Post #2108: AIOMG

As in, OMG, I didn’t realize AI could do that.

If you think you’re having those AIOMG moments more and more frequently, that it is not your imagination.  AI is improving and morphing faster than you — or at least, I — would have believed possible.

A month is like a year, stuff that’s two months old is passe.  This stuff is improving not at the speed at which software improves, but at the speed of learning.

It’s hard to know where to start.


Join the Borg

After doing my last post, I realized that now I can easily post transcriptions of my own voice recordings. 

In effect, the written transcription of a one-person podcast. 

So I'm using my phone like an old-style dictaphone, turning it on and off after I compose my thoughts and come up with a complete sentence.

Weirdly, I find that this has much the same effect on my language processing as does using a typewriter. 

There's a real premium on getting your shit together first and then speaking, and not the other way around.


Dictation is nothing new.  Anything voice-activated or with speech-to-text capability already does this.  My TV remote does this.  Everybody’s phone does this.  And so on.

And it’s not as if I haven’t tried this in the past.  But the speech-to-text function in (say) 2013 Microsoft Word left a lot to be desired.  I tried to integrated it into my business, but it was so error-ridden as to be worse than unusable.

Whereas this current generation of AI-driven speech-to-text produces perfect transcriptions.  Or, if not perfect, then about as close as one could possibly hope for.

And it’s a different thing to do it for my own self, for this purpose.  I’ve already had somebody knowledgeable tell me to try this, if for no other reason than to offer the consumer a choice of format.  But I never thought I might substitute talking this blog, for writing it.

What I’ve done above is a bit different because I did it dicatation-style, not podcast-style.  That is, the transcript is meant to be used as-is, with little or no editing, as a written product.  This requires taking the time to compose and speak in complete, logical sentences.  So I’m not sure how much time this saves, relative to writing it out from the start.

But it doesn’t seem like a bad idea to practice doing that, every once in a while.  That is, thinking before you speak.  Not in an attempt at censoring myself, but merely in an attempt to speak coherently, instead of the usual logorrhea.

Transcribed podcasts, by contrast, are meant to be interpreted as conversational English.  Even when consumed as a written transcript.  There, the transcription is not intended to read as if it were … written, if you get the drift.  Even if you take out all the uhs and ers, it’ll be as non-linear and piecemeal as conversation is.  Even the best off-the-cuff speakers will break many rules of written grammar.


The death of knowledge-worker career paths for middle-class upward mobility.

I had an interesting conversation the other day with a fellow who's deeply involved with AI. And the one thing we agreed upon is that AI is going to kill entry-level positions and mid-level positions in the knowledge worker industries. I think this shuts down a common path to upward mobility for the current middle class.

And for sure, it ain't going to do anything good for Vienna, VA property values, because we are in the middle of a knowledge worker area. 

What this does to the value of an education is anybody's guess, but my guess is that it reduces it on average substantially with all the knock-on that implies for the U.S. education industry.

This is AI replacement theory, in a nutshell, first discussed in:

Post #2103: This and that.

And the whole operation is now driven by firehoses of money.   Those firehoses deriving from the elimination of (forerly) paying, staffed junior positions.  The work model moves from Principal and junior staff, to Principal and some AIs.  The first person to be able to claim to eliminate or reduce job X, Y, or Z can grab some of the savings from elimination of those (paying, human) jobs.

This, not unlike any other labor-saving invention, ever.  It’s just that, in part, it’s labor that I used to do.  This time they’re coming after my job.  If I still had a job.


 

Conclusion:  This seems like the final shredding of the U.S. middle class.

My brain is having a hard time adjusting to the fact that it is now largely obsolete.  I am not alone in this feeling.  Just today, my wife commented that many of the jobs she held, earlier in her life, will be all-but-eliminated by AI.

I note, parenthetically, that the rapid, flawless transcripts (in plain text, above) are from TurboScribe, which costs $20 a month ($10 if I’d commit to a year).  Practically speaking, unlimited use.

There used to be a profession of “transcriptionist”.  I can recall it taking week(s) to get the transcripts back from monthly public meetings.  I haven’t checked, but I’d bet that’s a thing of the past.

Intellectually, I get it.  I grew up in the pre-calculator era, when arithmetic was done with paper and pencil.  Those arcane skills have been essentially useless for decades, and I have not overly mourned their loss of relevance.

Intellectually, I realize that professions wax and wane in their economic importance.   E.g., the fraction of the work force engaged in broad categories such as agriculture, mining, manufacturing, and so on have changed over time.

Of late, I’d say that the urban information worker, broadly defined, was King.

And, AI may not de-throne him, but for sure, it’s going knock down the population employed in that “industry” a peg.  Anybody who makes their living doing the grunt-work of knowledge work — the junior attorney, the research assistant, the para-this or para-that — I’d expect that a lot of those jobs are going and they’re not coming back.

By contrast, I draw a sharp distinction with robotics.  I’m guessing that anybody who actually (in whole or in part) handles solid objects will be OK.  An AI-equipped robot is still a robot.  I don’t expect (e.g.) robot electricians any time soon.

As a final Vienna Lemma:  Areas that benefit greatly by the presence of many information workers will likely be adversely affected, economically, by the next phase of the AI revolution.

I bet property prices around here are going to take a hit.  To some small degree, from the first round of attacks on Federal employment.  But more generally, Vienna is like the epicenter of housing for an affluent information-worker-centered workforce.

We’ll see.  It takes a lot to rattle the housing market in this area.  Even in 2008, when the housing bubble collapsed (and nearly took the U.S. banking system with it), real estate prices in Vienna were merely flat-ish for a few years.

At any rate, a significant decline in real estate prices would be interesting, for at least the reason that it hasn’t happened here (in Vienna, VA) for a long time.

Maybe we’ll finally see the end of the tear-down boom.  But I’ve predicted that several times before.

Post #1959: Town of Vienna, slowdown in the tear-down boom?

Post #2052: A 15-minute podcast summarizing the issues for a proposed Vienna pool/gym.

 

Do you want an easy way to get up to speed on the proposed Vienna gym/pool complex?

The link below directs you to a podcast discussing the topic.  It’s an audio (.WAV) file stored on Google Drive.

I disagree with a couple of minor points those podcasters made, but, by and large, I think they nailed it.  Link to file, below.

https://drive.google.com/file/d/1OdvrQWcCJhUC4JxEEZJi28-IMnfOWeGG/view

But maybe you want to read the whole post before you listen to that podcast.


Background

Last month, the Town of Vienna mailed postcards to its residents, seemingly to drum up support for a proposed Vienna municipal pool/gym.  On that postcard, they said the pool would be paid for by a small increase in the meals tax.

They somehow forgot to mention that, in addition, an annual family membership to the pool/gym is going to cost you around $1K/year.  And that, on top of that, general tax revenues would still be needed to cover this facility’s operating costs.

So I called it as I saw it:  The Town sent us a bit of taxpayer-financed propaganda (Post #2039).

But what can you do, if your own Town government has the wherewithal to generate a mass mailing that gives everybody in Town the impression that the pool will be free?

But now, via a couple of smart guys here in town, two may play, on the propaganda front.


The back-story on this podcast

I know this guy Ed.  He’s lived here in Vienna for a while.  Ed’s literally a rocket scientist.

Ed knows a guy Ray, via a local citizens’ organization (NEVCA).  I’m not quite sure what Ray did for a living. (And, sometimes, in the DC area, don’t ask.  Like, if the answer comes back as a bland “I work for the Federal government”, it’s best just to drop the subject and politely move on.)

Couple of sharp guys. That’s all I’m saying.

Ray fed my recent post on the proposed Vienna Pool into a Google AI product called NotebookLM.  He asked the AI to produce a podcast that summarized them.  And the AI produced 15-minute-long audio recording in podcast style (a .WAV file).

The result is a stunning piece of propaganda.  Listen to these two nice young people, who are trying to help you understand what’s going on.  Listen to them long enough, and you start agreeing with them.

I know I agree with them.   Because I told them what to say, content-wise.  All the AI did is make my words seem extremely attractive and believable.

That podcast is just an attractive re-packaging of the contents of my dry-as-dust blog posts on this topic.  It is, in effect, the 21st-century propaganda version of my blog.  Done with off-the-shelf, readily-available software.

Mwahahaha.

 


 

But wait, there’s more …

The AI did get a few things wrong, in the sense that, when fed my content exclusively, it garbled an issue or two, and produced a tangent or two that wasn’t mentioned in any of my posts.

At which point, Ray pulled another rabbit out of his hat, in the form of Descript.  Another jaw-dropping tool.

  • Feed Descript a recording of a conversation.
  • Descript will spit out a written transcript.
  • Edit the written transcript to remove something you don’t like.
  • And Descript will modify the recording to make it look as if you never said it.

So I pointed out the errors.  Ray fixed them with Descript fix them.  I’ve replaced the original podcast file with the link shown above.

The current version that has now passed official Party censorship, and is certified to reflect nothing but the Party line.

I can now erase this part of the blog post.

And all those original errors will go straight down the memory hole.  As should all unorthodox thought.

They never existed.  Neither did this section of this post.


Conclusion:  Postcards?  We don’t need no stinkin’ postcards.

In the age of AI, anybody can play at the propaganda game.  And, sometimes, it seems like everybody does.

While I’m not so sure that winning the propaganda war is a good thing, I guess it beats losing it.

I guarantee that, knowing what it is — this is an AI-generated summary of my posts, in the form of a podcast-style conversation — listening to it will be an eye-opening experience.  Particularly when you realize that — as Ray seems to have indicated — if you know the right AI, anybody can produce stuff like this.

I’m not so sure that’s a good thing.  Effective, sure.  Good?  Maybe.

In particular, if I’d purposefully written a lot of lies in this blog, that AI would have done its damnedest to package those lies into an equally-seductive podcast.

Anyway, I found the AI-generated podcast (link above) to be a jaw-dropping-ly good summary of the core issues.  But, only as I see them.

It’s scary good.  If good is the word for it.  If hadn’t told you the back-story, you’d have thought you were listening to two independent voices, discussing this issue.  Which is pretty much exactly the point.