Thinking About Chatbots and Thinking

What is thinking? This is something I have been pondering ever since the Danbee.ai project. The goal of the Danbee.ai project was to build “a platform for making chatbots.” A conversational AI, a chatbot, is surely the most typical image of AI that we have in mind, isn’t it? So before pushing the story further, let’s take a look at how chatbots actually work. What a chatbot does is understand what the user says and respond with an appropriate answer or question. Understanding what the user says is the crucial part here, and the most important key to doing so is the “words.” By analyzing which words were uttered, we can figure out what topic the conversation is about and what the user wants regarding that topic. Everything else is either secondary or lies beyond the chatbot’s domain. You might wonder whether the answer matters too, but that answer is usually just a matter of pulling out something already entered into a database, or hooking up another search engine like Google. Processing the searched data to create new information, along with judgment and decision-making, become problems beyond the chatbot.

How does that feel? Disappointing? At least it is to me. The artificial intelligence I expect is not something like this, but something far more complex. Of course, chatbots are revolutionary. You can operate a machine with words. And they work astonishingly well. At least for words they already know. Just like a human. And this is exactly where my concern lies. A chatbot can come close to a human with nothing more than simple input and output? If so, the reason is one of two things: either the chatbot is at the human level, or the human is at that level. But either way, it doesn’t matter much. In other words, the level of everyday conversation that most people have is generally about this. After confirming which words the other person is saying and what they want in relation to those words, you spit out the information stored in your head. Difficult tasks like complex information processing are absent or rare. If so, what use is a human who is merely at the level of a chatbot? People are expensive and chatbots are cheap. And for a human to be equipped with high-level information-processing ability also costs a great deal.

By now, it’s about time for me to wake up from the dream too. So let me end this story with a banal declaration. People with intelligence at best around the level of a chatbot will be weeded out. Those below the chatbot level go without saying. For people who manage only obvious input and output, or who struggle even with that, a bleak time will soon arrive. And the ability to process information, judge, and make decisions will be regarded as ever more precious. Perhaps a dreamlike haziness will turn out to be the very proof of being human. Of course, it won’t do to be merely vague. It would be good to embody the contradiction of sharpness within ambiguity, probably…