In this blog post, we explore whether the day will ever come when artificial intelligence can truly think like humans and converse naturally.
From ‘Star Wars’ to ‘Transformers’, many science fiction films feature robots as a central theme. In reality, numerous robots have been developed, making our lives more convenient by assisting with cleaning or enabling intricate surgeries. However, the robots in movies are quite different from those in real life. For example, in ‘Transformers’, scenes where the protagonist converses with robots appear frequently, and most robot movies include dialogue scenes with robots. To have a conversation, one must understand the other’s words and be able to provide an appropriate response.
A : “Did you eat?”
B : “I’m on my way to school right now.”
In the example above, ‘A’ and ‘B’ are speaking to each other, but they are not actually having a conversation. Conversation is not merely exchanging words; it is the process of properly conveying one’s thoughts to the other. The reason such scenes are common in movies reflects humanity’s dream of conversing with robots. The field of study necessary to create a ‘thinking machine’ capable of communicating with humans is artificial intelligence.
Most people have heard the term artificial intelligence, but few understand its precise meaning. Artificial intelligence refers to intelligence created by humans or other intelligent beings—that is, artificial intelligence. AI is broadly categorized into strong AI and weak AI. Both are computer-based artificial intelligences, but strong AI can actually think and solve problems, whereas weak AI cannot. This article aims to explain strong artificial intelligence, which is necessary for creating robots capable of conversing with us.
In discussions about strong artificial intelligence, philosophers have debated from a philosophical perspective whether it is possible to implement human intelligence or consciousness in machines. A representative experiment supporting the possibility of strong artificial intelligence is the ‘Turing Test’, and a counterargument to this is the ‘Chinese Room’ experiment.
The Turing Test was devised to answer the question “Can machines think?” It is based on the claim that “if a computer’s responses cannot be distinguished from those of a human, then the computer can think.” In other words, if a machine’s responses are indistinguishable from those of a human, that machine should be considered intelligent.
In contrast, the ‘Chinese Room’ experiment demonstrates that a computer’s responses do not necessarily involve actual thought processes. Briefly, this experiment places a person who does not understand Chinese in a room equipped with writing tools and a pre-prepared list of Chinese questions and answers. When a question written in Chinese is slipped into the room from outside, the person inside writes the corresponding answer from the prepared list and sends it back out. The person outside may feel they have communicated with the person inside, but in reality, the person inside has merely copied the prepared list without any thought. From this, it follows that even if a computer can answer questions, we cannot conclude that those answers stem from intelligent thought. In other words, the Turing Test alone makes it difficult to determine whether a machine can truly think. The debate over whether machines can think has been active for a long time.
To discuss whether machines can think, they must first be able to understand the meaning of questions. The technology enabling this is machine learning. Machine learning, as the name suggests, is about giving computers the ability to learn, allowing them to perform actions not explicitly coded. Humans expand their knowledge, understand new situations, and adapt to change through learning. Machine learning is an attempt to implement this human learning ability in computers. For a computer to understand sentences, it first requires vast amounts of data. Countless data points, each containing the appropriate response to a specific question, are stored in a database. Big data and data mining techniques are then used to identify statistical rules or patterns within this massive dataset. Through this process, the computer becomes capable of providing the correct answer to a given question.
Using this method, it is possible for a computer to converse with humans based on pre-prepared data. However, if we ask whether a computer can ‘think’, opinions may differ depending on the definition, but it is highly unlikely at present. This is because it is difficult to process data that does not exist. For instance, babies might combine ‘disappeared’ and ‘is’ to form the expression ‘became there’. Such creative expressions are beyond the capabilities of computers. Of course, current technology can already defeat a world chess champion or win quiz shows. However, chess is played by storing all possible moves from various records in a database and utilizing them, while quiz shows involve analyzing questions and searching a database for relevant information to find answers. In other words, it’s about finding answers based on vast data, not the machine ‘thinking’ like a human. Yet, just as smartphones became commonplace within a few years after the novelty of flip phones wore off, we can hope that someday we’ll be able to converse naturally with robots.
