Scary Monsters, aka Physical Artificial Intelligence: Five Months’ Progress with Robots

This AI subarea is not only one of the oldest, but the most graphically visible.  And, in honor of our publication date, the closest to monsters.  But how good are they really?

We start with “Delivery robot autonomously lifts, transports heavy cargo” (Kurt Knutsson, Fox News, May 26^th).  So you don’t want to pay those mid-six-figure incomes to union dock workers, or even way over $100k to UPS package handlers?  This technology, LEVA, can “load and unload cargo boxes without any human help” by “securing the box,” then lifting “itself back up and” carrying “the load to its destination,” handling up to 187 pounds and dealing with stairs and “rough terrain” on the way.  Alas, nothing here about availability.

Dealing with a current problem, “John Deere addresses farm labor shortages with autonomous tractors” (Sophia Compton, Fox Business, also May 26^th).  Although they have been made in at least prototype form for almost four years, we can’t tell from this article if, or when, you can buy one.

More clearly in the present tense is “Humanoid robots handle quality checks and assembly at auto plant” (Kurt Knutsson, Fox News, June 24^th).  “Kepler Robotics has officially introduced its Forerunner K2 “Bumblebee” humanoid robot at the SAIC-GM automotive plant in Shanghai… in a recently released video, the K2 is seen moving confidently through the plant, performing detailed quality checks, and handling assembly operations that demand both strength and precision.”  It “can load stamped parts, manipulate mechanical fixtures, and adapt to new tasks using a combination of imitation and reinforcement learning.”  In addition to “tactile manipulators with an impressive 11 degrees of freedom per hand, and flexible fingertip sensors that boost its dexterity,” it “leverages a cloud-based cognitive system that enables it to learn new tasks quickly and coordinate its movements with full-body awareness.”  Fundamentally, largely because of AI, better than the industrial robots of decades ago.

Continuing along more general but similar lines, from the same author and source, “Job-killing robot learns at work, and it’s coming to the factory floor” (July 5^th).  Although the previous example would match the title, this time it’s Hexagon’s AEON, also “humanoid” and designed for “handling repetitive and error-prone tasks,” which allows “raising the bar for productivity and workplace safety.”  It uses Microsoft Azure and “Maxon’s advanced actuators” to get “remarkable agility and dexterity,” along with “spatial awareness,” and its “intelligence grows over time thanks to a self-learning loop.”  But no availability information, and described by Knutsson as “new.”

Did you know that “There Are More Robots Working in China Than the Rest of the World Combined” (Meaghan Tobin and Keith Bradsher, The New York Times, September 25^th)?  That was “more than two million… in Chinese factories last year,” per the International Federation of Robotics, with 300,000 “new” ones.  Charts of “annual installations of industrial robots,” one for China and one for “rest of world,” show the second one way ahead from 2015 to 2019, somewhat higher in 2020, and behind from 2021 to 2024.  Because of a “national push,” “over the past decade, China has embarked on a broad campaign to use more robots in its factories, become a major maker of robots and combine the industry with advances in artificial intelligence,” and now has “five times as many robots working in its factories as the United States.”

This time’s stunning speed achievement is from a product that “is a collaboration between Crest Robotics and Earthbuilt Technology, “Australian construction robot Charlotte can 3D print 2,150-sq-ft home in one day using sustainable materials” (Kurt Knutsson, Fox News, October 12^th).  The author described that as “the speed of more than 100 bricklayers working simultaneously.”  The material it uses “comes from sand, crushed brick and recycled glass,” forming “a structure that’s fireproof, floodproof and created with a far smaller carbon footprint than traditional building methods.”  Its “future versions” could even build “moon bases for research and exploration.”  Until then, though, it needs to become available, as it, representations to the contrary, “may be years away from building its first full-scale home.”

Apparently in progress now, though, are “The Robots Fueling Amazon’s Automation” (Karen Weise, The New York Times, October 21^st).  In that company’s “most advanced warehouse in Shreveport, La., employees touch products at just a few stages, such as taking them out of shipping boxes and placing them in bins,” whereupon “the Sparrow robotic arm looks into a bin of items, picks the one it wants and puts it in another bin,” sometime after which “the robotic arm called Robin places packed packages on a small robot called Pegasus, which shuttles packages to drop down specific chutes depending on where they will be shipped.”  After that, robots sort them and “autonomously” take “them to shipping docks.”  No doubt there will be further improvements, and perhaps more steps automated.

How can it be that “Robots power breakthrough in pregnancy research, boosting IVF success rates” (Angelica Stabile, Fox News, October 23^rd)?  The automata “assist in the (in-vitro fertilization) lab,” which incorporates a great deal of other AI, by “preparing specialized plates to sustain embryos,” at which they are “10 times more precise in preparing (them) than humans.”

These are wonderful developments.  My only gripe is that, as so often happens in the information technology field, the difference between current and projected future obtainability is often blurred.  Is it fair to credit a product with being able to do something, if it has not been publicly rolled out?  What steps remain for the manufacturers of, in the cases above, LEVA, robotic tractors, AEON, and Charlotte to sell their products commercially?  How much low-error time after that would it take for the makers, and us, to declare their product productively deployed?  We don’t have much for answers to any of these.  Until we do, or see successful robotic sales and deployment, we should limit credit to the Amazons, Bumblebees, and IVF robots.  The other companies can see why – if they can get there, we’ll do the same for them.

Electric Vehicles – Almost One Year’s Telling Stories

It’s been the quietest year of several for electric vehicles.  Are they settling down, or just reacting to changing governmental policies?  How predominant, if at all, will they be late this decade and beyond?

To set the tone for 2025, we saw “Tesla Annual Sales Slip for First Time as Competition Grows” (Jack Ewing, The New York Times, January 2^nd).  They “fell slightly in 2024” for “the first annual decline in the company’s history, as rivals in China, Europe and the United States introduced dozens of competing electric models.”  Total Tesla deliveries were off about 1% to 1.789 million, and, as of press time, “still accounts for nearly half of all electric cars sold in” America. 

The next was even gloomier, as Ivan Penn asked if “Electric Vehicles Died a Century Ago.  Could That Happen Again?” (The New York Times, May 26^th).  The author’s reasons for concerns were that “The Trump administration and Republicans in Congress are working to undercut the growth of electric vehicles, impose a new tax on them and swing federal policy sharply in favor of oil and gasoline.”  He mentioned that “the oil industry has enjoyed numerous tax breaks,” but electric cars, starting with their now-discontinued buyer subsidies, have as well.  He focused on EV’s being less “macho,” but there has been much more than that to why “electric cars may be                                                                         in trouble, at least in the United States.”

“The EVs We’ve Lost” (Wired.com, July 19^th) told us that “shifts in economic policy and manufacturing have led major automakers to cancel upcoming electric vehicle launches in the US.”  Whatever it is, consulting company AlixPartners “dropped its 2030 sales predictions for battery-electric and hybrid card by a whopping 46 percent compared to last year’s projections.”  With that, the following, many of which have been absorbing money for several years, will not be produced: Ford Three-Row EV SUV, Honda Five and Seven-Seat EV SUV, Mercedes-Benz MB.EA-Large Platform, Nissan and Infiniti EV Sedans, Volvo All-EV Lineup, Maserati MC20 Folgore, Apple Car, and Fisker Pear.  These are still in progress, but believed delayed: Buick EV, Ferrari EV 2, Lamborghini Lanzador, Lamborghini Urus, Porsche 718 EV, and Tesla Model 2.  That’s a lot.

Something healthy, and good for electric car buyers, is that we are seeing “Used E.V. Sales Take Off as Prices Plummet” (Jack Ewing, The New York Times, September 13).  In contrast to delivery numbers above, “sales of used electric vehicles rose 40 percent in July from a year earlier, according to Cox Automotive, a research firm.”  Those too, though, were subsidized, with customers “rushing to take advantage of a $4,000 tax credit that can be applied to used electric vehicles that sell for $25,000 or less.”  Used sales prices will be a good indicator of how highly EVs are desired by people who don’t already have one, which is perhaps obvious but reveals valuable information about the extent of their market.

As September rolled to a close, we got the judgment that “Electric Vehicles Face a ‘Pretty Dreadful Year’ in the U.S.” (Neal E. Boudette, The New York Times, September 29^th).  The author, backed up by analysts, expected that the end of various federal tax credits that month would cause sales figures “to plummet in the last three months of the year and then remain sluggish for some time,” as that and other industry developments show “a stark turnaround from the heady days a few years ago when many automakers believed electric vehicles were poised to take off.”  Additional manufacturer cancellations named here included Honda’s electric Acura, Stellantis’s “battery-powered” Ram pickup, and importing of Nissan’s Japanese Ariya electric SUV.

It was time for another writeup on “How Much It Costs to Drive an E.V. and a Gas Car in Every State,” and, courtesy of Francesca Paris and the October 8^th New York Times, we got one.  It, however, considered only fuel cost, so anyone serious about this issue will need to assemble and properly interpret data on depreciation and other expenses.  This study, though, found that charging or filling up for 100 miles ran averages of $5.26 for home electricity, $6.15 for hybrids, $12.80 for “standard” gas cars, and $15.62 for “fast charging.”  Factors mentioned for possible individual consideration were “cheaper electricity rates at night or for E.V.’s,” home charging when power comes from solar panels, regional electricity-cost differences favoring some west-of-the-Mississippi states, different gasoline prices, and differing fast-charging rates.  The states with the cheapest, relative to gas, home-charging prices were all in the West, with the most expensive five all in New England.  When gas was compared with fast charging, the most favorable to EVs were Florida and four in the Pacific, while the worst were scattered: Arkansas, Wyoming, the District of Columbia, Vermont, and Maine.  It is noteworthy that New England, which culturally is one of the areas most favorable to electric cars, has the most expensive electricity, and the mountain states of the West are opposite in both ways.

What overall?  Even without the subsidy losses, electric vehicles, in the United States, were not poised to become the norm.  They seem solid as a minority preference, but that’s all we, automakers, legislators, and presidential hopefuls should expect.  They have more gyrations to go through before we know just how large a share they will command, but it won’t be a majority.  On that the data, tangled though it may seem, can agree.

What’s Happening with Driverless Cars, Good and Bad

Although from a late-teens perspective autonomous vehicles haven’t done much of anything, as I have reported they are succeeding in several cities.  What else has been going on with them?

In what should be good news, “GM restarts driverless car program more than a year after Cruise robotaxi incident” (Greg Wehner, Fox Business, August 11^th).  In a one-off event well before their decision to leave, “a Cruise Origin robotaxi… struck and dragged a woman about 20 feet.”  The automaker claimed here they’re “accelerating the development of autonomous driving technology capable of operating without human oversight,” and per Bloomberg will “be focusing on developing driverless cars for personal use instead of for a robotaxi service.”  As the firm’s “sources… reportedly” said, “the first steps should be to develop hands-free and eyes-free driving with a human inside the vehicle, but ultimately the company is working to have a car that can drive without anyone at the wheel.”  That sounds like returning to where they were, if probably incorporating improvements the taxis have discovered.

Soon thereafter, we watched as “Stellantis hits the brakes on Level 3 autonomous driving tech over soaring costs” (Nora Eckert, USA Today, August 26^th).  That wasn’t defined in the article, except that it “enables drivers to have their hands off the wheel and eyes off the road under certain conditions,” which “would allow them to temporarily watch movies, catch up on emails, or read books.”  That sort of thing seems scarier than it did even years ago, and indeed was “never launched,” but the company “stopped short of saying that the program was canceled.”  Clearly an indefinite delay.

Per Charlemagne in the September 6^th Economist, such technology is important enough that we can call the continent’s slow pace with it “Europe’s Sputnik Moment.”  Robotaxis, which are “starting to feel humdrum in Guangzhou or Phoenix” remain perceived as “science fiction in Warsaw or Rome,” as they are “barely being tested” there, and exemplify “how far the continent has fallen behind” and how “Europe has become too dependent on China and America.”  However, this interpretation is unfair, since cities with robotaxis have new road and highway systems and generally fine weather.  I have heard nothing about them being planned for New York or Boston, in which self-driving vehicles would fare little better than in the much older cities across the Atlantic, and as well have more people walking and using public transportation.

“The one thing that’s free in Las Vegas – but it requires taking a gamble” (Deirdre Bardolf, Fox News, September 21^st) is a ride on a Zoox robotaxi, provided by Amazon.  The vehicles, which distinctively look like “toaster(s) on wheels,” have been available for just over five weeks, before which they progressed from serving “company employees” to helping “friends and family members,” before opening to “anyone with the Zoox app.”  At press time, Zoox was “collecting rider feedback, testing its user interface, refining its pickup and drop-off infrastructure and working to gain the public trust in driverless transportation.”  All strongly positive, even in a city with conditions, as above, unusually well suited to robotaxis.

Forbes, though, printed something called “Tesla’s Full-Self Driving Software is A Mess.  Should It Be Legal?” (Alan Ohnsman, September 23^rd).  In order for company CEO Elon Musk to get “his jaw-dropping $1 trillion pay package,” he must put “1 million Tesla robotaxis on the road and 10 million active (full self-driving) users over the next decade” – a tall order for technology described as “error-prone,” as during an hour-and-a-half Los Angeles test it “ignored some standard traffic signs and posted speed limits, didn’t slow at a pedestrian crossing with a flashing sign and people present, made pointless lane changes and accelerated at odd times, such as while exiting a crowded freeway with a red light at the end of the ramp.”  One observer called it ”just a prototype” and said “it’s not a product,” yet it stays as “driving-assist systems are unregulated.”  The previous paragraph gave; this one took away.  A week later, we saw “Two US senators urge probe of Tesla’s Full Self Driving response to rail crossings” (David Shepardson, Reuters), in response to “a growing number of reported near-collisions.”

“When a Driverless Car Makes an Illegal U-Turn, Who Gets the Ticket?” (Michael Levenson and Laurel Rosenhall, The New York Times, October 1^st).  Two policemen in San Bruno, California, “saw a car make an illegal U-turn right in front of them,” but “a ticket couldn’t be issued,” since, although “California approved a law last year allowing the police to cite autonomous vehicles,” it isn’t in force yet, “did not specify any penalties,” and “citation books don’t have a box for ‘robot.’”  Indeed, “there are no clear rules in California,” although “Arizona has a state law that allows the police to issue traffic citations to driverless vehicles, just as they would to regular drivers.”

What can we make of this motley collection?  One takeaway is that robotaxis, when in their carefully chosen environments, are doing superbly.  Another is that elsewhere they are not, erring with such as obeying signs that I would have thought the software’s 2010s closed-course training would have long resolved.  A third is that those programming and implementing autonomous vehicle technology need to change some things they are doing.  Until the results improve, driverless software will be limited to robotaxis and, with warnings to them to never stop paying attention, in cars with drivers.  Could all that substantially improve?  Maybe in a year, maybe not in ten.  Don’t bet on it – unless the odds you get are good enough.

Artificial Intelligence’s Gigantic Financials, and the Effects They are Having on Others

A great deal has recently been announced about the often-staggering sums of money associated with AI.  What have we heard, and what does it mean?

On August 27^th, Tripp Mickle saw “Nvidia Sales Jump 56%, a Sign the A.I. Boom Isn’t Slowing Down” (The New York Times).  What was, at least as of that date, “the most valuable public company in the world,” and had the month before reached $4 trillion market value, sold $46.74 billion from May through July with “profit” of $26.42 billion, the latter up 59%. 

On the same date and in the same publication, we read that “The A.I. Spending Frenzy Is Propping Up the Real Economy, Too” (Lydia DePillis).  “Companies will spend $375 billion globally in 2025 on A.I. infrastructure, the investment bank UBS estimates,” and “that is projected to rise to $500 billion next year.”  Per the Commerce Department, “investment in software and computer equipment, not counting the data center buildings which have “overtaken office construction,” accounted for a quarter of all economic growth this past quarter.”  The boom has especially helped building materials companies, along with “electricians, engineers and heavy-equipment operators.”  The sector is so strong that “the most significant constraint on data center growth is more likely to be supply:  The energy, water, workers and technical equipment required to construct and run them are all getting more expensive.”

The Economist, on September 13^th, took a warning stance with “The $3trn bet on AI,” saying that “even if the technology achieves its potential, some people will lose their shirts.”  The piece, though, did not focus on the near-certainty of some companies losing their market share or their perceived potential, but asserted that even “in the rosiest scenario,” many shareholders “would face big losses,”; if worse, “the flow of capital could slow; some startups, struggling under the weight of losses, could fold altogether,” and “a lot of today’s spending could prove worthless,” as “more than half the capex splurge has been on servers and specialized chips that become obsolete in a few years.”

One area less considered is “How Wall Street’s Big Bets on A.I. Are Driving Interest in Huge Parking Lots” (Patrick Sisson, The New York Times, September 16^th).  A company, Gray, “building 22 data centers,” reports that each one requires “a space nearby that’s large enough to store millions of dollars’ worth of tools, generators, and tractors and trailers, in other words “a couple of acres of gravel or asphalt near highways, ports and other shipping infrastructure.”  That is nothing completely new, as e-commerce already “requires huge spaces to park inventory, shipping containers and the vehicles used in last-mile delivery services,” meaning that rents for suitable industrial outdoor storage facilities are up, an average of 123% in the past five years, and “owners of anything that resembles” such properties, such as “old truck stops and auto repair yards,” “have been hearing more from brokers.”

In response to differing perceptions, Cade Metz and Karen Weise told us, in the New York Times on September 16^th, “What Exactly Are A.I. Companies Trying to Build?  Here’s a Guide.”  Those mentioned in the article, which prints out to 13 pages, are “A Better Search Engine,” “Tools That Make Office Workers More Productive (and Maybe Replace Them),” “An Everything Assistant,” “A.I. Friends,” “Scientific Breakthroughs,” and “A.I. That’s as Smart as a Human, or Smarter.”  As I have documented, instances of these objectives have already succeeded, and companies expect many more.

Have you wondered “What Wall Street Sees in the Data Center Boom” (Ian Frisch, September 20^th, additionally in the Times)?  Analysts, right or wrong, consider that “data center capacity has become a barometer for both the health of the tech market and the risk of an A.I. bubble,” so they have been enthused about AI for most of this year.  Yet there are causes for concern, namely that “even if A.I. proliferates, demand for processing power may not,” “some worry that costs will always be too high for profits,” “it’s not just Silicon Valley with skin in the game,” and, as we will see, “the stakes extend beyond finance.”

As Schumpeter revealed in “AI’s $4trn accounting puzzle” (The Economist, September 20^th), depreciation rates have an enormous impact on industry profitability.  Not all companies agree on “the longevity of all those fancy AI chips they are installing,” especially now that Nvidia has “said it would unveil a fresh AI chip every year rather than every couple of years,” and the useful lives of servers have proven controversial.  Using reasonable assumptions, the author estimated that if “the entire AI big five” set server depreciation at three years, “their combined annual pre-tax profit would fall by $26bn, or 8% of last year’s total,” which could “amount to a $780bn knock to their combined value.”  Serious business those accountants are doing.

Since the most effective way for businesses to assure that their customer relationships will stay good is to buy those they sell to, it was no surprise that we can expect “Nvidia to Invest $100 Billion in OpenAI” (Tripp Mickle and Cade Metz, The New York Times, September 22^nd).  That will be “part of a wider effort among tech companies to spend hundreds of billions of dollars on A.I. data centers around the world.”  Yet that raised the question “Is A.I. Investment Getting Too Circular?” (Andrew Ross Sorkin et al., The New York Times, October 7^th).  Do such deals
“raise questions about the robustness of the artificial intelligence boom”?  Possibly, if, as a “prominent short seller” put it, “Don’t you think it’s a bit odd that when the narrative is ‘demand for compute is infinite,’ the sellers keep subsidizing the buyers?”  That is a good question, and the answer could be that nobody else, even in 2025, has enough money.  After all, when “deals with AMD, Nvidia, Oracle, CoreWeave and others promise to give the ChatGPT maker more than 20 gigawatts of computing power over the next decade, roughly equivalent to that of 20 nuclear reactors,” and “the electricity needed to support that compute could cost about $1 trillion,” they’re getting way up there.

An idea to consider from a Yale Law School professor and Budget Lab president there is if “There are Two Economies: A.I. and Everything Else” (Natasha Sarin, October 6^th, also in the Times).  In a year when the American population has been bifurcated into rich and not rich, and split ever more sharply into pro-Donald Trump and anti-Trump sections, it is nothing too strange to suggest that related economic forces have done the same.  Money spent on AI-benefiting capital “may reach 2 percent of the gross domestic product in 2025,” a 20-fold three-year increase, and it alone may be pushing national economic growth from 1% to “almost twice that.”  “There are signs that the non-A.I. economy is under duress,” shown by problems with inflation and jobs, and “it’s possible that other parts of the economy are being held back by A.I.’s dominance,” as opportunities in that area may be hogging capital.  If indeed “the A.I. boom is masking Trump’s policy blunders,” we may need it to succeed even more than we think.

Overall, artificial intelligence is cutting new channels in our economy wider than any before.  The courses they take will be different.  Its outcomes and major impacts inside and outside that field are still very much unknown.  Sometimes, all we can do in such circumstances is to hang on for the ride and stay away from the banks as much as we can.  That is where we are now.

No AJSN Today Per Government Shutdown

 I will publish September's American Job Shortage Number within a couple of days of receiving the Bureau of Labor Statistics' Employment Situation Summary and its supporting tables.