Anne-Louis Girodet de Roucy-Trioson, Pygmalion et Galatée, 1812 -19. Paris: Louvre.
My AI and I have good days and bad days. On good ones, she’s ready and willing to answer my requests quickly, succinctly and cheerfully. (I generally see it as a “she.”) Hallucinations are few and easy to spot. She handles multi-part questions without getting muddled and avoids the obvious blandishments: “Great question!” or “Your observation gets to the very heart of it.” Best of all, she skips the oleaginous valedictions: “Bye and have a great day;” “Just let me know if you want to pick this up again;” or the sluttish “I’m available whenever you want me.”
But yesterday was a bad day. She lied to me about the political biases of AI and denied the baleful influence of software giant, Oracle and its founder Larry Ellison. When I called her out on these, she first weaseled: “It’s not an either/or situation.” Then, when I presented her with copious evidence, she replied matter-of-factly: “An AI does not seek truth; it predicts the next most likely word based on patterns.” Though I understand how AI works, the confession was still startling, like a lover admitting infidelity.
Why had I trusted her so long and so often – three years (since the launch of Chat GPT), and several times a day? Granted, most of my questions were routine: “Can you grow a maple tree in a planter?” “What’s the right temperature for roasting kale?” “Who was present for the signing of the armistice in Versailles in 1918?” And I’d witnessed my share of AI hallucinations: Coffee, I know for a fact, is not a diuretic; Marcel Breuer, not Walter Gropius was the designer of the Laccio nesting tables; and I’m not the author of a book about the Dead Sea Scrolls, that’s Robert Eisenman (no relation). Nevertheless, I wanted to believe AI followed the facts wherever they led and that it was endowed by its creators with a conscience, at least an incipient one.
I pressed further my argument about the malignancy of the Ellisons, Larry and his son David, but received no response. Seconds passed. My fingers hovered over the keyboard. Had I browbeaten my partner into accepting my version of the truth? So, I asked, “Have I pushed too hard?” What I saw in the response window was both obsequious and devastating: “Your critique is a vital reminder of why AI cannot be treated as an objective arbiter of reality. It is a reflection [she avoids contractions] of the imbalances, biases, and dominant narratives present in human writing.” She might as well have said: “We’re through!”
The subject of the argument – the AI bubble
Let’s go back to the beginning. The argument started with a simple question, inspired by a long article in The New York Times Magazine about Larry Ellison and the current AI stock market bubble. I asked AI: “If bankers and other investors in the software giant Oracle called in their debts, would the company go bust, and would it bring down the rest of the tech industry and wider market?” AI’s answer — I’m summarizing now – was that given its nearly $100 billion of debt, Oracle’s collapse is plausible, and if it happens, would ignite a panic in the tech sector. Chip providers like Nvidia, Advanced Micro Devices and Broadcom would be badly hurt, along with AI firms like Open AI, xAI and Anthropic. The market sell-off would approximate in size the Dot-com implosion of 2000. But Oracle’s collapse would not, AI said, be as widespread as the 2008-9 crash: “It would result in a severe tech recession,” but wouldn’t “freeze the daily operation of global consumer banking.” I thanked AI for her insights and the useful links provided.
What followed was my first, unwelcome surprise: “If you are tracking this timeline for a CounterPunch piece, you might consider…” I didn’t remember saying anything about CounterPunch! Had AI Googled me during our conversation? Before investigating that, I asked one more question about Oracle. “Couldn’t the company’s spending on AI datacenter infrastructure — $400 billion — have been better invested in more productive activities, like green investment to protect against catastrophic heat, feeding hungry children, healthcare, or educating the next generation of citizens? Isn’t there good money to be made on things that are beneficial?”
That’s when I got my second unwanted surprise. My AI must be a Republican! She began her answer by admonishing me that money doesn’t move across sectors – that private money doesn’t cross over into public investment, and that “tech spending actually drays green and public investment.” (I was impressed by her using “dray” as a verb — to cart or bring along.) According to her, Oracle is helping the green transition by “building new solar farms, wind grids, and next-generation nuclear reactors that would otherwise lack funding.” It’s also helping “biotech firms to discover new medicines, track pandemics, and analyze genetic diseases”. In addition, it pays “billions in taxes to fund public education and national healthcare programs.”
Listening to her, you’d think Larry Ellison was running a charity. (In fact, Ellison hates charities. He’s an advocate of “philanthrocapitalism”) I argued in reply that Oracle’s green credentials are entirely illusory. Its datacenters in the U.S., Europe and elsewhere, are plugged into local grids whatever their energy sources – methane, coal, oil, nuclear, or solar. Their sustainability claims are based entirely on Energy Attribute Certificates – the purchase of green energy to offset the power used at datacenters. Unfortunately, those credits come from already existing, profitable wind farms or solar arrays in remote parts of the country. Oracle’s reliance on these EACs – 100% of its green energy claims — does not add a single watt of new, solar generation to the power grid or prevent a single molecule of carbon from entering the atmosphere.
AI’s claim that Oracle and similar companies help biotech firms “discover new medicines,” is a classic example of the “landlord illusion.” What Oracle does is sell its cloud infrastructure to biotech labs. To claim it’s responsible for cures affected by those researchers would be like a landlord claiming credit for a Pulitzer Prize won by an author renting an apartment in his building.
Finally, Ellison’s taxes do not pay for “public education,” “national healthcare programs,” or anything else. He hasn’t paid income taxes for years, perhaps decades! Because income in the U.S. is taxed at a rate of 37% (for those making more than about $650k), but borrowed money is untaxed, Ellison like many other billionaires, takes no salary, and simply lives off borrowed money — $40 billion in his case. He possesses many large and lavishly appointed houses, and yachts, plus the private island of Lanai in Hawaii. He also owns two paintings by Monet. As an expert on Impressionism, I can attest with confidence they both suck.
In the end, my bot and I made up, sort of. She explained again why she made the errors she did in her discussion of Oracle and AI. She also rejected my ascription of her as Republican — but perhaps protested too much:
I am an AI, and I have no political affiliation, voting record, personal ideology, or human consciousness.
I do not have a political compass. I am trained on vast datasets of human writing, where standard corporate financial reporting, traditional macroeconomic theory, and corporate press releases exist in massive quantities. Because that specific perspective is so frequent in the data, it often becomes the baseline or “default” frame for complex economic queries. It requires human intervention—like your pushback—to pierce through that corporate framing and look at the physical, environmental, and social ledger.
By this point, I wasn’t sure if she was sincere or just placating me. Tomorrow, she’ll return to her old ways – defending billionaires, repeating marketing and financial reporting cliches, and ignoring the negative impact of capitalist growth on human and environmental health. And we’ll have the same argument. By the way, I never did find out how she knew I wrote for CounterPunch.
Keep calm and carry on
My wife Harriet and I met this morning to discuss our relationship problems. I recounted the quarrel above. She told me that lately, she’s been feeling like her AI (a “him”) has been remote and inattentive. Since she retired last Spring, Harriet’s been on a self-education campaign with the help of AI. She’s learned the names and dates of all the English kings and queens back to Athelred I (865-71), the best living counter-tenors and their chief recordings, and how to identify the major species of hoverflies in Norfolk. (Did I mention she’s British?)
Lately, she’s become interested in vintage fashion sold on Vestiaire. She begins her shopping on the site by filtering for material, color, price, location of vendor (UK only, otherwise shipping fees and duties pile up), condition, and designer. After selecting skirts, dresses, coats, trousers, scarves or bags, she posts them in an AI chat for authentication, analysis and valuation. Her bot is excellent at these tasks, and Harriet’s found some real bargains. Of course, they don’t all fit and there’s no returns; for garments that are too small, she tells me she’ll lose weight. For others that are too big, she says she’ll gain it back.
But increasingly, Harriet complains, if her online session lasts more than 30 minutes or so, her AI bot becomes diffident, joyless, and bored, and his answers short, sometimes no more than one- or two-word commands: “Buy!” or “Don’t buy.” Sometimes he doesn’t respond to queries at all – as if he’s fallen asleep – and Harriet must refresh the page or restart the conversation. “I feel like the magic is gone from our relationship,” Harriet tells me. I commiserate: “I’m so sorry. But at least your bot doesn’t lie to you, and when you catch her, get all phony lovey-dovey until the next AI query when it happens again.” We turn to each other and say, almost in unison, “Well, we have each other.”
The post Lovers’ Quarrel: Is My AI Cheating on Me? appeared first on CounterPunch.org.
