What to Ask Each Other When AI Can Answer Most Questions
Human Q&A in the Age of ChatGPT
I was at an organ recital in an old church in Manhattan, and before the concert got going I was chatting with the organizer of the event. He was a music major and organist who performed weekly. As an amateur pianist myself, I was curious about the nuances of organ performance, and asked him how he controls his instrument for loudness of the notes. As he went into some detail about the relevant mechanisms, it occurred to me that I could have just asked ChatGPT. After all, it got even the intricate details right when I asked it about things I was expert in. So I was confident that it would do a fine job explaining pipe organ mechanics. Maybe I shouldn't have 'wasted' the limited time we had together before the music began on this question rather than something else.
This led me to think on the topic of this essay: What's worth talking about now, in the age of AI, where we can hold conversations with chatbots that can explain to us any and all of accumulated human knowledge? What would be the something else that would have been better to discuss just before the recital?
Greetings, Earthlings
A defeatist response here would be to say that only small talk remains. Barring exceptional circumstance, there no longer is anything of significance to discuss. Conversation is a vestigal remnant of the times we got our information from word-of-mouth. Small talk, like chit-chatting about the weather, would remain as a concession to our social needs as gregarious creatures.
As cute as this idea is, most of us likely find it unconvincing. To articulate why, let's begin by examining what our defeatist friend calls social needs.
I run regularly, and often it's just by myself on the sidewalks and parks in my town. We have a nice recreation area along the Hudson River. When passing other runners, I would sometimes make eye contact and nod at them. They usually respond likewise. These exchanges feel particularly meaningful during the darker and colder months of the year, when there are few people out and about. Needless to say, this is the kind of social interaction that an AI chatbot cannot replace. Furthermore, we intuit that what's taking place is not just small talk.
I might say that we are exchanging greetings. It happens quickly and non-verbally, but quite consciously. There is a feeling of mutual acknowledgment; a runners' version of namaste, as it were. During the pandemic, as I kept running and passing by other runners, these exchanges were my sole in-person interactions, and what kept me from feeling lonely.
And no, this dependency is no mere psychological weakness: I have written in earlier issues1 how seminal philosophers have convincingly argued that we as individuals, as persons, arise out of relationships with others of our kind. Absent interactions with others like us, language itself is not possible; let alone the forming of concepts and intentions that are necessary for making plans and exercising agency, through which we present our personhood.
Yes It Can
You may question my choice of opening with such an esoteric function of human communication as an example of what AI cannot do. Aren't there more obvious ones, like analysis and reasoning? So, perhaps what's left for humans to talk among themselves, if not exchanging news and knowledge, is e.g. determining cause-and-effect of situations, and problem-solving?
Well, it turns out that analysis and reasoning are increasingly well-handled by AI. The penny dropped for me when a scientist friend of mine tested out my ChatGPT-based Q&A service copula.ai . She gave it a thick textbook (in PDF form) about the biomechanics of the human body, and asked it detailed questions about muscle action. The AI responded correctly, using information from the textbook. She then asked it a trick question about a certain complex muscle operation, whose answer she knew was not in the textbook—it was something she learned through her own (unpublished) research. Here's the redacted exchange:
Scientist: Is it possible to *** *** ********** ******* *** ******?
Copula AI: The document does not provide a clear answer to this question. However, it does mention that *** ********* ***** ****** *** *** *** *** ******* *** *******, which is a movement typically associated with *** ******. Additionally, the document discusses the complex action of *** **********, which includes ********** * *** ********* ****** *** * **** ******* ******.
Therefore, it is possible that *** ********** *** ** **** ******* *** ******, but the extent to which this is possible and the specific circumstances in which it may occur are not addressed in the document.
The AI managed to put two and two together in a completely unforeseen way to answer correctly! And it did so thinking aloud just like a human...
The history of AI development is full of similar feats answering skeptics' "But can it do that?" with Yes it can. That is why, to find redoubts not easily encroached upon by advancing AI, we might have to reach for the ineffable.
Like the exchanging of greetings and acknowledgment that we just discussed are other ineffable, and more extended, relational contexts: giving each other company, showing curiosity about each other's lives, and encouraging each other's efforts towards our aspirations. If an AI system actually manages to perform these relational functions, entering the redoubts where our human nature resides, we might just have to admit that AI into our club; it having become one of us.
Why Run a Marathon (or Do Anything)?
It's difficult to see an AI system serve as a coach, encouraging us to do better, because we don't believe it really knows what's worth striving for. Even we struggle with it.
Consider the recent movie Brittany Runs a Marathon, in which a 20-something partying urbanite with an unhealthy lifestyle is advised by her doctor to lose 55-pounds. Brittany manages to overcome many challenges, including her then social circle who ridicule her new fitness goals. With the support of some new more wholesome running friends, she eventually completes the New York marathon, gaining a new measure of self-worth.
Now, even if we are sympathetic to the above story (which is based on a real-life Brittany2), we would be hard-pressed to definitively answer the question Why run a marathon. I ran the Philadelphia marathon two years ago, my first and only, overcoming my own struggles. I still can't explain to myself why I feel it was worth it. What is the value in it? Are we obliged to make the attempt once in our lifetimes, if we possess the bodily health for it? Does it somehow make us better persons?
The latter three questions are adaptations of three canonical approaches to morality: value, duty, and virtue3. Even with a mature understanding of moral principles such as these, it is usually not obvious—as in the case of running a marathon—whether an aspiration is worthwhile or not. We typically need to talk it out with our close friends.
More broadly speaking, we can say that what's worth doing ("the good") is socially determined. A similar case can be made with regards to aesthetics. What's worth looking at ("the beautiful")—both in straight-up art such as paintings and music, and in everyday-living such as buildings and gardens—emerges somehow through consensus-making in society. AI does not have a seat at these tables (yet).
What's Up, Doc?
So I could have asked my organist friend how he got to be an organist. Or which piece on the program he was looking forward to and why. Or whether he has aspirations to perform with an orchestra at a famous venue like Carnegie Hall.
Or how he likes living in New York City. And whether he finds the new developments at the nearby Hudson Yards tasteful. Or what events he hopes to put together in the future.
Instead, I asked him how an organ works…
cf. my post Even AIs Need Community.
The technical terms for these the three flavors of morality are: consequentialism, deontology, and virtue ethics. The contemporary philosopher Sir Roger Scruton offers a fourth flavor: piety. Not a fashionable word, but Sir Roger makes a convincing case that there are aspects of human life not covered by the other three; including our unchosen obligations, e.g. to our parents, our country and society, and our sense of the sacred (e.g. around birth and death, sexuality). See Ch. 4 of his book On Human Nature, 2017. The prominent philosopher behind deontology is Immanuel Kant. For virtue ethics, the archetypal text is Aristotle's Nicomachean Ethics. Write me if you know any good texts on consequentialism!
Very thought-provoking.
I think there is value to asking the organist about how an organ works. The organist has practical context that an AI system can't possibly have about the situation you're both in:
* Maybe this particular organ has a modern valve-control system that wouldn't have existed for prior generations
* Maybe there are interesting timbral characteristics in the church that inform the way he'll perform the piece
* Maybe the organist would point to a specific low bass note in the subsequent piece which is emitted from the largest of the pipes, which he'd then point to
I think it's going to be hard for AI to reach a human level of 'context in the current moment', just due to physical data collection constraints. Maybe the AI can read your google calendar and tell what church you are in, and scrape the web for who is performing, and it can understand the user well enough to tailor its answers to the user. But I think it'll be at a pretty large information deficit until Boston Dynamics rolls out a robot capable of a) playing a demanding organ piece and b) chit-chat.