Tesla profit soars. And also craters

Investment Analysis Clubs Macro Economic Trends and Risks
141

Odd logic. No human worker has been replaced by a humanoid robot because we don’t yet have a general purpose humanoid robot. These are being developed.

Here’s a question for you. What is the value of AI in a machine that can only do one thing? My answer is “Not much”. Abilities like machine learning, neural networks, and deep learning increase in value as the range of potential functions of the robot increases.

A specialized robot that does one thing doesn’t have to be smart. That’s why we have so many such robots even before AI became a thing. But with AI, we now have the potential to make a single machine that can drill holes, paint a wall, make coffee, dig a hole, and clean the bathroom. Amazon will soon be able to lease a humanoid Bot that can not only move containers from point A to B, but also fill those containers, wrap packages, sweep the floor, empty the garbage, do security patrols or whatever management needs done at the moment.

Humanoid robots with human-like capabilities means we don’t have to radically change the design of our office buildings, hospitals, restaurants, stores, or factories to make things more automated. A general purpose humanoid robot with human-like capabilities means that a single base product can be used in any industry or business where the work was previously done by humans. That’s a pretty big manufacturing advantage and a pretty big market.

The correct answer is because we lack robots with the capabilities of a human. Combining AI with a humanoid form factor with human-like dexterity that can use human tools will satisfy that lack.

Not true if you consider learning through neural networks. The advantage of a robot having a human form factor increases as the task becomes more complicated. That’s because teaching a robot how to do a task is most efficiently done using unstructured visual data. It is how humans learn, by watching others do things. This is more easily done if the learner resembles the teacher.

Bet your non-human like robotic meat processor will require pretty complex coding to process a cow and will have to have major modifications in the code to then do a duck. In contrast, my humanoid robot butcher would just need to watch a few youtube videos…

1 Like
142

I just watched the landmark movie Terminator for the first time in decades.

Superb, and suddenly I realized that it had become the primary idea structure through which we view automation including both AI and Robotics.

Perhaps take a break and take a look and then come back?

david fb

2 Likes
143

Nice, but not even close. Isaac Asimov really set the stage with his I, Robot book series in the 1950s.

4 Likes
144

Yes, but we don’t have the Three Laws. So the future could look more like the Terminator than I, Robot. Remember, those who deny the existence of robots may be robots themselves.

3 Likes
145

My wife and I love that movie and watch it every couple of years, along with Point Break and Roadhouse.* She always tell the story that she came home from work one night and flipped on the TV before taking off her coat and boots. The Terminated was just starting. The movie ended and she was still wearing her coat and boots.

*The fact anyone could even consider remaking Point Break ** or Roadhouse is an abomination. I refuse to watch.

**Yes, The Fast and the Furious is technically a Point Break remake, but they changed the name and that’s enough for me.

3 Likes
146

We’re starting to repeat old ground, but just to recap:

  1. Amazon isn’t going to be able to get a robot that can do all that “soon.” Our current state of AI is nowhere close to being able to do that.

  2. Amazon doesn’t use it’s human workers that way - their pickers are pickers and their security guards are security guards. Why would it assign value to a robot being able to switch jobs like that? Better to have a robot that can do each of those jobs better than a human (a picker with more than two arms, a security guard that’s faster than a human) and be optimized for it.

Why? My non-human robotic meat processor could have exactly the same brain as yours. It just wouldn’t have legs, or be designed with the aesthetics to look like it has a separate head - just arms on a rectangular box.

Again, this is not a debate about whether an AI brain (if we can develop an advanced enough one) would improve the efficiency of robots. Absolutely it would. The issue is whether it makes any sense to bother putting that brain in a humanoid body, rather than putting it in a body that’s optimized for the kind of work it’s going to be doing. A robot that’s going to be sitting in a room breaking down meat 24/7 - and never doing anything else - does not need feet or legs or a humanoid-shaped torso or head.

1 Like
147

You are asking the wrong question. As you say, non-human shaped robots are everywhere and has been the dominant automating technology for a century now. Yet there are more tedious, repetitive, and dangerous jobs than ever for humans. You gave meat slaughterhouses as one example. This indicates that humans, with our human form factor and capabilities, are more cost effective for a lot of tasks than specialized machines. The question should really be in what jobs will the humanoid general purpose robot be more cost-effective than non-human-like specialized machines.

Observations of the real world suggest a great many as seen by the large number of jobs being performed by humans at or near the minimum wage, one definition of labor that could be considered “menial”.

The rapidly growing hospitality, health care, and elder care industries will likely be among the earlier beneficiaries of humanoid robots as these are greatly affected by a declining work force and the latter two particularly impacted by cost pressures. Humanoid bots that can fix beds, do laundry, fold sheets, serve food, empty bed pans, and clean/sterilized in environments optimized for beings that look human will significantly reduce the workload for the human staff.

Why not? Elon made a bold attempt to make his first gigafactories far more automated than legacy automakers and very publicly failed using conventional non-human-like specialized machine technology. He would probably know better than anyone the usefulness of a general purpose humanoid bot for making a Tesla vehicle.

Why wouldn’t those who actually make automated manufacturing factories know what is needed to make an even more automated manufacturing factory? Why would one expect a lawyer to be more expert in such matters?

2 Likes
148

Because Musk has a very mixed track record in being able to assess the practical capabilities of technology, and to project it in the intermediate term. As your example illustrates, he completely overestimated the state of existing tech in designing his first factories. He completely missed the state of technology for things like the solar roof and tunneling, assuming that current tech could do things that it just can’t do. So too with Autopilot/FSD/robotaxis/Tesla Network - he’s consistently failed to correctly assess where the state of his own tech was, promising (and acting upon the assumption) that the tech will be ready within a short time for the last five years or so. And he almost always makes that mistake in one direction - assuming the tech is (or will be) further along than it is.

Musk is one smart guy…and he’s also a canny businessman and a salesman. It’s no accident that their special event “Days” have shifted from Battery Day to Autonomy Day to AI Day. Even in the AI space, Dojo used to be the new hotness until it wasn’t - now Musk is talking down the odds of that being useful and noting that their Nvidia spend will be several multiples higher than Dojo. Optimus serves a valuable role for Tesla even if they never build very many, and all of these “skunkworks” projects in large companies produce far more misses than hits - so I don’t assume that just because this is something that Musk is currently excited about that this is going to be a thing that Tesla really is going to make a lot of.

1 Like
149

We joke, but that is a legitimate fear. I also wonder about the interface between neuralink, the direct control an external computer by the human brain, and humanoid bots with increasing functionality. People controlling with their mind unattached robots has potential for all sorts of unintended consequences from possible near immortality to some hybrid consciousness with heightened abilities formed by the merger of a human brain with electronic AI.

Yikes

Few want to die, but there was stability, perhaps even comfort, in knowing that one had to make do with what one was born with and had about three score and ten years to do it in.

2 Likes
150

The best surgeons often have relatively poor success rates because they tend to be asked to do the most difficult cases. That doesn’t mean one should doubt their judgement or skill.

Musk’s celebrated “failures” have generally been at overestimating the pace of of technological change and the timing of what can be accomplished. He generally gets the narrative right however. The Model 3 did demonstrate that an EV sedan could profitably compete in SUV-loving America, just later than projected. The Model Y did demonstrated that a premium EV could become the best-selling vehicle of all types, just later than projected. SpaceX did demonstrate that a private company could build a reusable rocket and provide dependable and profitable space transportation, just later than projected. StarLink did demonstrate that a private company could put up a functional global space-based internet system, just later than projected. The Supercharger did demonstrate that a single private company could in a few years create an international charging system that is rapidly becoming the de facto standard, something no one projected.

We’ll just have to wait and see how Boring, solar, FSD, robotaxies, etc work out but to be honest, I think Musk has already demonstrated a pretty impressive track record, at least equivalent to that of Steve Jobs, Jeff Bezos, and Bill Gates, when it comes to foreseeing the future and creating products for that future.

2 Likes
151

Because business activity fluctuates during the year. During slow periods. machines dedicated to a single task shut down while general purpose robots can be directed to do other things that increase productivity. Companies will do a cost/benefit analysis as to the benefits of task flexibility vs specialty efficiency and my guess is most companies will find it beneficial to have some number of general purpose humanoid robot workers. The only question is how many.

One big reason this is getting repetitive is that you won’t acknowledge my argument (with links) that it is significantly easier to teach humanoid robots complex tasks than non-humanoid variants. I think that is a significant advantage so believe it only fair that you state whether you disagree and why or agree but don’t think it important. This is relevant to you statement below:

See above about learning. We also come back to the question of whether meat processing companies go through seasonal fluctuations in activity. Would their meat processing machines really be active 24/7 all year long, or are their periods of inactivity when it would be beneficial to have a general purpose humanoid bot that can shift to other functions? One could even have a system where bots move from the processing plant after the slaughtering season to the ranches to help raise the next herd.

2 Likes
152

Okay. Yes, I disagree with your premise, on several levels. First, many jobs don’t use all of the human body. Again, using meat processors as an example, that job doesn’t require the humans to use any part of their body other than arms and hands. So a non-humanoid shaped robot that was nothing more than a box with arms and hands could learn how to break down meat just as easily as one that was shaped like a human.

Secondly, I’m not sure that you need the same body parts even for machine learning. Again, using meat cutting as an example, sure you could train an AI that had a human hand how to cut chicken both by watching the human hand hold a knife and seeing where the knife goes. But you could just as easily skip that first step - instead of a hand, just have a knife attachment with full range of motion at the end of the armature. The robot can still watch all the videos and see where the knife goes, but now it doesn’t need to learn the intermediate step of “move fingers to move knife.” Instead, it just does “move knife.”

I expect the former - most large scale plants operate year round (which is part of why you can buy meat all year round).

There are seasonal jobs in the world, of course. The machines doing those jobs might be idled out of season. But I wouldn’t think there’s enough of those types of jobs, and enough unfilled jobs nearby for the out of season work, to make it worthwhile moving those machines around rather than just idling them. There’s such an enormous amount of logistics and work involved in, say, moving some ski slope maintenance bots from Colorado in the winter to a chicken farm in Arkansas in the summer that it’s not likely to be much of a value add - and one that’s likely offset by (again) the extra cost of having legs and having to use hands instead of just a box that has an integrated knife attachment.

1 Like
153

Sure! After the shift line shuts down (meat processing plants have to clean and desanitize the lines once every 24 hours) you could have them sweep the floors. Maybe someday they could disassemble the machinery, as humans do, sterilize it (like humans), and reassemble it before the morning shift arrives - but I’m going to guess that’s a long way off given that the cleaning procedures are different every day depending on what’s happened on the line that day.

But sweeping the floor during the overnights at a car factory? Brilliant. Not exactly a high use deployment of capital, but heck, even humans sometimes take a second crappy job to make ends meet. :wink:

154

There was a meat packing plant near the town where I went to college and they were always hiring, so I knew lot of people who worked there over the years. They ran three shifts all year. If they weren’t killing animals, workers were cleaning the plant.

2 Likes
155

I’d wonder whether you put it in the body at all. Why not stick the brain on the wall and have multiple arms and hands and eyes near by doing the work, coordinating their actions.

3 Likes
156

That is brutal work. I wonder if we want a robot doing it though because it will take us further away from the food chain. Some people would probably like that, though, because they wouldn’t have to think or see where their food comes from.

Andy

2 Likes
157

I doubt its the human form factor that is important. It’s the human brain. The form factor has to come along for the ride.

–Peter

1 Like
158

People already don’t want to think about it. They just want to buy meat on nice shrink-wrapped styrofoam trays.

I bet the meat packing industry would love robots. Nobody wanted to work there long term.

1 Like
159

A childhood friend of mine during high school worked in our local meat slaughterhouse. He retired a few years ago from the Marines as a Brigadier General in charge of recruitment.

2 Likes
160

This is called Cross Training.

My nephew works for Amazon.
He’s done time in several roles. Often changing roles from one day, or week to the next, but occasionally within a day/work period.
He’s a complainer. And prefers to only do one thing. So when I see him, his conversation is about how “abused” he is by having to do several different roles.
But- Amazon has purposely cross trained him for multiple roles. So they don’t have to hire a single worker for each specific task.

  • (Hmmm. Does that sound like “Union shop”?)

I myself worked at Whataburger, a fast food place as a youngster. I cross trained for ALL positions:
Order taker, fries, grill cook, hamburger builder, drinks, cash register …
food prep, supplies ordering and storage, cold walk-in maintenance, … Nightly inventory, opening/closing, store cleaning and maintenance, … ?more?.
We (not just me, but most of us whom the boss “trusted”), cross trained for all tasks. I was often a floater - doing whatever needed doing.
I WANTED to be “multi-useful” to the boss.

Cross training was a thing at most jobs I’ve had. Cross training was smiled on. Cross training was the way the bosses identified and groomed the next bosses. Ie working the way “up the ladder”.

I worked at a local factory after the fast food career. I cross trained for all tasks. I eventually became a “trainer”, teaching others, and new hires the task for which they were hired. As people became proficient, I’d train em for the next level.
I usually did one role per shift, but sometimes I’d do several.

Later I worked in field research in environmental sciences. I was a research associate and was EXTREMELY multi-roled.
Setting up research sites, trouble shooting, assembling equipment, digging holes, cutting brush, building stuff, cinder block wiers, coding data loggers, collecting data, spreadsheet macros and data entry, ordering stuff.
Multi tasking.

So, I think a robot/s that is multi-roled might be mighty useful.

You’ve said several times that Musk making the kind of robot he-Musk thinks is best, is wrong. What kind of robot should Musk make?
If I’m the boss, I get to do me. I’m not gonna do you. Doing you, is your job.

Making legacy robots is legacy business’s job.
I get the impression that Musk doesn’t do “legacy”.
Musk looks at legacy and thinks “I can do better.”

You point out that Musk has failed.
One of Musk’s of failures is a reusable rocket that lands where he wants it to land.

Another is Tunnels that are in use.

And economically viable, functional EVs. Cause he recognized the disruptive functionality of Li ion batteries.

Oh, and structural battery packs.
Steer by wire, the new wiring bus that replaces the legacy wiring harness.
The list of not-failures goes on and on, with many many “in house designed” tech (patented) that’s NEVER BEEN DONE BEFORE".
Ie, not “legacy”.

No. Tesla products are not “perfect”.
Musk does Agile, which is "deliver a functional product that can be profitable, while continuing development toward “better” ".

I’m not sure “it’s perfect, we can stop now” is part of Musk’s vocabulary.

:slightly_smiling_face:
ralph owns TSLA shares.
While there’s lots of FUD thrown around, so far, I’m not selling any.

5 Likes