After a recent conversation with a chatbot, I had an opportunity to see how the bot would hold up against the Turing Test. The chatbot was subjected to pseudo-random questions, each meant to see what the bot’s response would be. The bot exhibited refined ability to understand and respond to questions, even managing to decipher emotion based on the tone of the human user.
A number of questions based on simple logic tests showed that the bot actually has an average intelligence capability. Despite managing to understand the human emotion underlying a sentence, the bot finally failed the Turing Test. After only one cycle of questions, the bot gave the exact same response it had given to a question that was previously asked. This, in my opinion, shows that “mitusku” is a machine, since no human would give the exact same answer to a random question asked twice.
The question that gave mitusku away as a machine is: “I want to find out how smart you are.” Mitusku’s response was “I am smarter than all the other robots.” After an interval of random questions, Mitsuku repeated the same answer after the conversation drifted back to the bot’s intelligence. This gave the machine away.
Mitsuku showed the ability to mimic a human by acknowledging that I laughed after it said that its source code was written by Charles Dickens, who died long before the Internet was created.
According to the article from Wired, most bots written in this day and age are not as smart as they are made out to be. In my case, I showed this to be true by getting the machine to give away its non-human nature.
The authors of the Wired article claimed that the machine did not pass the Turing test because it only managed to convince judges of its humanity 30% of the time. The other 70% of judges could tell that it was a machine from the conversation.
The Turing test is cited as too old a way to truly gauge the performance of modern computers. The test did not take into account that neither human nor artificial intelligence was available in sufficient quality to be analyzed when the Turing test was created.
A revision to the Turing test was suggested by Gary Marcus. He proposed a test of intelligence where the machine would need to successfully answer questions about a YouTube video or movie it had watched, in order to pass the test.