A Look Into the Implementation of AI Chatbots Attempting to Emulate Human Behavior
As time passes, nature takes its course. The sun goes on shining. The plants go on growing. And another constant I think I can add to the list is the continuity of scientific discovery. As science grows more and more advanced, and people go on understanding a little bit more about the Universe than the day before. How is this relevant? As technological aspects of society become more and more advanced, bots and other programs harness capabilities that become increasingly adept at emulating human behavior, and as the gap between artificial intelligence and human behavior gets smaller, we can be left wondering what it truly means to be human. The Most Human Human: What Talking with Computers Teaches Us About What It Means to Be Alive by Brian Christian explores just that, making it a pretty interesting read.
Alan Turing was a prominent computer scientist that was highly influential in the development of computer science and artificial intelligence. He designed the Turing test, which is conducted every year to test the latest AI programs’ ability to mimic human thought in an effort to answer the question as to whether machines can think. The format of the test is such that a judge will communicate through a computer terminal with a human participant and an AI chatbot for 5 minutes each. Without knowing who is whom, the judge releases his final verdict after 10 minutes of contemplation regarding both conversations in an attempt to guess which of the two is a human. In simple terms, an AI chatbot and a human are both trying to convince a judge that they are human through their manner of dialogue. The program with the highest number of votes across a panel of judges each year is given the title “Most Human Computer”. Similarly, the human with the greatest number of votes is given the title “Most Human Human”.
Large datasets are needed for advanced AI, and chatbots that respond appropriately to a user are no exception, especially when trying to imitate humans in the context of the Turing test. There are many different approaches for the use of data to train an AI program that can have a conversation with humans, so let’s take a look at an interesting approach called “crowdsourcing” which is used by Cleverbot. Cleverbot interacts with its users in order to improves itself by tagging user responses with inferred meanings and adding it to a database. For example, if Cleverbot says “Hello” then any response given such as “Hey” or “What’s up” is entered into an extremely large linguistic database, typically called corpus, and tagged as a type of greeting. Hence, as users further add to its well of knowledge, Cleverbot continues to make replies based on previous conversations.
If I had to describe Cleverbot in one word, I would probably use the word — dynamic. That is because Cleverbot is intended to imitate human individualism by using whimsical conversation to portray a unique identity. This is only made possible by formulating responses to users based solely on the most recent stimulus.
The following conversation is why Chatterbot can be referred to as “dynamic”:
And yet it can set off some alarm bells at times:
Questions about the chatbot itself indicate a lack of coherence caused by it replying with existential records in a database compiled from many different users. Its strength — crowdsourcing — is its fluidity in conversation which successfully imitates the human spirit, but it is also a weakness because it lacks the ability to keep track of its own identity.
Christian says it best: “This is a central trade-off in the world of bot programming, between coherence of the program’s personality or style and the range of its responses.”
Sources
Christian, Brian. The Most Human Human: What Talking with Computers Teaches Us About What It Means to Be Alive. Doubleday, 2011.