Get the Ricochet app
Ricochet will work fine in a browser on your mobile devices, but for the best experience please download our app.
I asked ChatGPT, Grok and Comet/Perplexity if AI can operate just using solar and wind energy sources. All three gave me suspiciously optimistic answers. Apparently, AI relies disproportionately on greenie internet sources that downplay the costs and downsides of “renewables.” If AIs are relying on what can be found on the internet to guide recommendations for AI tech's own development (and survival), would that result in self-defeating inferences based on bias and outright misinformation? [Insert your favorite Star Trek episode reference about a fatally confused computer or robot here.]
Grok estimates that supplying electricity for a continuous-use 100 MW datacenter would take about 500 acres of solar panels. Of course, at night that would require drawing on a really big battery, so some multiple of that acreage and some really good storage technology would be required. That is a lot of real estate, and with expensive, still-evolving tech highly dependent on rare metal resources. It is not surprising that Mr. Zuckerberg, Mr. Bezos and other big tech execs are buying up access to nuclear energy in lieu of wind and solar, even as they garner applause for their current visible instances of reliance on “renewables.” They must not be relying on AI inquiries to guide their planning choices.
Not just AI but collective human cognition appears to be crippled by online ideological filters. Like some ruling Procrustean monster (see, e.g., Wikipedia), every bit of content must be pruned or contorted to conform to some prevailing narrative. Much like war with East Asia, we have always believed that anatomy is overruled by identity, that all natural science is about climate change, history is merely a chronicle of transgressions by white men, and collectivist policies always work better than freedom.
At some point, the fiction that CO2 is the “control knob” and that predicted catastrophic change is already upon us will become unsustainable as dire predictions continue to fail to happen. And then even AI will know that. But that will not happen soon, and some really bad choices need to be avoided in the meantime.
Did you know that the U.S. Capitol does not rely on renewables? Like much else it has for itself, Congress has its own power plant that runs on natural gas but captures much of the excess heat as energy for congressional office buildings' HVAC. It is an admirably efficient instance of “cogeneration.”
You would think that would be a model for policymakers in lieu of far less cost-effective, less reliable, but ideologically favored approaches.
The three AIs I consulted do say positive things about this kind of cogeneration — if you ask. That is probably because greenies don’t want to praise anything that uses natural gas, and especially don’t want to have to admit that cogeneration with natural gas is highly efficient, so they don’t talk about it. The result is that there is far less greenie content on the topic to distort the AI perspective.
An object lesson about policy and innovation in a nutshell: (1) The giant agribusiness corporation Houweling’s is closing its California operations because it is no longer feasible to do agriculture in California. (Let that sink in—how bad does a government have to be to make what was once the most productive, agriculturally diverse, climatologically ideal growing region in the world unprofitable for growers, even greenhouse growers?). And (2) Houweling’s is building a 13.2 megawatt combined-heat-and-power project in Utah in which an adjacent natural gas electricity plant will direct its excess heat and CO2 emissions into a large greenhouse complex for more rapid growth and CO2 capture. How is that not an ideal approach worthy of widespread replication? Do the rich guys who take their private jets to climate conferences know about that? Would they be allowed to talk about it if they did?
Imagine the rate of progress in a world assisted by powerful AIs that are not fed a steady diet of ideological tripe and used by humans truly free to think and innovate.
Discussion (27)
To read this bio, please join Ricochet.
First, what we call AI is not truly intelligent (yet). What they are is LLMs that are out there scanning the Interwebz for answers that likely pertain to the topic. Thus, their answers are based on what is already available and most popular. If you press them, they may look further for alternative information or numbers. But, really, the programming is not that good.
For instance, see this:
To read this bio, please join Ricochet.
Oh, and taking the question from the title, or the first half of it, "Do AIs Understand . . ." No, they do not. They do not truly understand anything.
CHARTER MEMBER
To read this bio, please join Ricochet.
Grok epic fail.
To read this bio, please join Ricochet.
I think this is correct.
When browsers were first developed, they were "trained" to scan web pages for headings and bold type--hence the instructions to résumé and advertising writers to pepper their documents with proper nouns, numerals such as years, and lots of headings. Website developers used these "tricks" for "search optimization."
My recent forays into artificial intelligence (AI) copy on Google has led me to think that the major difference, inspired possibly by the excellent Google Books search algorithms, is that the AI browsers probe the interior of documents to find phrases. They can find data that is not in display type.
I have been impressed by the Google AI's "ability" to produce results that were actually pretty close to what I was looking for.
That said, it's not in any way original "thinking" or "reasoning." It's not "intelligence" as I define the term for myself. The browsers are simply going deeper than the preceding browsers were able to do.
I'm impressed with the results though. The Google AI is an amazing search engine. But it is not, to me anyway, "intelligent." Just good at finding and matching. :) :)
What I get the biggest kick out of is Microsoft's offering in the AI market. As recently as ten years ago, MS Office was used in 90 percent of the offices around the world. Entire countries were using it. :) When Microsoft went to the "cloud" and also began to license the use of their own resident MS Office software, rather than selling a disk to the user that the user would use on his or her own premises, it meant that all those documents were being uploaded to Microsoft's cloud.
What a goldmine for reality-based AI development. And sure enough, Microsoft's AI program, originally called Azure and now called Copilot, originally beat even Amazon and Google in the AI race. Google used their email program, but Microsoft used real fleshed-out correspondence and reports. Microsoft's AI is not as advanced in relation to those other two now, but for a while, Microsoft was giving Amazon and Google a run for their money. It was really funny to watch.
Microsoft had access to everything the world was actually doing and documenting. What a perfect data environment for AI development. It was pretty funny. Who knew the little slow car at the end of the line would catch up and beat the faster, fancier cars in the race! :) :) :)
To read this bio, please join Ricochet.
I don't claim nor want to philosophize about what and AI "understands." Not really the point of the OP. I have never found much satisfaction in discussion of consciousness as an "emerging property" or tiresome analogies about physical nerve tissue and physical microchips.
Declaring that AI is not the same as full human cognition (harrumph, harrumph) misses the point that (a) it will be increasingly harder to tell the difference between human and AI responses on any topic; and (b) the power of AI technology to digest and/or create large quantities of information and perform other tasks will increasingly create dependency. My wife will not drive unless the WAZE app is running on her phone mounted on the dashboard. WAZE does not have to "understand" anything to get her to rely on its guidance in this set of tasks.
Every browser and data storage system has moved to an AI platform. How that entity is "trained" influences what it "thinks" in terms of reference materials, criteria for relevance and legitimacy. And if users are increasingly dependent, then that influence is powerful.
The Comet/Perplexity product (called by some a "Chrome killer" which is likely premature) is designed to be a comprehensive single platform for almost everything--the ultimate assistant. (I don't like talking to Alexa, I use a number of different email addresses, five separate browsers with all cookies and history instantly cleared on all but one. So giving a single entity complete access and control is not happening as far as I am concerned--although it it is probably already happening in the background by the diligently grinding algorithms at Alphabet Inc. et al)
The larger point is that as a society we are developing tools and forms of interaction that make us more vulnerable to the effects of bad info. It is almost societal self-satire that the people who most loudly decry "misinformation" are the most likely to endorse bogus narratives about almost everything. And AIs "train" by listening to them.
You're seeing five of 27 comments. Sign in, or join for just $5 per month, to read them all and add your own.