(through web searches, camera input, etc.).
Machine learning represents the possibility of an AI program that creates
something totally unexpected and unintended by the original programmer,
and that may not need any further human involvement to be considered a
commercially valuable work. For example, a computer program could be
developed to determine the most effective sales pitch to give to potential
clients of a large corporation. The program could be given access to the
Internet and a database of sale pitches; the ability to select, combine and
generate responses to customer inquiries; and the ability to determine
whether the interaction with the customer resulted in a sale or no sale. Such
an AI program could develop a series of unique sales scripts with which to
respond to given inquiries, which might be useful for online sales, training
sales agents, etc., all of which would be possible with no further input from
In such an example the potential customers who input their questions
into the program might not even know that they are talking to an AI program,
and the customers would not be exercising any skill or judgment
toward the creation of the final AI-generated script.
In such a case, the programmer would likely have exercised skill and
judgment in creating the original AI program, but would not have had any
further input with respect to the changes made by the program itself. Would
a script developed by an AI program through machine learning be original
under the Canadian skill and judgment test?