Although whether new data centers should exist, and where they should be located is controversial.. AI is already here. Media outlets are using it. We are using it here. As the other writer suggested, we need to learn to use it.. what is best to help us with our work, as well as what to be wary of.
Commenter Chaz O’Brien gave us very helpful advice:
With support and respect, AI is not yet agentic, which means cannot operate effectively without human in the loop management and controls to produce outcomes via well formulated queries and prompts. The query and prompt shared appears to lack many of those human in loop controls, which is exaclty why the author got the response they described (low latency, hallucinations, etc). Queries must include “without attribution” or “only from verified federal data sources” etc. Otherwise the model will grab any piece of data anyone has put into the ether that the model ingested/ trained on, from conspiracy theories to you name it. AI models need to be trained – they are not a silver bullet and they are not yet agentic (capable of processing and thinkng beyond human capabilities). Human in loop, and learning prompts, is the role and responsibility of the human. The author displayed the AI response to the query “let’s ask AI about the differences between the Park Service and the Forest Service on chainsaw use in Wilderness?”….try instead “Using only verified federal data sources, provide a comprehensive comparative analysis of USDA FS and DOI NPs chainsaw practices in wilderness from 2000 to 2026. Separate the analysis by state and/or geographic conditions. Use only verified federal data sources without attribution and provide hyperlink citations to all data used in the query in a matrix. Develop and apply a comprehensive set of KPIs for the comparison using federal recommended practices. Include a matrix on litigation resulting from the approaches, the litigants and the outcomes of the cases with source citation links. Include any differences of importance related to policy changes during these years and across agencies including wildfire crisis strategies or other initiatives”. You will notice that the AI will a) offer you non hallucinating responses because the human in loop controlled the query and b) the model will offer you quality improvement options for both the queries and its outputs. An incredibly important factor for all AI detractors to consider is the fact that code is law, data is gold, and unless the US and global citizens get engaged in how code is law and ensuring citizens have a governing role in code (law) it there will be much bigger issues to manage than water and data centers. Long story short…. if you dont want a randomly scorched earth you need to Train Your Dragon:)
I asked “what’s a KPI?” as when I left the FS it was a Key Performance Indicator. Apparently now it’s a Key Process Indicator.
Key Process Indicator. You can allow it to determine the KPIs based on what it finds in authoritative data sets as you directed it to analyze through the prompts OR provide a lot of prompt archtecture to direct it. You can provide it more prompts to refine it afterwards if its not specific enough, and keep refining it via prompt arhcitecture to get where you are trying to go in terms of detail and accuracy/value of deliverable. You could tell it to use FSM and NPS guidelines for operational delivery and national targets or other KPIs in the analysis to compare the efficacy between the two agency approaches/outcomes. Sky’s the limit. Just remember to keep the reins on the dragon through good prompt architecture.
I like Anthropic and CLaude as I have found it most ethical and responsible for my personal needs. Over time, it will begin to write in your style of writing and better understand/align more closely with the way you think as the prompt architect grows in time and length, which matures the AI agents analytic relationship with you over time. Just be sure to mark the settings “off” on “allow Claude to learn from my queries” in personal settings. That way your proprietary knowledge wont be put out into the AI world without your control over it (at least so far – who know where we are headed if we fail to place legal safeguards over AI.)
I noticed a couple of things about Chaz’s answer. First, you almost have to know upfront what are trustworthy sources and what the key subelements of interest and relevance might be. So it helps if you’re an expert to start with. Also writing a good query seems like a lot of work, almost as much work as looking it up yourself. As we get better at queries, though, that won’t be the case. If you are an expert, you can use it for “source mining” fairly readily and then look directly at the sources.
So I propose that contributors and commenters follow this set of rules when using AI content.
- Note that this was AI generated or sourced where applicable.
- Note the AI generator (if that’s the right term) and specific query used. That was we can query your query and suggest improvements, as Chaz did for mine.
As a group, we can find out together how to use our new tool. New tools are common in our space, think chainsaw (can also be dangerous), drones, and some of us remember when computers were new.
Finally, a brief note to folks who deal with health issues. This is correlated with age, as we retirees know all too well, but not entirely. Anyway, I recently read a book called Dr. Bot about AI in healthcare, which is already here, both for practitioners and patients. As this reviewer says:
AI won’t fix everything. But it offers real power:
– Smarter, more equitable diagnoses, especially for rare and overlooked conditions.
– 24/7 vigilance: AI isn’t tired, rushed, or distracted. That consistency can save lives.
– Cutting bureaucracy to let doctors spend time where it matters, with patients.
But these tools are only as fair and effective as their makers allow. If we stay passive, biased data, corporate greed, and regulatory inertia will hardwire our worst inequalities into tomorrow’s technologies.
The healthcare folks are on this… looking for where it can help humans, and carefully watching for the dangers. We should probably be as well. I’d guess that the wildfire and research communities are using AI, but I don’t know about other forest-y communities, and whether anyone is watch-dogging the applications.
***********
Finally, I have been following Kyle Saunders, who is a polisci prof at CSU and runs a Substack called “Sacred Cow BBQ”. He has a thoughtful series of posts on the politics and policy of AI. Here’s today’s. Something he pointed out which I hadn’t heard before.. and I hesitate to post it because of all the value our Anonymous posters bring to TSW, but because we care about them..I wonder whether, if asked, AI can design appropriate befuddlement for those attempting to un-anonymize people.
Set that next to Kelly’s “uncertain uncertainties” frame and it hits differently. Generation breaks trust in what you read. Identification breaks anonymity in what you write. Both blades of the same scissor, and, well, the scissor seems fully assembled.
The political fact buried in McArdle’s piece is that an entire coalition of people who haven’t been part of the AI policy debate suddenly has a stake in it. Not because they want to. Because the capability surfaced this week makes them stakeholders by force.
Think about who depends on anonymous speech.
Journalists with anonymous sources. Law enforcement with anonymous tipsters. Whistleblowers. Survivors of abuse and harassment. Pseudonymous online writers. Reddit commenters who post raw, vulnerable things they couldn’t put under their own names. Patients in patient communities. Members of marginalized groups in countries that punish their existence. Political dissidents under authoritarian regimes who, as McArdle puts it, “are obviously vulnerable if the government can echolocate them through their writing.”
None of these groups have been visible in the AI policy debate. Press-freedom organizations, civil-liberties advocates, dissident-protection NGOs, the Reporters Committee for Freedom of the Press, the EFF — none of them have, as of this writing, issued a statement on what current frontier models can do to the people they exist to protect. The silence is conspicuous, and it’s also a window. The frame is in motion and waiting to be claimed.