Magic?

[Settle in; this is a long one.] I spent years writing (or directing writers of) software. The initial attraction to me, long ago, was the ability to control very precisely (and also rather “brittlely” at the time) a machine’s behavior to accomplish something, and to accomplish increasingly complex somethings. (I’ve always thought of software as “brittle” in that it can break dramatically in the presence of the tiniest flaw.)

At the beginning even the most mundane programs seemed a bit like ... uh … magic. At one point, in school, I dabbled in a bit of study of the ideas of deterministic machines (and algorithms), Oversimplifying, all the software we wrote was boringly predictable; fed the same inputs (e.g. tax rate, or shoe size, or album title) it would always do exactly the same thing. The result was precisely determined by the inputs and the software. And this was what we all wanted. We all wanted our paycheck amounts to be predictably correct.

Whenever writing this software, we know how to make it work, up front. Maybe we’re not sure how to make it efficient, or well-organized, or how to get it done in time, but the algorithms and the “recipes” are known. We might have a few ways we might do it. And of course it’s always full of bugs, even after we release it, but the feasibility is pre-determined.

When I was working in research at Bell Labs, one key directive was that we minimize spending our time building technologies that were already known or obvious how to build. This was a challenge, THE challenge, actually. But, in the end, back in the day, once we had an “aha” moment and started to invent some algorithms to carry out the aha, eventually the building effort became an exercise in determinism.

I’ve also spent hours and hours writing songs. The process is mysterious, vexing, messy, and certainly not deterministic. I talked about this in a very early blog post:

Every time I finish a song that I think is “worthy” (or pretty good at least) I worry that I’ll never write another good song again, ever. I have no idea how I did it. It’s a mystery what skill or tools or “muscle” I used to write it. I don’t mean I don’t know how to write a song – I could write ten songs a day (well maybe not the lyrics for all of them), but all ten would be worse than forgettable. They’d be unlistenable, laughable, embarrassing, cringeworthy…. I could assemble melodies and chords and rhythms (and some kind of lyrics) all day long, creating what are technically “songs”. I guess maybe I could just furiously write as many songs as possible, as fast as possible, and hope that one in every ten or fifty or a hundred would be worthy. But I’d much prefer to simply write one new song and have it be better than the last one, then repeat until I have an excellent catalog. But where is that “good song” skill? For the life of me I can’t locate it, and yet I find it fascinating that somehow I get better at it the more I work at writing songs. Somehow I’m improving a muscle or skill that I can’t locate.

The mystery and non-determinism are key to why I love writing songs – as long as I eventually create something that excites me. Then it really seems like … uh … magic.

We value the handmade, the artisanal, the one-of-a-kind. I think I read somewhere that, in the early days of mass production, consumers valued product uniformity, especially when it meant the uniform absence of flaws. And mass production has meant much lower prices. But these days, if we can afford it, we prefer handmade items. Or we will pay extra for customized products, even if in the end a machine makes them. Why?

Maybe in some way we are holding out hope that humans will prevail over the machine, or simply that we value ‘human-ness”.

A couple of years ago I had the thought that the singularity had already occurred but that we hadn’t noticed it because we were part of it, willing participants, willing contributors. So much of what many of us do is directed by, monitored or tracked by machines. This isn’t a new observation, but one way to think about it is that we’ve accumulated so much machine functionality pervading everything that we do that we may have passed into a stage where we are now just components in a vast, distributed machine-dominated existence.

We’re all cogs in a big machine
One so vast that it can’t be seen
It’s got no soul so its conscience is clean
Let’s surrender now

This “cog in a big machine” is most starkly illustrated in the gig economy. Or in the increasing use of technology to monitor employees in the workplace. And of course we monitor ourselves more and more – sleep, step counts, calories, todo lists, schedules – more and more effectively by virtue of machines that are readily available, 24/7. Not to mention “doom-scrolling”. I wonder if we’re now at a point where too much of our behavior is governed by what is called “operant conditioning” . Jaron Lanier warns of the effects in the case of social media (“…when people are acting under an algorithmic system that is designed to engage them to the max. It’s a symptom of being part of a behavior-modification scheme.”).

Love the life it promises
We’ll be using till our death
Surrender
Responsibility refused
We are both user and used
Surrender

We are both user, and used. Get it? Get it?

In addition to this relentless infiltration of the machine, something that’s a bit new is the emergence of AI everywhere. Machine-learning powers more and more of our everyday tools. A lot of us talk to “smart speakers” multiple times a day. A friend of mine is experimenting with fashion design using the AI-powered image generator Midjourney. Midjourney was also used to create prize-winning artwork. An AI-generated portrait sold for $435,000 at Christies. AI is becoming a real collaborator with humans, maybe. You can see ABBA “live” as avatars. And although “she” is not especially new, Japanese pop star Hatsune Miku is purely “synthetic” and performs in large convert venues as a hologram.

Oh and let’s not forget deepfakes, which have been used to create fictional scenes with Ronald Reagan, Johnny Carson, John Lennon, and others in Apple TV’s series For All Mankind. A.I. is now creating full-blown recipes.

Along with your search results, Google search anticipates what other questions you want to ask, then both asks and answers them. AI-powered language abilities, in particular exhibited by OpenAI’s GPT-3, are increasingly impressive. The Guardian used GPT-3 to write an article about AI’s being harmless to human beings; it subsequently published it, with appropriate disclaimers. And GPT-4 is due any day now.

What about A.I. generated music? This is also a hot area these days. Songwriter and rapper Aloe Blacc talks about his interest in composing music assisted by machine learning in an episode of Charlie Melcher’s Future of StoryTelling podcast. Check out some examples of music created in various styles by OpenAI’s Jukebox, their music-generating neural network. Following are a couple; there are more on their site:

Interestingly, the Jukebox team note that “while the generated songs show local musical coherence, follow traditional chord patterns, and can even feature impressive solos, we do not hear familiar larger musical structures such as choruses that repeat.” Writing and arranging a pop song – composing memorable melodies or grooves, deciding how many verses, whether or not there’s a bridge, the general structure, how to get the song to “take off” when you get to the hook, etc etc. – these things are really hard and challenging and mysterious, as I’ve said. And it’s quite thrilling when you get the whole song to work.  I wonder what I will think if the machine figures out how to do it. In a relatively recent interview in Wired magazine, Gary Kasparov addresses this question as it relates to chess:

When you lost to DeepBlue, some people thought chess would no longer be interesting. Why do you think people are still interested in [young chess prodigy] Carlsen?

You answered the question. We are still interested in people. Cars move faster than humans, but so what? The element of human competition is still there, because we want to know that our team, our guy, he or she is the best in the world.

It feels like we’ve moved past determinism with some of the AI generated stuff. Do these creations and capabilities now bear some imprint of “humanness”? Is it making progress in that direction? If yes, then what is the source of that humanness; is it from the software designers and the training data, or is it emerging from the machine? Does it matter? Will we have to redefine what humanness means? Is that a manifestation of the singularity?

If you’d made it this far in this semi coherent collection of musings (and I thank you for that), I leave you with this: Back in the late 1980s, when I was at Bell Labs, I wrote and recorded a song called “A.I.” The hopeful, cheerily optimistic lyrics talk about having the AI make your phone calls, do the laundry, make “nutritious, tasty meals”, choose your clothes, tie your shoes, make “playtime from ordeals”, drive your car (“you’ll never exceed 55”), be your intellectual ally in various ways, tell jokes and sing songs, provide psychoanalysis in confidence.

In the late 80s, when I wrote the song, synthesized speech wasn’t readily available as it is now. But Bell Labs had an in-house text-to-speech system used for research, and I “sweetened” my recording with the voice of the A.I.:

It’s a bit hard to make out, but in the bridge the A.I. is overheard negotiating for improved working conditions:

I believe I’ve asked you for a raise.
And I’d like to have a full hour for lunch, plus weekends off.
Also, don’t you think two weeks of vacation is rather short for someone who puts in the long hours that I do?

Bottom line, I guess, is that, with the right attitude, all of it is MAGIC.

OK, that’s it (whew), so THANKS FOR LISTENING. And tell your friends…