AI is OK
By Rick Boettger
I have some sad news. GPT 3.5 is not anywhere near smart enough to take over the world. It is more like a diligent student who learns exactly what the teacher is saying, without an original thought anywhere to be seen.
This saddens me because I thought it could be my new best friend. I ran some of my most provocative ideas by it to see if it could actually discuss them with me. This would be a real treat because I have not found any humans who can (or should I say “care to”).
My most provocative position that everyone knows about is on the Big Bang Theory. This is indeed a theory, not a fact in the class of “the sun rises in the east.” I am sticking with Steady State, aka Continuous Creation. This is a shock to my MIT friends, as I am the only person they personally know who does so.
Why? Because my extensive science training (started with Nuclear Chem at MIT, finished with a dissertation on probability, discussed below) makes me think this is the most nutso theory in my lifetime. It has led to nothing but increasingly nonsensical positions, most hilariously that 85% of the universe is “dark matter,” something mathematically demanded by the Big Bang Theory but which cannot be detected.
Study State posits that the universe has always been here, continuously recreating itself as photons become quarks in deep space. The BBT came into being purely because an electromagnet hum exists which, bizarrely in my view, theorists could only attribute to a massive explosion many billions years ago. That’s it. They’ve run from there with no helpful, predictive science coming out of it.
So I asked GPT to compare the theories, to take sides. It refused to! It just stated the theories, accurately and succinctly, bought the background radiation as dispositive, and said now almost everyone believes BBT. What a good, obedient student!
So my next provocative idea was on the economy. I got a book published in 1994 called The Deficit Lie, a precursor to what is now called Modern Monetary Theory (mine had a cybernetic component against inflation, which the current version does not understand yet). So I asked GPT to discuss the National Debt and the National Wealth.
It accurately got the easy number of the “national debt,” $28 trillion as of its last data suck. But it would not give me even an estimate of national wealth today, though it did so a week ago (??!?). It told me to check other sources, which it had easy access to—for example, the Fed says $163 trillion, the World Bank says $285 trillion. And certainly no comparison or comment.
Now to me, the fact that our wealth is 5 to 10 times our debt is why none of the scary predictions of our debt’s ills ever happens. As I for years painstakingly tried to explain to a baffled public, it is as though you have a $100,000 mortgage on a half million to million dollar house. You are not only leaving your children the mortgage, but the house as well. This wailing about burdening our children is nonsense, because of the immense wealth we are also giving them.
My next two tests are both wanky. One is my dissertation, on decision making. My dissertation created a new way of looking at probability. I asked GPT to compare Fuzzy Sets and Second Order probability theories. Here it did much better. It actually accurately described the difference, as though it had read my stuff. But it couldn’t take the next step, which is to realize that all normal probability is just a singular version of second order, with an asymptote at zero. Nor any practical application, like my math on the chances of Challenger disaster, which I calculated at one in 17, vs standard probability’s estimate of 1 in 300, based on the same data. So, again, the AI is an attentive student, with zero creative chops.
It did much worse on the second geek problem. I asked it about the relation between of the 20/80 rule (20% of the people do 80% of the work, etc) to the normal distribution. It accurately understood and named the Pareto Rule, and it understands the Gaussian function in the normal distribution, but it missed my insight, that the 20/80 rule is a mathematical artifact of the normal distribution—wow! This was exciting for me to intuit and then work out mathematically (I leave this to you, students—hint: use integrals) but it is not something that tickled the AI at all.
I next asked it to compare the Butterfly Effect to reverse causation. My IQ 200 single question test is “Discuss the butterfly effect.” If you have a 200+ IQ, you’ll point out it is a misprision of reverse causality (of course). Giving away the answer and asking it just to explain it drops the IQ measurement to around 160, and the AI couldn’t even do that. Just dutifully defined them both, la-la on any relation.
Finally, I tried to get it to take a position on the Ukraine war. It refused, saying it’s only good through Sep 2021. Okay. So I finessed by asking about the Domino Theory, which it accurately described as having been thoroughly discredited after the Vietnam War, where it had been used to frighten us to keep fighting too long. I then asked it if the fall of Ukraine would lead to Russia taking over Europe. In this answer, no mention of the Domino Theory, just like the current war enthusiasts, who warn that the Russians will be a the gates of Paris if Ukraine falls, but carefully also never mentioning what they once vividly called their Domino Theory.
So: instead of being a threat to go all Skynet on us, this generation of AI is more like an eager puppy, a teacher’s pet, who wants nothing more than to suck up the common wisdom and parrot it dutifully back, may I have a treat please.
Confession: after asking GPT 3.5 all these questions, I noticed a button on the bottom left of the screen to “upgrade.” It turns out that version 4.0 is available. But it costs $20/month. And this version is so far from being my new best buddy, I think I’ll try to live long enough to see, say, version 5.0 and buy that one. Anyone who already has 4.0, please give it a shot, and we’d both be happy to have you rub my nose in it if indeed the 4.0 is as much of a smarty pants as I am. I would so love to have a fun pet to cavort with me in my fields of intellectual delusion![livemarket market_name="KONK Life LiveMarket" limit=3 category=“” show_signup=0 show_more=0]