All posts by erk

I love writing, computers, puns... and can sometimes be spiritual, or mindful.
Editing, The Craft of Writing, The Writing Process

First drafts

Leave a comment

In short, writing a novel consists of the following steps/phases/whatever you want to call them, hurdles?:

  1. Come up with an idea, research etc. to get enough to know what to write about. Here, you may also want to polish that idea into a synopsis, snowflake or other “writing plan” (there are tons of different ideas out there on how to do this)
  2. Write the first draft.
  3. Edit the first draft. (+ likely more researching).
  4. Now we’ve reached the point where we either involve editors etc. for self-publishing or try to shop the book to an agent or publisher.
  5. More editing, more rewriting, more more until finally you push the “publish button”. I’ve also heard, breathing into paper bags might be involved in the before or after of pushing that button…

Not to mention marketing a novel or a career, regardless of if you are self-publishing or working with a publisher/agent.

Today, I’m focusing on step 2 and 3.

I’ve seen some interesting questions from people struggling with their first draft. The interesting thing about these questions is that they are seldom “first draft questions” in the sense they are questions about how to perfect the text in a time when perfection will be your enemy.

I once read a book (I think it was by Sanaya Roman) about how we avoid being successful by piling on enough hard tasks that it will be impossible to change.

Let’s say you decide to quit smoking. Your devious mind, all against change, suggests, if you really want to be healthy you need to eat better as well, and exercise, and hey, it’s healthy with social interactions, join a club or two, and that garage needs fixing too. Now that you’ve decided to become a better person, let’s add the garage and the garden to the list as well.

You quickly decide quitting smoking seems like a major task, and maybe you could postpone it a bit.

While writing a novel is hardly like quitting smoking, it is a change in its own way. You may very well fear what would happen if you became a famous bestseller and had fans chasing you down the street, or maybe more commonly nobody wanted to read what you wrote, and it was in fact crap even though it seems so great right now in your head.

Your mind will want to protect you against that change, just like any other change.

One way to do that is to claim the first draft must be PERFECT or you might as well quit right now.

Hence, all the questions about proper verb usage, dialog writing, theme, tension, style, etc. etc.

We worry about the details in order to prevent ourselves from finishing the first draft and risking realizing that it is crap and useless and no good and by no means ready for publication.

Now the hard truth. But for it to sink in properly, I give you a quote from Hemingway himself:

“The first draft of anything is shit.”

Hemingway via Arnold Samuelson

That’s a bit harsh, but what it really means is that in order to have a text ready for publication, you’ll have to edit it into shape after the first draft is finished.

Others have also said, writing is rewriting, and you can’t edit an empty page.

So, rather than worrying about the text not being perfect in the first draft (or God forbid, sending your unedited first draft to a publisher or agent assuming you’re no good when they won’t reply or send you a form mail back) you produce a first draft that stinks seven ways to Sunday, and then you edit it, rewrite it, and polish it until you’ve removed everything from that slab that wasn’t David.

Here’s another quote if you feel the above one is unkind to your newborn first draft:

“Every first draft is perfect because all the first draft has to do is exist.”

Jane Smiley

This, of course, is not true for the final novel, but without a first draft you have nothing to edit into a final novel.

Computers and Internet, Science and Technology

Artificial general intelligence (AGI)

Leave a comment

Will Artificial general intelligence (AGI) be created any time soon?

No.

Here’s why…

What is AGI?

Let’s start with the question of what Artificial general intelligence (AGI) really is? Google has this to say:

“Artificial general intelligence (AGI) refers to the hypothetical intelligence of a machine that possesses the ability to understand or learn any intellectual task that a human being can. It is a type of artificial intelligence (AI) that aims to mimic the cognitive abilities of the human brain.”

There are two things to note in this quote:

The concept “Artificial general intelligence“.

You have probably heard the term AI. It’s usually being used, quite generously, for things that, half a decade ago, were known as Machine Learning (ML).

So, since AI (that originally was supposed to be used for intelligent machines) now is used for the ML equivalent (for sales purposes) the intelligent machine must therefore be called AGI instead.

I’m sure, when the sales logic dictates it something like “Artificial real general intelligence” or “Artificial general intelligence for real this time we promise” will be used instead of AGI that would then be used about some fractionally more capable ML.

AGI is supposed to mimic the cognitive abilities of the human brain. Mimic the human brain.

A roadmap to AGI

In essence, there are two things we need in order to create an AGI or super-intelligent AI:

  1. A correctly designed digital brain
  2. Sufficient input (learning data) to that brain

We’re unable to do both of these today. Let me explain…

What do we know about brains?

I’m in a unique position since I’ve studied both computer science, AI and psychology, and I’ve tried, and failed to write a paper incorporating computer science and psychology. Mostly because the institution where I tried to do this (a CompSci one) had zero knowledge of or experience with psychology.

When I see software companies creating digital “brains” I am always reminded of this experience, and I’m always seeing this total lack of understanding of how the human brain works.

Don’t get me wrong. Nobody knows how the human brain works!

No, really. Nobody!

Here’s what we know about the human brain:

We know how a neuron works. How synapses and axons work. How signals are likely being sent and received, but get this: We’ve never seen a neuron in an undisturbed brain in real live action doing what it usually does.

Not to mention a whole network of them.

We know different parts of the brain does different things.

This has been measured using something called functional magnetic resonance imaging (fMRI). This measures the blood flow in the brain when the person being scanned performs different tasks, thus making it possible to designate areas of the brains as different centers (language, movement of the body, tactile skin sensations of the body, vision, etc.)

Finally, there have been experiments where the brain of a fruit fly was mapped.

This was done by slicing the brain in ultra-thin slices, photographing them, feeding them into a computer (ML software) and building a 3D map of the brain.

What does this tell us about the brain? We know where things happen, we have a pretty solid theory on how the smallest component, the brain cell, works we also have a hopefully rather correct, map of the brain… of a dead fruit fly.

We have no clue what happens on the neuron level in a live, undisturbed brain. We cannot, like in computer programming, step through the different signals and stages taking place in a brain when something happens.

With this “vast” knowledge of the human brain, we’re about as likely to understand how it works as we are to understand the internet by knowing how the smallest component in a CPU (a NAND-gate) works and that different parts of the computer whirs and clunks when we do different tasks on it. Oh, and we also have the circuit board of a 1980ies calculator to assist us in our understanding of “TikTok Memes”…

But AI companies are working on it!

Sure. There are some invention going on when talking about artificial neural nets and ML/AI research. (Though most significant endeavors seems to center around adding more artificial neurons and making the models slimmer, faster etc.)

This research has been going on for almost 70 years now.

Nature has been evolving brains for over 500 million years.

And what we have right now isn’t even able to understand simple objects and that they are a cohesive whole. To understand what I mean, ask an AI to produce sources for its statements. Most of the time those sources will be wrong, or will not support the statement they are supposed to support. This is because the AI does not have an object understanding of articles. And they are generally bad at sources anyway.

I don’t doubt that researchers will be able to create a digital brain that will be able to do as good as or even better than a human brain, but I think it’s rather silly to imagine this will be done tens of millions of times faster than nature did it.

Sure, give it a century or two… or at least until we do actually know how the human brain works… but trying to create a digital brain when we don’t know how the organic version works… that’s something only people that have no clue we don’t have that information would ever dream of trying.

Language is not meaning

Another thing worth noting with respect to the efforts of AI companies to develop AGI by creating larger and larger LLMs (Large Language Models) is the very nature of language. Words and sentences does not contain all information about the world. They refer to memories of experiences we’ve had as human beings, and they get a lot of their meaning from this.

Take a simple example:

"The ball slipped out of the child's hands and bounced down the staircase."

You get an image of what is happening here, how it’s happening, and what some consequences of this could be. But not because you’ve spent hours reading up on balls, bouncing, staircases, and childhood traumas related to balls going where they shouldn’t.

You understand this sentence because you’ve seen balls bounce down staircases, maybe even had a ball that bounced down a staircase where you lived, maybe even had it break something as it reached the end of the staircase…

An LLM/GPT reading everything it can about balls and staircases could never experience this sentence like you do.

All the text on the Internet: a month of visual data…

Let’s look at this problem from another angle.

Did you know that today’s LLMs are trained on almost all available text on the internet?

There are a few problems with that, like what do we have on the internet, on social media platforms and on and on? Hardly objective data. So an AI trained on this will hardly be objective. Nor will it be able to do much original thinking…

But it’s worse.

Comparing textual data (what LLMs are trained on) with visual data (one type of data a human child would encounter and use in order to understand the world) we can quickly see the differences:

“In a mere 4 years, a child has seen 50 times more data than the biggest LLMs trained on all the text publicly available on the internet.”

We can draw the conclusion that text is very light data wise and an AI that would even have a slim hope of gaining a human understanding of the world would have to use not only internet text, but video, audio and even tactile senses.

In effect, it would have to be a robot, living among humans in order to become human like.

This requires some developments in robotics. Give me a robot cat, as dexterous as a real cat, and as light weight, and maybe it’ll be able to understand humans. Unless we condemn it to live in our mobiles.

Robots of today are hardly that agile or light on the foot. They are more like your grandpa on steroids. Though there are some quite interesting developments of “Super dogs”:

To be honest, though, they still seem arthritic… only faster…

Will there ever be AGI?

Here’s a thought experiment:

Let’s say you’re a forklift manufacturer, and you’re looking into AI to create an autonomous forklift. You want to give it a room with shelves and have it do all the heavy lifting from moving actual packages to organizing and inventorying the space.

Pretty much, you want to be able to just leave stuff off on the front of the storehouse for it to organize in there or ask for it to bring stuff out you can pick up from the front of the storehouse without having to bother with how that happens.

This type of product would cost you research and development, however, if I told you, just add another 10 million (or 10 billion?) dollars and I can make your forklift critique Dostoyevsky or do childcare or collect stamps?

Why would you want all that, and why would you pay extra for it? You want a forklift and the better if it isn’t conscious or self-aware or have a ton of extra interests. It would be cruel and unusual to lock a living being into that warehouse for its whole life!

People tend to see AGI as this humanoid robot (or digital human) that can jump into any kind of situation and solve any kind of problem, but they don’t realize that’s going to cost extra and my prediction is that it’s going to cost so much extra very few companies or venture capitalists will be willing to pay for it.

And at the other end of AGI rests the nightmare of dealing with a self-aware conscious being. Can you really treat it however you want? Does it have rights? Will it become a digital slave? What if it wants to unionize?

I wouldn’t want to pay an extra million or billion to buy me those kinds of problems when all I wanted was a more autonomous fork lift.

The problem with AI

The biggest problem with AI isn’t that it’s going to take over, out-smart us, or necessarily steal our jobs.

The biggest problem with AI is that we think it’s super smart and give it tasks it cannot handle.

We’re already doing that:

So while AI can be a threat or a promise, I think the human level AI will not be around for a long time. Though, AI (or ML) is here, and it already impacts our lives and if we don’t understand what it is and what it isn’t we risk giving it more power than it should have.

I sincerely hope decision makers do understand that the people peddling “almost humanlike intelligence” have no clue how human intelligence actually works.

Nobody does.

Header image: By El contenido de Pixabay se pone a su disposición en los siguientes términos. Pixabay, CC0.

Creative Writing, Editing

A Nod Every Second Page

Leave a comment

In the second draft of my current WIP, my characters nod about once every second page.

Statistics

I found this out by looking at statistics in Scrivener. The second draft is 233k words (yes huge, I’m working on getting it down to somewhere closer to 150k right now). Of those, about 575 words are nods of different types. Nodding is in fact the most common word in the WIP.

However, this is statistics. How bad is it really on the page? For me, having been around the text for years, that may be hard to figure out. For a beta reader, they may get irritated by the text without really understanding why.

It is, nevertheless, important to separate statistics from the actual experience of the text. Maybe this is no problem at all? Or is it a subconscious showstopper? Maybe every person reading the text will see nothing but the nods. And heads will be rolling? (Pun intended…)

Regardless, looking into fixing some of these nods should improve the text, even if not taking care of them at all might not kill the text.

So how do we take care of the nods?

Fixing Repeated Words

The simple answer to getting rid of all the nods is, of course, to replace them with other actions, and to replace with many different actions.

That’s simpler said than done, of course, but there are help out there.

I mulled this over for some months and realized I already had the resource to fix this on my computer.

My solution is spelled, “The Emotion Thesaurus,” by Angela Ackerman and Becca Puglisi. There may be other resources like it out there, but this is invaluable for this problem.

However, and there is an important however, you need the PDF-version to make this work. (Though the PDF-version is really cheap—$7 as of this writing—so I suggest you get it).

You need the PDF-version because my solution to this problem is to search for “nod” (or whatever other thing your characters do) and work backwards to find synonym actions. More about this below.

The Emotion Thesaurus

The Emotion Thesaurus is a list of emotions (anger, fear, happiness) defined by their corresponding physical signals and behaviors, internal sensations and mental responses. There are also sections on long-term responses and signs the emotion is being suppressed. Among a few other things.

Replacing Nodding

As I stated above, in order to come up with alternatives to nodding, I searched my PDF-copy of the Emotion Thesaurus for “nod” and a list of sections (emotions) came up.

There seems to be a use for “nodding” in a large proportion of emotions, so the next step is to try to determine what the character is really feeling in every scene where there is nodding.

In one scene, my POV-character comes back from a negotiation, having to tell her boss (the president of a far future sci-fi nation) that they didn’t get all they bargained for, and he’d been pretty mad about that before, even calling them traitors. So here’s the original:

He pressed his lips together but then nodded. "We knew it was a tough starting bid."

Things that seem to fit in this scene are:

  • Acceptance
  • Determination
  • Indifference
  • Resignation

I did a first, quick filtering out of things like admiration, gratitude and self-loathing since it was not at all what the character felt at that moment. Then checked each potential entry for what it was about (not being a native English speaker sometimes require some effort on that front) and how the nod was being used to figure out if it was a good replacement or not.

Nodding Because of Acceptance

In my case, I decided the best emotion was acceptance. The thesaurus had just “nodding” as one physical behavior for acceptance.

The president was accepting that the bid didn’t get through, deciding it wasn’t really that big a deal. My POV-character thought as much, and she too was happy to see him giving up that notion.

The text had already been down the road of the president getting pissed on the negotiations, and besides, he had other things on his mind in this scene, so… acceptance seemed to be the best emotion.

For acceptance, there are a number of possible physical actions that could replace nodding:

  • One’s shoulders and torso loosening slightly as tension ebbs
  • Taking in a cleansing breath
  • A smile that grows
  • A light tone of voice
  • Open body posture (arms away from the body, legs slightly apart, chest out, etc.)

I picked the first one but shortened it down to the person just relaxing.

He pressed his lips together but then relaxed. "We knew it was a tough starting bid."

The text already contains a piece of dialog equivalent to the character accepting the situation, so the nod was only there from the start to add some body to dialog. Now I get the same result with another type of physical response instead.

Sidenote: Yes, that sentence actually shows and then tells, in a way. Maybe in a future round of editing, I’ll decide to change the dialog to something else. I’ve come to realize, doing editing, that my texts will likely always have issues even if they get printed. I think I’m suffering from perfectionism…

POV-Characters

If the person nodding is a POV-character, there are also internal sensations like the chest loosening or feeling lighter and mental responses like being cautiously optimistic.

Of course, had this been a POV-character, I might also have used a thought or two on the matter. Without thoughts or dialog, things would otherwise have become pretty confusing. But also adding a physical response adds to the section and for a non POV-character dialog, action or body language is pretty much all there is.

Final Words

Finally, I’d like to add that right now I am not editing the WIP for nods. In fact, I am working on its size.

Editing for word usage will, in my current plan of editing (that I hope to be able to post about some day) be the final step of polishing among several steps going from an overall analysis of the text (is it even readable?) through the structure of the text (distribution of acts and plot points) and characters to looking at scenes and finally details like paragraphs, sentences and words in the very end (where I may have had to remove both scenes, plots and characters because they didn’t work—it’s easier to kill a darling if you haven’t spent months coddling it…)