| by Inderjeet Mani |
Conversation, of the natural, face-to-face variety evolved by our ancestors, is said to be an endangered species.
The growth of texting, especially among the young, has already lowered the frequency of our natural conversations, with the negative impacts on personal health and social bonding discussed in books like The Village Effect and Reclaiming Conversation. Our ubiquitous online text conversations with each other are continuing to evolve stylistically, differing drastically from natural conversations.
Meanwhile, the sales and customer service representatives we used to chat to are being steadily replaced by chatbots (with annual “savings” estimated at $65 billion), and eerily human-looking chatty companions are already being envisaged as partners for us when we grow old. Capitalist efficiencies and demographic trends will force us to engage even more with these artificial companions.
What does the growth of mass-produced robotic conversation imply for future conversationalists?
The age of scintillating conversation seems long past; we are unlikely to see another Diderot, the 18th-Century Encyclopédiste whose conversation was said to be “enlivened by absolute sincerity, subtle without obscurity, varied in its forms, dazzling in its flights of imagination, fertile in ideas and in its capacity to inspire ideas in others.” But many of us still relish genuine and heartwarming face-to-face talk.
Must such traditional conversation face the fate of dying languages and vanishing species? Will traditional chats become more like vintage films, to be cherished mainly by the cognoscenti? Or will they continue to evolve and thrive in the midst of ubiquitous Frankenchatter?
The current ascent of chatbots derives from key improvements in the machine’s ability to imitate face-to-face conversation. These include dramatic achievements in speech recognition and the ability to mimic our behind-the-scenes ability to keep track of context through the twists and turns of a conversation.
Bots also aim, if not as yet to be empathetic, at least to be cooperative, based on politeness policies that are in sharp contrast to the frequent rudeness of inter-human chat.
For conversation beyond specialized transaction domains, systems face more difficult and interesting longer-term challenges, including understanding conversation on any topic and building trustworthy relationships with us. To improve their capabilities, they are leveraging the intimate knowledge they have of us, including our conversations, while being increasingly endowed with more convincing personas based on demographic and linguistic features harvested from the big data trove.
Even if the bots are fundamentally limited by the data they are weaned on, they may co-evolve with humans to adapt to each other. Humans have already adapted to the poorer linguistic capabilities of computers, “code-switching” as native speakers often do among languages or varieties thereof. Over time, we may slide further in terms of online brevity, bluntness, and trendiness, all of which can simplify automatic processing.
But bots too will adapt. Much as couples may complete each other’s sentences, a devoted bot trained on a decent quantity of interactions should be able (as search engines can do today) to auto-complete as well as anticipate user topics. The bots are already starting to mine conversational wisdom from on-line corpora; a recent open-domain conversational system was able to display commonsense reasoning by training on 62 million sentences from movie conversations.
Given that humans have fine-tuned a variety of conversational strategies to help gain the confidence and trust of each other, systems are also able to automatically detect such strategies in human interactions, including self-disclosure, reference to shared experience, praise, and violation of social norms, all of which affect the rapport and relationship between interlocutors.
And like sensitive face-to-face listeners, the bots may also learn which topics are off-limits in particular cultures and settings, and when silence is preferred.
Such co-adaptation will allow traditional conversational abilities to flourish, albeit in novel hybrids, in our intercourse with machines. That makes it likely that these abilities will be preserved in any conversations we continue to have with each other, assuming we will still have time for them.