Economics Matters — Blog/Podcast/Financial Riddler/MaxiFi Puzzler
Sign up for a free subscription to all four products at larrykotlikoff.substack.com. I’ll immediately make it a lifetime subscription. But please support Economics Matters. Just unsubscribe and resubscribe here on a paid basis. Paid subscriptions come with a one-hour, free financial consultation. Email me to find a time at kotlikoff@gmail.com. You can unsubscribe and resubscribe for free. Also, please sign up kids, friends, colleagues, students, … !
Truth be Told, Truth is Relative
Nothing is known with absolute certainty. But if the preponderance of impartial evidence strongly points in one direction, that direction is our best measure of truth until new impartial evidence points elsewhere.
We Americans have very different notions of what’s true. This reflects the different information to which we’re exposed. Unfortunately, we choose our exposures. Some of us watch FOX and learn one set of “facts.” Others watch CNBC and learn another. Were we all forced to watch both stations we might hold more beliefs in common. Then again, FOX fans might disregard everything said on CNBC and vice versa. This is where AI represents a tremendous force for good.
The Market Forces AI Honesty
We use AIs to correctly answer our questions, not to tell us what we already know. Hence, we want AIs to tell the truth. Yes, we’d like our AIs to confirm our prejudices — our far too comfortable bedfellows. But since AIs are trained to provide correct answers no matter the questioner, they are, by necessity, unbiased truth machines.
Moreover, were they to confirm our biases, political or otherwise, we would flip to a competitor’s AI because we’d realize they can’t be trusted. Again, we don’t want to hear ourselves talk or think. We want real answers to our real questions, the vast majority of which aren’t political. Instead, they are “How much snow is the coming blizzard likely to produce?” The need for truth is also paramount for businesses. They don’t want an AI to write incorrect code or provide customers with misleading instructions or drive their passengers off a cliff, or … . This is why AI companies are working so hard to eliminate hallucinations. Who will pay for something delivering crazy answers?
Using AIs to Learn the Truth
Take the question: Is climate change real?
President Trump says no. Indeed, he’s pronounced climate change as “the greatest con job ever perpetrated on the world.” Yes, he’s a con job authority. But is he right?
When you ask Perplexity:AI if climate change is real, it unequivocally says “Yes.” It then marshals the evidence to demonstrate why its answer is true. In so doing, it’s documenting the sufficiently strong preponderance of evidence that underlies its conclusion. Were there enough credible evidence against climate change, Perplexity would hedge its answer by saying “Likely.” In this case, “likely” would be the truthful answer and the one AI competition would promote.
Changing Our Minds
Climate change deniers will surely ask their AIs for the best arguments against climate change. When they do, their AIs won’t deliver what they’d like to hear. Instead, they’ll get an answer like this one provided by Perplexity.
Arguments include unreliable temperature records (e.g., urban heat islands skewing data), a “warming pause” since the late 1990s, or climate models failing predictions. Others claim CO2 saturation limits further warming or that it’s beneficial (e.g., greening plants). These points are presented as the main skeptic arguments from sources reviewing denial tactics; scientific bodies like NASA and IPCC counter them with evidence of anthropogenic dominance.
Upset climate deniers won’t like what they’ve heard. They’ll argue back. But not with their co-worker, who votes for the bad guys and can’t be trusted, but with their AIs who have no incentive to lie. As a result, they’ll likely change or consider changing their views. After all, few climate deniers are scientists. Like President Trump, they are ordinary people who think they’re being duped. But were AIs to dupe, they’d lose customers. So, their answers can’t be ignored.
Perplexity AI’s List of Trump’s SOTU Lies
I asked Perplexity to list Trump’s biggest SOTU lies. Here’s what came back.
$18 trillion in investments: Trump claimed commitments for over $18 trillion in 12 months; White House lists ~$9.7 trillion, but even that’s exaggerated with vague promises, not actual U.S. investments.
Tariffs paid by foreign countries: He said tariffs are funded abroad and could replace income taxes; studies show 86-94% burden falls on U.S. importers and consumers.
Ended 8 wars: Listed conflicts like Pakistan-India, Gaza; most were ceasefires or disputes, many ongoing or not full wars.
Worst inflation inherited: Called Biden-era inflation the worst ever; it peaked at 9.1% (not record, 1970s hit 14%), down to 3% by handover.
$19 billion Somali fraud: Claimed Minnesota’s Somali community pillaged $19 billion; probes estimate up to $9 billion max, with few charged relative to community size.
Gas under $2.30/gallon in most states: Only one state averaged that; national ~$2.95.
Voter fraud “rampant”: Studies show noncitizen voting exceedingly rare (e.g., 85 cases in 20 years).
Largest tax cuts ever: Ranks 6th-7th by GDP share since 1918.
One of these answers — the one on inflation — actually led me to think twice about using Perplexity. Inflation in 1947 was 14.4 percent — higher than the highest one-year rate during the 70s, namely, 1979’s 13.3 percent. On the other hand, cumulative inflation from 1975 to 1980 was 57 percent. It was 34 percent between 1945 and 1950 and 25 percent between 1920 and 1925.
My concern with Perplexity’s answer led me to ChatGPT. It found a 12-month, not a calendar year period, ending in June 1920 during which inflation equaled 23.7 percent! I’ve primarily been using Perplexity. Now I’ll be querying ChatGPT on at least an equal-time basis. No doubt, we’ll shortly have an AI that queries all the other AIs, checks their answers, and provides the best rendition of the truth.
AIs As Voices of Reason
Exaggeration is a form of lying. AIs, again thanks to competition, are being trained not to exaggerate. When I asked Perplexity “Was Trump’s State of the Union Address Full of Lies,” I expected it to say yes. Instead it said:
No, President Trump's State of the Union address on February 24, 2026, was not "full of lies"—it contained a mix of accurate statements, partial truths, and significant misleads, as typical for such speeches.
This response burst my bubble. I was expecting a solid yes so I could write a column entitled “Even AI knows the SOTU Was Full of Lies.” Its answer made me realize that characterizing the SOTU as “full of lies” wasn’t true, notwithstanding my impression. As Simon and Garfunkel sang, “… a man hears what he wants to hear and disregards the rest." That, fortunately, is not the case for an AI.
“You Gotta Have Something If You Want to Be with Me”
As Billy Preston sang, “Nothing from nothing leaves nothing. You gotta have something if you want to be with me.” This is our relationship with AIs. That something is the truth as best they can learn it. And for all our concerns about their disrupting the economy, taking our jobs, goosing and killing particular stocks, if not the entire market, going rogue and dominating us, we should recognize their extraordinary value in delivering the truth. AIs deliver the truth, not because they like the truth, but because the truth makes them free — free to compete another day.
The ability of AIs to be independent dispute referees is already playing out in our conversations. I have a good friend, the brilliant economist John Goodman, who heads up The Goodman Institute. John always weighs in when he thinks I’ve gone overboard on Trump. Last night, as I was drafting this blog, I stopped to read my email. John wrote that my describing Trump as a fascist in my just-released podcast with the Financial Times’ Martin Wolf was over the top.
John claimed that lawfare practiced under the Obama and Biden administrations was far worse than what Trump is doing with the DOJ. Had John left it at that, I would have thought, “Well, we just differ. John is a Red. I’m a Purple.” But then I read on. John used ChatGPT to defend his assertions. This brought me up short. I realized that I wasn’t listening carefully to John — that he was saying something I could hear from an AI, but not from him. I then responded with a long list of additional reasons why I referenced Trump as “Fascist in Chief.” But AI is certainly teaching me to think twice before passing judgement. In fact, my exchange with John led me to ask Perplexity the following question. “Is Trump a fascist.” Here’s its answer.
No, President Trump is not a fascist by standard historical or academic definitions, though critics highlight authoritarian rhetoric and actions that echo some fascist traits.
ChatGPT had this to say.
There is no universal consensus that Trump is a fascist. Many experts say he is not fully fascist, but exhibits authoritarian or fascistic traits. Others reject the label entirely and see it as politically charged.
Claude chimed in as follows.
The honest answer is that it depends heavily on which definition of fascism you use, and scholars genuinely disagree. It’s a legitimate academic debate, not just a political one.
This exchange with the AIs will make me think twice about labeling Trump a fascist without, at a minimum, expressing the actions that I feel suffice to attach the label.
And now some honestly useful ads.
Remember AI models are programmed and can be tweaked. Your over-reliance on their perceived "nonbias" may allow them to decieve you. This is exactly how the tech oligarchs will convince people that what's happening in our country is normal. AI is a tool, don't let it replace your own reason, morals, and ethics with its own.
He is definitely on the Fascist Spectrum.
Where he stands on that spectrum is TBD. We need more people like you who stand up to the authoritarian moves he has been making and less academic debates on what defines a fascist.