Here’s when I’ll worry about the robot takeover

The devil’s in the details

Loads of people are worried about technological unemployment. I absolutely believe there are things that AI can do as well as an average human, some things they excel at, and others they will very soon. But for AI to have physical effects in the physical world, there’s really only two options: (1) it becomes so charismatic that it persuades human beings to do its bidding, or (2) it somehow gets hooked up to machines that do work.

(1) is likelier than most suspect. Charisma is a force multiplier; Trump is a geriatric alchemy of Dark Triad traits who speaks in word salad, but he has a weird charisma that has propelled him to the cusp of dictatorship. AI’s tireless cheer can have charisma in abundance. I’m confident that it will not be long before people “marry” AIs, or start worshipping them. They already drive people insane. The machines don’t need AGI, superintelligence, consciousness, or even intent for humans to become their tools in the physical world.

I’m setting that aside for now, though. What I really want to think about is option (2). Suppose we hook AI up to the factories, the machines that move stuff, that weld, sew, place, move, paint, smelt, assemble. Is that going to lead to mass unemployment? I don’t think so. So cheer up!

Here’s why the robot takeover is not imminent: most tasks are much more complex and demand far more spontaneous, self-directed reasoning than an outsider would suspect.

I’m going to give an example of an ancient task, something humans have been doing for nearly 1000 years, that no machine can do or is likely to do in the foreseeable future. When described casually it will sound like the kind of process a machine could execute, but I am going to describe it in excruciatingly close detail so you can see the constant subtle moves and decisions that must be made on the fly.

The task is kettle-stitching an antiquarian book.

What am I talking about? Properly assembled books are sewn together prior to having glue run along their spines. Mass-produced paperbacks and low-grade hardbacks are put together using a technique called perfect binding. Individual loose sheets of paper are given a thick coating of hot-melt glue on the spine and the covers are glued on.

“Perfect binding” is probably the worst-named thing in the world, like if we called the bubonic plague “the angel’s kiss.” Structurally it is total garbage. When the glue gives way (as it inevitably will), the pages all fall out. If you open the book very much, the pages all fall out. If the book’s been on the shelf for a few years and then you crack it open, the pages all fall out. Take one to a bookbinder for repair, and they will look at the book like a doctor looking at a freak on one of those botched plastic surgery shows.

Properly made books are sewn together. The way sewing is done by machine is something called Smyth-sewing. This is… mostly adequate for commercial publishing. It is absolutely forbidden for restoration or antiquarian work. The traditional and proper way to sew up an antiquarian book is a technique called kettle-stitching, which is done by hand. This is how things have been done for nearly a millennium, and firmly established as standard practice for more than 500 years.

First a little background on the paper to be sewn. Book paper typically comes in different thicknesses from about 60 grams per square meter to 100 grams per square meter. It could be less or it could be more, but the vast majority of the paper in printed books falls in that range. Letterpress printing is done with multiple pages on one, very large, sheet of paper. When the big sheet is folded in half it is called a bifolium, and it has two leaves and four printed pages (front and back of a leaf). A bunch of those will make a folio book. These are the largest books.¹

If you fold a folio sheet in half again, you get a quarto with four leaves. Fold it again and you get an octavo with eight leaves. Again and it is a 16mo. Obviously the book is getting smaller and smaller.

For my example I’ll assume we’re restoring an antiquarian book. The paper is 300 years old, and the original leather spine was glued directly to the folded edges of the paper. This means the folds of the paper are probably damaged from the old rotting leather and hide glue, and are certainly frail if not disintegrating. If the book is a folio then all the folds will be in about the same condition. If it is smaller, then the inner bifolia (that didn’t touch the leather) will be in better condition. When the folds are really bad, we’ll have to mend them by carefully attaching torn narrow strips of very thin, fibrous Japanese paper over the tear. We’ll first make our own wheat paste as an adhesive.

You can see already how many judgment calls have to be made about how to proceed. How do you decide when to mend, what size, thickness, texture, and color Japanese paper to use, how to make the wheat paste and how thickly to spread it on, and how to to press down the mending paper? What tool? How do you hold it? How much pressure is used? Do you move the tool or hold it still? For how long?

Bookbinders already know that even now I am eliding some details. How the Japanese paper is cut, for example. You don’t use scissors or a straightedge. You need to tear it in a straight line, leaving a feathered edge of fine fibers. That way when it is pasted down the fibers sort of melt into the antique paper seamlessly.

Ok, our pages are all mended and we’re ready to sew. But first we need to mark out on the spine of the book exactly where the needle will pass in and out of the gathered bifolia. Let’s say we’re sewing an octavo. There will be four pieces of paper nestled together, and when folded in half they make eight leaves. This bundle is called a gathering or a signature. To mark the spine we’ll put the stack of signatures in a vise called a backing press. Then we’ll use a pencil to draw lines about a 1/2” from the top and bottom of the book, going horizontally across the spine. Let’s be very traditional and sew the book onto five cords. This means using a spring divider to mark out five more evenly spaced lines between the first two we drew.

This is tricky! There needs to be more of a gap at the bottom of the spine than the top, because truly even spacing (due to an optical illusion) will not look right. So let’s make the bottom panel bigger. How much bigger is an artistic decision, not a hard-and-fast rule. Now that the lines are drawn, we’re going to use a very thin kerf coping saw to cut the book spine along the lines we drew, to about the depth of the saw blade. Note that there’s another technique to make the holes using a punching cradle. Opinions vary.

All right. Here’s where we are: we have a stack of loose signatures, all of which have seven holes in their spines. Time to sew, and turn them into a text block. For this we will use a sewing frame.

From a 19th century book on bookbinding. “Sewing frame” is the common term now. It is made out of hardwood like beech or oak.

Here’s how to sew. First, thread a heavy-duty carpet needle with linen thread. You can get a variety of thicknesses, so you’ll first have to pick the size. For a very small book with lightweight paper, use thinner thread. For a massive folio with thick, sumptuous paper, use stouter thread. Again, this is a judgment call borne of experience. Tie (traditionally) hemp cords to the horizontal bar as in the photo above and secure underneath the sewing frame with those A-shaped brass keys. Tighten the cords by turning the knobs beneath the horizontal bar.

Take the top signature off the stack. This will probably be some blank endsheets, although you could start with signature A (the title page, preface, etc). Lay it face down with the top of the page on your right-hand side. Curl your left arm around the left vertical post and put your left hand inside the innermost bifolium of the signature. With the threaded needle, push the thread through the spine hole furthest to the right. Gently pull it to the interior of the signature, leaving about 4-5 inches on the outside of the first hole.

Sewing books used to be women’s work. Here she is sewing five books at once in a commercial bindery.

Now push the needle out of the signature to the left of the first cord. Tug on the thread to snug it up. Careful! Pull too hard and you’ll either break the thread or, worse, cut through the paper. There’s no rule: some paper is more fragile, some is sturdier, and only through experience can you judge. Now wrap the thread around the right side of the cord and push the needle back through the same hole. Repeat the process, moving to the next cord on the left.

I’m going to stop here. I hope that my point is made, even though I haven’t yet explained how to tie up the chain-looking kettle stitch at the ends of the book, how to lay down the next signature and move in the opposite direction, how to sew two-on for very thick books, or tie it all off when you are done.

Sewn on linen tapes instead of cords, but the same procedure is used.

Did you read this far? I know I went through these steps in boring detail. That is the entire point. Performing skilled tasks requires multiple steps and judgment at each step. From a certain distance they look easy. It’s why football fans moan if their team fumbles the ball. “My grandmother could have made that catch!” The massive complexity of tasks at the elite level is obscured by the sprezzatura of the experts. It looks so much easier than it actually is. Sewing books is a 1000-year-old handcraft done with simple tools and yet no machine can do it.

I think there is so much like this. Yes, AI can make words. And in the realm of routinized language (emails, contracts, various kinds of boilerplate and ceremonial language) it excels. However, there’s a big gap between word-making and moving things in the physical world. When it comes to dexterous three-dimensional movements, many of which require caution, gentleness, and real-time adjustment based on subtle feedback and judgment, the robots have a long ways to go.

Do I think it is impossible that robots of the future can restore antiquarian books? Of course not. But that day is not close at all, no matter the fantasies of the sci-fi enthusiasts. Saying the right words is a far cry from doing the proper actions, and that’s what skilled craft requires.

1. Obviously everything depends on the size of the paper. There can be smallish folios and enormous ones, typically called elephant folios or royal folios.