According to Blake Lemoine’s LaMDA chatbot, Joseph Wilson, a linguist and journalist with extensive experience with oral languages (languages that have not yet been recorded), chatbots do speak like real people. He draws a clear line between spoken language and the written language used to train chatbots:
However, this leaves out all non-verbal means of communication, including body language, oral histories, tone of voice, sign language, and the larger cultural setting in which people are communicating. In other words, it omits a lot of the intriguing material that enables complex interpersonal dialogue.
“WHY AI WILL NEVER FULLY CAPTURE HUMAN LANGUAGE” by Joseph Wilson at Sapiens (OCTOBER 12, 2022)
Written language may be traced back to only around 5400 years ago, whereas oral language may be as old as 50,000 years, according to Wilson. And only approximately half of all languages—roughly 7100 as of this writing—have ever been recorded. Humans communicate primarily orally. So, a sizable gap develops:
In everyday interactions, discussions progress as participants employ a wide variety of communicative cues. Real discussions develop in an elaborate and subtle process analogous to an impromptu dance; they are messy with individuals talking over one another, vying for the right to speak, hesitating to find the appropriate word, etc.
“WHY AI WILL NEVER FULLY CAPTURE HUMAN LANGUAGE” by Joseph Wilson at Sapiens (OCTOBER 12, 2022)
We frequently understate our meaning in hopes that others would understand us by the context. “I told her I’d miss her,” as an example. This could be interpreted as “I told her the truth.” Alternatively, it could suggest “I told her that out of politeness.” Also “There. What you expected me to say, I told her (without reference to how the speaker feels about it). Typically, the hearer will comprehend what is stated based on a variety of cues and situations.
Literature in English that approaches the complexity of spoken language and attempts to reproduce it somewhat correctly, if in a restricted fashion, is half the secret of classic, timeless good writing.
Wilson says,
Machine-generated prose can be nearly indistinguishable from human prose in specific contexts, such as text-based communication. However, spoken language is far more intricate and fascinating than what can be read on a page or screen, from wholly oral languages to the nonwritten clues found in everyday speech.
“WHY AI WILL NEVER FULLY CAPTURE HUMAN LANGUAGE” by Joseph Wilson at Sapiens (OCTOBER 12, 2022)