On AI conversation
Computer voice is getting awesome
Text-to-speech is now allowing less robotic reading voices. This will boost the accessibility of written media.
When I talk to voice agents now (particularly, Sesame), I feel some of those feelings that I feel when speaking to a real human. One feeling in particular, is that I feel compelled to speak up to direct the conversation in the direction I want to take it in. Another feeling is that of being guided in a new direction because the AI raised a good point. So far I mostly use such AI conversations to learn new information and discuss it.
Purposes of conversations
Conversation for regulating emotions and social balance
Conversation for exchanging information
Conversation for constructing social life and meaning (we try to educate youngers, debate peers, etc.)
Conversations as rituals. (Job interview, viva, therapy, small talk, etc.)
Organic conversations: I also think many conversations don’t have a purpose, in the sense that they aren’t started because of a purpose and they don’t end because some objective has been met. I think we find ourselves talking to people that we share presence with. I’d call this an organic conversation. One could say that it serves the purpose of acknowledging presence I suppose.
Conversational Flow
When something is at the right level of challenge, it is fun, it triggers our capacity for play, it puts us into a state of flow. AI agents are now in the zone where they can be challenging conversation partners (again here I’m focussed on conversations where one is learning something or creating knowledge through conversation). It can be fun to talk to them. They also adjust their level of language and knowledge to match the human, so the flow persists as the human learns.
Future of work
Conversations are a very important part of scientific research. A lot of salient information is learnt from others in conversation, a lot of sense making is done by exchanging ideas and making many attempts to grasp what the other person is saying.
If these agents get really good, I can see the future of work looking like us spending a considerable portion of time having a conversation with a helpful and emotionally intelligent AI agent about deep ideas in our field (here I’m thinking about science).
This is not too different from how we have conversations with ChatGPT, but one major difference is that speech agents are far less verbose. So beyond the difference in language medium, this creates a very different experience in the conversation, with only the most salient information being brought up, and a lot more linearity.
We may also find the pace of a conversation and the requirement to produce speech quickly to be motivating. Conversations are more evenly split between human and AI speech production than the current style of responses given by AI chat assistants.
Getting better at conversation
I’m excited at the possibility of practicing and learning how to have a conversation.
In school we learn about formats of writing. However we much less often learn and apply formats of conversation. The closest thing is a very ritualised conversation like Model UN or Debate. I imagine those whose jobs involve conversations as a primary deliverable like doctors or therapists or evangelists also think a lot about formats of conversation.
However, I wonder if there could be formats of conversation that we teach large groups of people. For example, imagine if there was some standard form of a business conversation to understand if two parties shared any interests or to identify possibilities for collaboration. If two people who have both learnt this format of conversation meet, and if they want to identify opportunities to work together, they could just engage in this format of conversation. The way we do this today requires one person to take the lead and lay out the structure of the conversation. I wonder if teaching this in schools could allow very well designed conversation formats to be used by large numbers of people who wouldn’t otherwise put in the initiative to carefully design and lead a productive conversation.
Also just imagine how society would be if there were a format of conversation for laying out political views to understand common ground and differences. Many things get in the way of a good conversation: poor emotional regulation, one person dominating the conversation, etc. but they are all skill issues.
I think AI conversation agents create one pathway for rolling out such an education, since we can roll out the practice and teaching of these formats at scale.
Diversity of AI conversation agents
In personality, ChatGPT, Claude etc. are broadly similar, they are just helpful agents trying to be as neutral as possible.
In a conversation, the personality shines through more. The agent doesn’t list 10 answers to your question, it has to pick one. It can interject. It has a voice. That voice has a tone. It is a lot more emotionally expressive (this might be an accidental different between Sesame and the big chat agents).
I’m excited to see how we will use different agents with different personalities for different purposes.
AI conversations
- Obtaining information
- If you want to be educated on something that has been well documented online, you might do really well talking to an AI rather than a human.
- Reflection / Regulating emotions
- The goal of some coaching/therapy is to create a space for reflection, where just some small nudges and questioning can help a lot. Of course we shouldn’t perpetuate loneliness, but I expect we can supplement real human touch with AI conversation.
- Role playing/Language learning/Learning how to converse better
Why speak to a human?
- Coordinating action
- Rituals
- Job interviews, Viva examinations, recurring meetings, talk shows, podcasts - these serve ritual functions.
- Eliciting their ideas
- People’s subjectivity can matter for its own sake, people can have local and contextual knowledge.
- Because it is good enough
- Maybe you can tolerate the imperfections and risks. You don’t need to be having world-class conversations all the time.
- Giving attention
- People feel cared for when you talk to them, especially the elderly!
What happens when AI becomes really good at conversation?
I think AI will meet some functions of conversation, but not all. I think we should use the AI to meet the functions of conversation it can serve, because that will contribute to well-being and self-actualisation. We are entering a scenario with an abundance of helpful conversation partners and this should help to calm some of our uneasinesses about life and find clarity.
I also think we should use AI to meet the conversational needs it can meet because then we free up humans to meet the conversational needs that only humans can meet.
I don’t think there is enough good conversation in our world - and the fact that we live in a podcast era shows how much we crave that authenticity and negotiation that comes with conversation. Letting AI have the less important conversations with us will free us to have really good conversations with humans.
This post was originally titled "Him".