What is the Loebner Prize?

If you had to speak to a human and a computer online at the same time, would you be able to tell the difference between them?

Doing so might be a bit more difficult than you initially think, especially if you cannot see who you are talking to.

The best example of how this can play out is the Loebner Prize, which is a Turing test that was launched in 1991. If you are actually not yet familiar with the name “Turing,” the recent movie, “Imitation Game” is actually a good place to start. To most, Alan Turing was the de-facto inventor of the chief processes that still drive our computers. In connection with this, he is also celebrated for his invention of the Turing test.

While the concept of running a Turing test to determine the relative level of a computer’s humanity has existed since 1950, it has only been applied to AIs since about 1991. The principal way of doing so has been the Loebner Prize, which follows the same process, except for the fact that it involves AIs.

In the end, once all of the AIs involved have been judged by pre-set criteria on their humanity, those who come the closest to the equivalent of a perfect score are given prizes. These prizes are not, however, something for the machines like a numerical grade on their performance. They’re actually large sums of money which are awarded to the AI teams who created the winning systems.

Wait, you might be saying now.

What is this criteria that these judges use? How does it look in practice?

The judges for the Loebner Prize apparently convene some time before the competition in order to reach a consensus on how the Loebner competition will be structured. Judging by various sources on the subject, this always ends up being a traditional, text-based conversation.

Therefore, to understand the Loebner Prize, it would be best to further define what this sort of conversation involves.

Imagine a question and answer session between humans and machines that is only in a textual format. Beyond this, it is enough to know that both the humans and the machines answer the same questions.

Furthermore, in the end, the machines are then scored based on how much their answers differ from their human counterparts. The less of a difference there is, the more human or human-like they are determined to be. Despite the fact that this process is highly subjective, it is actually quite telling in terms of the difficulty of differentiating between humans and machines in this context.

In future pieces, we will delve into specific cases of this, including the advantages and disadvantages of Turing tests, to help us shine some light on the possibility of Artificial General Intelligence. In doing so, we will also examine the possible connection between teaching AIs collaboration through an improved sort of Turing test.