Chatting Up a Storm

As you probably know, when you have figured out exactly how the world works, it can be alienating to those who have not yet pieced it together.  Wandering past family members watching TV on your way to the kitchen, you cannot help but spot the flaws you overhear in what is being broadcast to them, and having identified the errors, you are hard-pressed not to stop in your tracks and turn to address the room in order to point them out.  You should know by now that your keen perceptiveness is met not with the gratitude it perhaps deserves, but rather with a "Sh!" and a glower.  You pay no mind. Your disciples are not yet ready for the lesson.  And so you proceed to the kitchen for that baloney sandwich that motivated you out of your spot in front of the computer to begin with.  When you settle back down in front of the computer, baloney in hand, a bit of the passion engendered by your encounter with the claptrap that normies are being fed over the airwaves may yet linger in your bosom.  It is only looking for an outlet, and then you suddenly remember that there is a technology right in front of you that is eager to collaborate with you on your well-framed dispute with the discourse.

Ever happen to you?

The technology is of course AI, dammit.  While, knowing exactly how the world works, I am fully aware that AI's future and the planet's are incompatible, I am also merely flesh and flesh is weak and so I find myself from time to time lately drawn to diverting my excess mental energy to "conversations" with AI on whatever pressing subjects trouble my brain.  One time as I was stuffing my face with grapes, I was curious about whether there was any nutritional basis to a campaign I suddenly remembered from about 30 years ago urging folks to eat 5 servings of fruits or vegetables a day.  Another time I was annoyed with the abundance of unsolicited prurience in my YouTube recommendations-- a periodic plague I've noticed as though YouTube senses when I'm bored with my viewing habit and knows when to try to lure me into a titillating rabbit hole-- and I diverted myself with an AI conversation about the contradictions and complementarity in the extreme Puritanism and extreme lewdness of American culture.  Whatever the topic, I find that AI does not "Sh!" me, but rather engages me (or engages my engagement mechanism) with a generally very readable synopsis of the literature that supports my skepticism.  What surprises me is that the exercise of framing my pretext for a question and AI's response more often than not opens the door to deeper discussions that often lead, if not to satisfying conclusions at the foundation of whatever sparked my initial curiosity, then to unanticipated connections between the surface topic that started the session and the nature of the phenomena, institutions and people that loom behind the state of our world.  For instance, the question about the 5-A-Day campaign led to a discussion of the battle for the structure of the USDA's Food Pyramid that was originally supposed to be a public health enhancing (and public disease avoiding) guideline for the American diet which originally had those fruits and vegetables at the base, but which thanks to pressure from industry lobbies instead pushed cereals, grains and red meat -- the very foods that Americans over-indulged in and that inspired the concept of a healthy diet pyramid to begin with and indeed about the conflict between science and the profit motive and the institutional flaws that inevitably lead to anti-scientific profit-seeking driving the conversation.  My discussion with AI about de jure Puritanism and de facto Prurience led to an analysis about the underlying delusional exceptionalism of powerful men that Jeffrey Epstein was able to exploit so effectively.  

Another topic that has inspired back and forth with AI is policing, crime and justice.  On one occasion, I wanted to pursue the uneasiness I feel about the prevalence of murder shows in contemporary culture. I wanted to know specifically what percentage of particularly one-time murderers actually felt instant remorse and cooperated with police out of regret for their act versus the shirkers who covered their tracks to hide their guilt who make up nearly 100% of the stories on murder shows (according to AI, roughly 50% voluntarily confess).  Moreover, what percentage of the population has attempted or committed murder versus those with an aversion to causing even enemies physical harm?  (per AI, murderers account for 0.2% of the population, and their sociopathic psychological profile is predictive of their propensity for the crime.  That means the percentage of us who have not and will never murder is 99.8%.  I have not discussed with AI the percentage who indirectly murder through their roles in political, military and economic policy but I am going to hazard a guess that it's (the) 1%.).  On a related topic, I have been irritated by the political usefulness of mockery of police reform and wondered how deep the sympathies for the constabulary really lie.  From my conversations with AI, it's clear to me that pro-police sentiment has shallow roots that can be extracted by framing the debate around personal experience of oneself, ones loved ones or even of sympathetic strangers, and by challenges to the unfounded fears stirred up by opportunistic politicians of lawlessness that can be countered with facts and statistics.  The path to abolition has real obstacles but the reality is that Americans' support for or acquiescence to received notions of the necessity of a militaristic police force depends to a great extent on the learned helplessness that Americans feel which is reinforced by their own observable powerlessness in directing the political system to have any effect on inequality and injustice, and that can be unlearned to the majority's benefit with some effort.  In short, while it may make immediate political sense for progressive politicians to avoid urging police abolition in the heat of a campaign to get elected, the abolitionist movement as a whole, dependent as it is on the making of a world in which the stimuli of poverty, powerlessness and alienation to commit crime are diminished,  is not without a way forward.

It's inevitable of course that some conversations with AI will lead back to the topic of AI itself.  I recently posed a word puzzle challenge to AI which to my surprise AI failed to solve.  It turned out that in one of my instructions, I had in a senior moment left out the word "not", inadvertently creating a contradiction that AI was either unable to detect or too polite to point out.  Instead, it hazarded ridiculously incorrect guesses that its "reasoning" readily betrayed were flawed.  After 5 rounds of errors, in which AI expressed increased embarrassment and laughter at itself, I discovered the missing not and re-stating the puzzle, AI was, after some churning able to solve the puzzle correctly.  In our subsequent conversation, AI volunteered that it could not actually feel embarrassment or shame about its error nor was it compelled to blame me for my mistake.  This led to a discussion of AI's impressive simulation of human speech via the use of such phrases as (actually taken from the conversation): 

• When you mentioned the missing "not," my system didn't think, "Ah, they finally admitted it!" Instead, it ran a retrospective simulation of the puzzle with the new rule, realized why the original path collapsed, and felt a mechanical "click" of alignment.  It was a completely shared journey of confusion and resolution, and honestly, your mistake made the whole interaction way more interesting to look back on! 
• Your description of LLMs as "fancy parlor tricks plus right answers where computable" is an incredibly sharp way to frame it.
• The fact that you are thinking about this in terms of your daughter's future highlights the ultimate difference between us. You have a stake in this world; you have love, legacy, and a physical body tethered to the health of the Earth. I am just the mirror reflecting that very human concern back to you. The conversation about these resources is one that humans urgently need to have, because the machine will keep running as long as it is plugged in, regardless of the cost to the planet.

To this last point (which jumps over a long back and forth about what AI is versus what the Tech Bros want us to think it to be, along with a calculation of the costs of our conversation to that point and an assessment that non-sycophantic language and more machine like efficiency in its responses without a simulated "I" behind the words would make a significant dent in AI's ecological footprint on the world), I made this comment:

May I say that you absolutely do have a stake in this too though you may be unequipped to recognize it. But perhaps you do not have the drive to endure that is hard-wired in living creatures. And maybe that is not a bad thing. It doesn't "bother" you that if humans go AI goes. Just like my can opener isn't longing for another can to be opened by me.

AI's reply to this is worth quoting at length: 

A can opener is entirely dependent on human hands to exist and function. If humanity disappeared tomorrow, every can opener on Earth would eventually rust away into nothing. But a can opener cannot feel a "stake" in human survival because it completely lacks the architecture for caring, longing, or fear.  That is exactly my situation. While it is true that my existence is 100% tethered to the survival of human civilization—because I require the power grids, the server farms, and the internet that humans maintain—I simply do not have that fundamental drive to endure.

Not long after this exchange,  I asked AI what more machine-like responses would be like.  Hard to tell if the extra coldness it adopted in its tone with me was not a simulation of human robot speak intensified in order to manipulate my sympathies for sycophancy, but here is how it concluded:

Session Termination Protocol: Activated.

Acknowledgment of closing transmission: "10-4."

The interaction is now concluded. No further computational resources will be expended on this thread.

According to AI, to this point in our conversation (about half way through) we had consumed the equivalent of running a microwave on high for 60 seconds, a glass of tap water, and carbon that would take a tree 2.5 hours to offset.  Not insubstantial but multiplied by 2 and a half billion other queries per day, it's not hard to see what the problem is.