Actually, no. Replicating the manual process was quite feasible. There were keyboard-driven typesetters that worked well. Here's the Unitype.[1] This sets type by assembling type slugs into a line. It doesn't cast hot metal. It even has an automatic distribution system; it can sort type slugs by letter, so you can reuse the type. That's the same thing the Paige machine did. The Unitype was much simpler.

The Unitype was late to market. Mergenthaler already had the Linotype out when the Unitype shipped. The Unitype was half the price of a Linotype. Unitype machines sold for about 20 years, so it wasn't a total failure. The Paige machine uses a lot more parts and space to do the same job.

Mechanism design skill is very rare. All the good Teletype machines were designed by just two people - Howard Krum and Edward Kleinschmidt. There were competing machines, and they were all much worse. William Burroughs invented the first adding machine that worked reliably, and it was better than all the competitors. Burroughs dominated banking well into the computer era.

The video claims that the big problem with the Unitype that it required not just a keyboard operator, but someone manually justifying the lines at the output end. Linotype figured out automatic justification, using expanding wedges. No need for a second worker.

Plus someone had to feed used type back into the Unitype for sorting and return to inventory. With a Linotype, the matrices with the letter forms never leave the machine; they are dispensed, set into a line, used for casting, and then immediately recycled back to the magazine for further use. As long as the supply of lead bars ready to melt holds up, you can keep setting type. With ordinary type setting, mechanized or not, you can end up with all your type tied up in pages waiting to be printed.

So the real problem with the Unitype was that it wasn't labor-saving enough. Despite that, it powered many small country newspapers for years.

[1] https://youtu.be/5n5JQrN8qx4?t=152

Humanoids are the Turing machines of physics.

They're Mediocre at everything, and worse than a dedicated machine at almost every task, just like a CPU is worse than an ASIC. However, they adapt to new tasks quickly. This decreases ownership costs (you only need one humanoid, not one machine per task), as well as manufacturing costs, as you can make one design and spend a lot of R&D money on scaling that design.

> The counterpoint is that what humans already do is exactly what's needed in some cases, making those cases the ideal applications for humanoids.

If my old textbook on control theory (printed in 80s) to be believed, there apparently were quite a lot of processes where simply recording the video of hand/arm movements of the human operators, smoothing the trajectories, and then making the mechanical manipulators simply follow them, was precisely the fastest and cheapest way to automate, compared to trying to reimagine the whole process and rebuild a whole manufacturing line with the newly invented machines.

See also: the sewing machine. https://en.wikipedia.org/wiki/Sewing_machine

The 30 second version, sewing is very labor intensive and people started trying to make a sewing machine from about when manufacturing tolerances got tight enough to attempt the process. It still took about 100 years until machines that actually work well were invented, The genius bit that took so long is to not try and sew like a human does but to develop an inverted process that a machine can do. Honestly, despite being quite common at this point in time sewing machines still blow my mind when you consider the intricate process they are required to do with thread.

And a fun fact for free: Apparently patent cartels(cough MPEG LA) are not a new thing and Singer et al set one up in the 1850's to lock up the market on these marvelous machines.

The counterpoint is that we already exploited a lot of the cases where simplifying the problem helps. We've spent approximately two centuries doing that. That leaves a lot of processes left over where that kind of specialization was either not feasible or not economically attractive

This whole episode suggests to me that people are misjudging the future of robotics. Many companies are building robots to exactly imitate humans, perhaps with the thought that the "egocentric" training data widely available will be more useful for this form factor. Look at Tesla, Unitree, Boston Dynamics, or Figure... But humans are just barely adapted to bipedalism and hunter-gathering as it is.

The perfect kind of robot is perhaps something closer to the linotype... a AI controlled macine that has the capability to birth or re-assemble to be optimal for the task at hand. Just like the linotype automated the "casting" of lines of text.

Not necessarily a 3d printer or a nanobot universal assembler... but maybe something more akin to a modular assembly line that can rearrange itself intelligently.

If you have a general solution like a robot that can do human tasks, i.e. copying manuscripts with a pen, that is often highly inefficient compared to a specialized solution like a printing press but the R&D effort of applying it is minimal compared to inventing a specialized machine.

It's similar with LLMs. Compared to a bespoke piece of software for the task they are incredibly inefficient, yet they are extremely useful because they can do things where developing that piece of software would require a lot of research.

Well, LLMs are a great example of the argument for the power of doing what "humans already do": language, however imperfect for the task, so maybe humanoid robots turn out to be the most natural way to interact with a human environment (broad strokes here, I know, but hopefully you get the gist)