IDK. All this talk about unobtainium and flying cars sounds a lot like what people might have said about Microsoft in 1990. Back then, lots of people looked at computers and decided they were marginally useful in only narrowly contrived business situations or for entertaining children.
Well, yes, they were marginally useful in 1990. At that time, accounting departments were trying to use computers to track receipts and finding themselves running out of storage space because the computer only had a 20 megabyte hard disk, and needed somebody to spend $500 to put in another 20MB so they could store 2000 more receipts! And back then everything was eventually printed on paper anyway, so you basically had a better typewriter with a backspace feature that cost a fortune and didn't do much else.
At that time even the most wild-eyed futurist was underestimating the economic upheaval that was coming, and suffering from a failure to imagine what the next decade of development would bring. Nobody then could have known that thousands of successive manufacturing improvements would bring processor speeds over a gigahertz within ten years, or that the cost of memory and storage was going to plummet for two and a half decades. Similarly, few 1990 people could have envisioned business applications made up of networked computers like TurboTax, PayPal, Ebay, Amazon, or Salesforce. Back then file sharing was something you did with a floppy disk, which was no more convenient than a stack of papers.
With AI, we already have sufficiently powerful hardware. Now it's a matter of being creative enough in programming to get around a handful of fundamental limitations that reflect differences between computer processing and biological neural networks. In other words, it's more of an advancement in the art and culture of software design than the overcoming of hard physical limitations with the size of transistors, hard disk sectors, or volatile memory banks. It would seem like these software design characteristics could be easier to resolve than the engineering challenges of fitting more data onto a magnetized spinning metal plate. Software has already advanced far beyond what was imaginable in the past, even though it still uses the design paradigms of the past.
Moreover, at some level the AI can be used to design better AIs, which parallels the introduction of computers to design better computer components in the 1990s.
By 1990, a whole generation of people had been exposed to sci-fi concepts of verbal robots with consciousness. What they were seeing instead was the better typewriter that cost a fortune and had a steep learning curve. Their disappointment was as palatable as their disappointment with the 1960s predictions of flying cars and moon vacations. Yet it would prove to be wrong to lump every sci-fi plotline into the same "it'll never happen" cynicism. Computers performing tons of our work for us happened. An economic transformation and new industrial revolution happened. Supercomputers that could fit into your pocket happened, and then they became verbal assistants who could tell you the weather, tell a joke, explain a concept, or give directions.
Maybe AI is happening too.
