
“Epic” is how a lengthy article in the Wall Street Journal last week described the current investment in AI. In today’s dollars, it dwarfs the investment in the railways in the 1800s. It dwarfs the investment in electrifying America in the early 1900s. It dwarfs the investment in the interstate highway system in the mid-1900s. It dwarfs the investments in the internet at the end of the last century.
So, went the gist of the Journal’s article, it must all be an investment bubble – right? – that will come crashing down the way Pets.com and other internet stocks did.
Or didn’t. Bear in mind that Amazon, Facebook, Google and Microsoft are internet companies, too.
A competing article in the Journal last week describes how Walmart plans to manage AI. They say AI will change every job in the company – all 2.1 million of them. They anticipate substantial growth in their revenues and store count, but see their employee count staying flat. They intend to use AI to do more work without more people.
Along the same lines, the Chief Executive of Ford Motor Company said last summer, “Artificial intelligence is going to replace literally half of all white-collar workers in the U.S.”
The average person has limited experience with AI. They do know that when they need a gas station, they no longer have to type “gas station” into Google Maps. Instead, they can tell AI, “Find me a gas station,” and – voila! – it does. It’s like having a wife who can read maps!
(Ladies, please direct your correspondence to WGates@Microsoft.com.)
Several criticisms are often leveled at AI. One is that it’s great at gathering information off the internet, but its conclusions are only as good as the information it gathers. This criticism is valid. How could it not be? Like you and me, the machine is only as good as the information it relies upon.
On the other hand, the machine’s use of information is getting better and better as the algorithms mature. It is learning, for example, that quantity does not equal quality. Just because something is said many times on the internet does not make it right, and just because something is said seldomly on the internet does not make it wrong.
It makes this discernment both by considering the credibility of the information sources and also . . . [drum roll] . . . by reasoning.
That’s right, AI can think. It can look at a piece of information and say, “Nah, that cannot be accurate. It cannot be accurate that it takes days for sunlight to reach the Earth, given that the Earth is X miles from the sun and light travels at Y mph.”
In my judgment, that constitutes thinking. The machine is not specifically asked how long it takes for sunlight to reach the earth. Rather, in the course of answering the question it is asked, it rejects information that it reasons cannot be accurate.
Here’s another example of AI thinking. Already, you can give it information about a building site for a house, such as the location, the topography and the boundaries, and tell it:
“Give me some birds-eye views (yes, it will understand that colloquialism) of potential house designs for a client who likes midcentury architecture and passive solar, and wants four bedrooms and a wine cellar. Oh, and bear in mind the Building Code of Pitkin County, Colorado and the HOA rules at this address.”
In seconds, the machine will churn out diagrams of such houses. It doesn’t scour the internet for diagrams to copy; it generates its own. It becomes an architect – one with the benefit of Frank Lloyd Wright, Leonardo da Vinci, Antoni Gaudí, and all the others firmly in its “head” together with an instantaneous ability to figure out the workability of the designs it conceives.
If you want to tinker with a design, it will let you do so. You can say, “I like this one, but it’s kinda tall. Can you make it shorter and with a bigger footprint?” Or, “Let’s get into the HVAC and plumbing details on this one. Give me some schematics.”
To me, that’s high-level thinking again.
In medicine, AI already has the capability (though it hasn’t been tasked with this yet) to have on-file a patient’s lifetime medical history. A technician could say, “This patient is now experiencing sharp pain in his left-side torso and recurring headaches. What do you think?” AI might respond:
“It’s not his left kidney, because this patient had his left kidney removed in 2013. I recommend the following tests . . . By the way, be careful with poking him – he’s on blood thinners. And he’s had claustrophobia in the MRI chamber before. Note his family history of diabetes.”
To me, that’s high-level thinking yet again.
Ah, you say, that’s all just problem-solving. The machine still cannot dream, cannot feel. It knows the cost of everything, but the value of nothing.
Maybe, but the same can said of many people.
As for AI’s ability as an aesthete, I asked ChatGPT the following (with deliberate misspellings):
“Make me a 3-dimentional wall hanging about 3 x 5 feet made out of scrap steel welded together to make an abstract sculture.”
Here’s what it came up with:

I probably wouldn’t hang this on my wall, but, then again, I probably wouldn’t hang on my wall what passes for modern abstract masterpieces in museums today, either.
Now a word about the purported downside of AI – the Luddite notion that it will put everyone out of work and so we’ll all starve to death.
Economists know this is bunk. Technology certainly produces dislocations. The invention of refrigeration put thousands of ice men out of work. The invention of the automobile put millions of buggy-makers out of work. The invention of the internet is gradually putting late-night comedians out of work.
But overall, these technological wonders improve the efficiency of society – and, therefore, the wealth of society. If an invention can improve a worker’s efficiency by 50%, that doesn’t mean half the workers get laid off and starve. In the big picture, it instead means workers can get paid the same for working half the hours, or get paid double for working the same hours, or some blend of those two outcomes.
That’s what has happened throughout history in response to technological innovation. We work fewer and fewer hours, even as we have more and more things. (Whether that makes us happier is different question.)
We also live longer and longer. With AI, could we live forever?
Maybe. AI might not just cure disease and treat injury, but also stop the biological mechanism of aging.
Or AI might have the ability to receive an upload of a person’s memory – his life – before his body dies. A memory in a durable machine that can interact with humans would seem no less valid than a memory in a failing brain that increasingly cannot.
Could that AI embodiment of a person, residing on the computer cloud (maybe Heaven really is in a cloud!) continue to interact with the flesh and blood world? I don’t see why not. And what it experiences would of course add to the experiences that were originally uploaded. “You” would continue to “live.”
The AI “you” would undoubtedly be the object of real love by flesh-and-blood humans (let’s call them “humies”). After all, people routinely experience real love for inanimate objects like dolls and teddy bears and sports cars. They could surely love an image that talks with them, especially if they loved that image before its humie got buried.
In receiving that upload of a person’s memory, would the machine also receive his soul? I cannot answer that question, nor, I suspect, can AI.
