How do you treat artificial intelligences? The implications are just as complex as the inevitable transformations this new technology will bring.
“Hey Siri, please read Michael’s latest blog.”
If your phone just did an annoying back flip because it said its own name out loud and then triggered another nearby iOS device to respond….you’re welcome. We live in strange times, as uncharted as any undiscovered terrain might be without a map. What I’m presenting today is a carefully focused spotlight on what might casually appear to be a tiny detail in this emerging world. I think, conversely, it’s a Rosetta Stone.
Here’s why. This whole thing turns on the use of a single word in my first sentence: please. In this context it’s not what you think. I mean, sure, it’s always good to be polite, and the conventions of civility in common speech are certainly appropriate for all of us to revisit. But let me re-cast the sentence with a subtle modification, and then we’ll discuss it.
“Hey Siri, read Michael’s latest blog.”
Not very different, right? It’s just missing that one word: please. If we’re being honest about this, who cares?, especially considering that I’m only speaking to a digital assistant. What does it matter if I skip a single, socially lubricating word when I’m talking to a massive, yet barely-out-of-diapers LLM housed in the cloud? I might be saying words, but nobody’s ears are really listening to what I’m saying. (And before anyone raises the thorny issue of how how potentially extraneous words asked of AI actually cost real money and energy, I’m sidestepping that matter for the purposes of this essay.)
The naturalist Aldo Leopold famously said, “Ethical behavior is doing the right thing when no one else is watching- even when doing the wrong thing is legal.” When I speak to an AI assistant—and this applies to any of those currently in use— I’m just issuing instructions to a machine in ways that may once have been entering through a long—very long—stream of keystrokes on my laptop. How could there be value in sounding polite when there’s nobody to be either respected or degraded by my behavior?
I’ll highlight three reasons why this matters.
Here’s the first. Does the habit of issuing a verbal command of any sort degrade my own behavior when I’m not paying attention to how I sound? If quality of action—any action— is a direct product of quality of practice, then my practice communicating requires me to maintain good communication habits. Said another way, if I’m only a decent person when I act like a decent person, doesn’t the habit of trying to sound like a decent person help me solidify the intentions behind my words? As Leopold suggests, this would stand to reason even when nobody else is watching…or listening as the case may be.
The second reason concerns a bizarre digital test of The Tragedy of the Commons. (Click the link if you don’t know this reference. It’s a vital thought experiment for our time.) Since large language models rely on iterative feedback to “learn” how to think, I must be aware that my own behavior with LLMs are fundamentally part of their evolution. How I interact with the tool will shape how the tool interacts with me. Taken a step further, my interactions with the tool will influence how it comes to “understand” its interactions with other people, too. It’s true that my own tiny list of commands will have only an infinitesimal influence on Siri’s ethical trajectory, but that’s the whole point of The Tragedy of the Commons. It’s only when everyone looks out for the value of the community’s common resources that the community can retain resources of value. Does my own lack of civility with my inhuman assistant train that assistant to sound less civil? If the answer is either “Yes”, or “Who cares?”, I worry that my own ethical center is hollow. More to the point, incivility to my AI means that it will not learn basic social graces, which will contribute to the overall decline of similar graces as it interacts with others. As already demonstrated, this leads to a negative feedback loop, in this case distributed across the broader base of users. Inevitably this behavior will permeate from the digital world to the real world.
The third point is the most critical. In practicing civil speech with my digital assistant, I’m essentially habituating myself to interact with other living, breathing people in a civil tone. In my estimation, this is not just an aesthetic preference, or a utilitarian cost/benefit evaluation. It is subtly different than my first point, too, precisely because it has to do my externalized relations. My first point has to do with my own sense of self, and how I might practice and thus reinforce my own behavior. In this third point, I’m extending my personal, ethical development to direct interactions with the world around me, and when I say “world”, I really mean “people”.
Here’s a refraction. When I’m on a TV or movie set, a whirlwind of activity inevitably swirls. No matter how well organized the team, no matter how good our pre-production planning, or how smart or even dedicated the crew, urgent concerns will inevitably present themselves all day long. Thrilling though it may often be, production can be an edgy business, with ticking clocks measuring ferocious financial burn rates, to say nothing of a million forces conspiring to impede achievement of creative excellence. To steal a line from The Bear, “Every second counts.”
That edginess means it’s easy to be short with people. Snippy. Curt. When the volume gets turned up, orders have a tendency to be barked rather than requested. It’s practically cliche how some production executives can tromp all over junior crew. (Drives me nuts. I never think this is okay.)
To be clear, you probably shouldn’t get into the game if you’re thin skinned. In the heat of battle, stuff has to happen, and that often means it has to happen without a lot of discussion. As much as I believe production sets should always support a deep sense of intense collaboration, they are not democracies.
That leads me back to my AI. When I ask my digital assistant to retrieve a document or read something out loud or schedule a calendar event, I never have to discuss it. But when I ask a production assistant or an actor or an electrical grip to do something on set, it’s not the same. I’m not necessarily inviting a conversation, and I’m also not asking for a counter-proposal, but I must always be aware that there is the potential for a response, a new idea, an observation, or even an unexpected emotion. I cannot bark an order without awareness that staff from other departments may also be clamoring for my crew member’s attention. I also cannot make a demand without some measure of awareness that the person I’m directing might have other forces pressing their attention, trivial or significant. I’m talking to a person, in other words, and no matter how crazy the moment on set might be, I’m still talking to a person. As far as I’m concerned, that statement alone invests certain categorical imperatives.
I try to remember to say, “Please,” and I try to remember that the word is not just a formality. I mean it when I say it, even though it’s just one single word. When I say “please”, I’m reminding myself that I’m speaking to a person, with all of the person-ness that every single one of us brings to the day. That means ipso facto I’m not talking to a machine.
Siri doesn’t require such social awareness to do what it was designed to do. It doesn’t require please, and it is not (to my knowledge!) insulted when I forget it. But as we all start to lean on various AI systems for more and more of everything, it seems to me that how we treat our robots will ultimately have a direct influence on how we treat each other.
ADDENDUM:
For those who wonder what the billionaire class of AI developers thing about this whole subject, the NY Times recently published a story you might find interesting. Funny thing is that I had completed THIS BLOG before that NYT story hit the ‘net. I simply had to smile.