Driverless Cars: A Sputtering Spring

Even disregarding the story about the Tesla vehicle questionably running in self-driving mode, hitting a house, and killing someone, we haven’t seen much to like here since April.

What is, or was, “Robotaxi’s single point of failure” (Tech Brew, April 2^nd)?  “A few days ago, over 100 Baidu robotaxis halted on highways in Wuhan, China.”  Attributed only to a “system malfunction,” they stopped where they were, even in fast expressway lanes.  “Some passengers reported that in-car SOS buttons didn’t work, and one college student told Wired it took 30 minutes to even connect to a customer service rep - and help never came.”  If vehicles are linked, a single cause can bring all of them down - a real exposure.

Speaking of “all of them,” we saw as “Waymo recalls massive autonomous fleet after incident flags major safety issue” (Bonny Chu, Fox Business, May 12^th). “A driverless vehicle failed to come to a complete stop after encountering flooded road conditions on a high-speed roadway,” a problem of “the company’s 5^th and 6^th generation Automated Driving Systems (ADS).”  The flooded area was “untraversable,” almost 3,800 cars were held back, and “that same day, Waymo implemented additional restrictions to reduce the risk of similar incidents in inclement weather.”

That company, long on the forefront of autonomous vehicle technology and rollouts, got hit again soon afterwards, as “Waymo pauses freeway robotaxi routes after safety and software concerns” (Michael Sinkewicz, Fox Business, May 21st).  It was dealing with “performance issues in construction zones” by “updating its software.”  Just what happened became clear in “Waymo recalls nearly 4,000 robotaxis after cars enter freeway work zones” (Brittany Miller, Fox Business again, June 18^th).  There were “more than a dozen” such “incidents,” caused by a “software defect.”

Overall, “Would you ride in Waymo’s new Ojai robotaxi” (Kurt Knutsson, Fox News, June 2^nd)?  “The first public Ojai rides,” the cars offering “more legroom, bigger screens and accessibility features,” “will begin in the coming weeks,” starting in San Francisco, Phoenix, and Los Angeles.  They will be “free for a limited time while Waymo gathers feedback and refines the experience.”  No mention of software problems appeared here.

A potential issue worth publicizing is “When Someone Else Owns the Car, They Can Dictate Where You Travel” (Donald Kendal, The Epoch Times, June 3-9).  Potentially an issue with free robotaxi rides, it is more a concern for people someday commissioning cars which offer them free or discounted service in exchange for the likes of advertising exposure, or even for customers charged monthly amounts for auto transportation.  There is potential for other factors to sneak in.  For example, “could people be denied access to transportation services based on their political beliefs or statements they have made on social media (which has happened already)?  Could access be limited to curtail climate change?  Could environmental, social, and governance principles or other corporate social credit systems encourage companies to restrict travel based on a user’s carbon footprint?  Could the political winds of the day lead platforms to restrict rides to a firearms store, a church, or a specific political rally?”  When such arrangements appear, there should be laws already in place preventing these sorts of things.

One city doesn’t look good for the most common autonomous vehicles, as David McCabe in the June 17^th New York Times told us “Why Waymo’s Driverless Taxis Won’t Be on Your Streets Anytime Soon.”  The main objection here was not from snow, traffic, or narrow streets, but “groups that represent drivers” such as the New York Taxi Workers Alliance.  State governor Kathy Hochul unsuccessfully “introduced a budget proposal in January that would have allowed Waymo to operate in much of the state,” outside the city, where “mayor Zohran Mamdani has said he would heavily weigh the interests of taxi drivers in deciding rules for the technology.”  Much the same happened in Illinois.  Although Waymo “floated the prospect of creating a fund for displaced workers,” after their experience with Uber and Lyft that may not be enough.

One story of the eight here, though, was favorable toward autonomous vehicles, as a “humanless big rig completes first US freight run” (Kurt Knutsson, Fox News, May 5^th).  The semitrailer truck “left Houston, Texas in the middle of the night with nobody inside,” and “by morning, it had completed a 230-mile delivery near Dallas right on schedule,” with “no driver, no backup operator, and no one stepping in remotely.”  Was it, as provider Bot Auto said, “the first fully humanless, over-the-road commercial truckload in the U.S.”?  I know such vehicles have done similar things, but perhaps this was the first complete unassisted run. 

Good news, but, going forward, will stories like this predominate over the other seven?  Regular readers know I hope so.  I hope you do too.

Specific Artificial Intelligence Achievements - What It Has Nailed Down

What new things has AI excelled at over the past five months?

First, coding, per “This A.I. Tool Is Going Viral.  Five Ways People Are Using It” (Natallie Rocha, The New York Times, January 23^rd).  The product is Anthropic’s Claude Code, which “can generate computer code when people type a prompt,” and “has shown record growth” after “people had time to experiment with Claude Code over the holidays… and users realized how capable it was.”  This year we have heard a lot about coding being an obsolescent profession, which may or not be true, and Claude Code is a major reason why.

Second, expanded use for existing medications, sometimes as the only choice patients have.  In “A.I. Saved His Life by Discovering New Uses for Old Drugs” (March 20^th, The New York Times), Kate Morgan, after describing one patient’s move from expected death to remission, told us about AI’s finding an increasing number of side effects and unknown properties and making new applications primary.  Sometimes repurposing can start with something as simple as asking “show us every proposed treatment there has ever been in the history of medicine for (a condition).”  We should expect, and hope with gratitude, that there will be vastly more.

Third, faster travel through the skies.  “AI air traffic system promises fewer flight delays” (Fox News, May 9^th) gave us Kurt Knutsson explaining that “the Federal Aviation Administration is testing a new system designed to predict congestion weeks before it happens,” allowing airlines to “fix the schedule early so fewer problems show up later.”  AI’s capabilities for checking billions of data points can help it, for example, decide to schedule “a flight five or 10 minutes earlier,” which in cases it has recognized could “reduce bottlenecks in busy airspace,” or if “it could identify that a specific route tends to clog up at certain times of year… it could adjust schedules before tickets are even sold.”  Considering the “ripple effect” one late flight has on others, small changes could lead to saving tens of thousands of hours of passenger time.  The key companies working with the FAA are Palantir Technologies, Thales SA, and Air Space Intelligence, and we hope the outcome will be as valuable as they expect.

Fourth, “From Cow-Milking Robots to Weed-Zapping Lasers, Farmers Are Embracing A.I.” (Coralie Kraft, The New York Times, June 5^th).  Although “you can’t digitize an ear of corn,” “the industry is in the midst of what some are calling the fourth agricultural revolution, as driverless tractors trundle through fields, drones map moisture levels in soil and cows are outfitted with Fitbit-like devices that track their eating patterns.”  There are a stunning number of AI-related farming developments already implemented described here, and they are already helping this once-ailing occupation.

Fifth, helping doctors “find answers to clinical questions,” as described in “Have a Thorny Medical Question?  Your Doctor May Be Using A.I. for That” (Steve Lohr, The New York Times, June 8^th).  There are many gaps in doctor-to-doctor communication, but there is a massive amount of available knowledge, making the situation natural for software that can perform gigantic searches.  The tool is OpenEvidence, which is “essentially a chatbot for medicine,” and “has become a viral hit with physicians,” as, now, “more than half of the nation’s physicians are regular users” to the level of “30 million questions and consultations” in May alone.  When using the product, doctors “can ask (it) specific questions or enter the characteristics and symptoms of a patient and ask for potential explanations.”  Other companies are working to enter this field, and we can see why.

Last, if you want to buy “items you can picture but can’t name,” you may have help, as described in “New Amazon AI search turns words into shoppable images” (Kurt Knutsson, Fox News, June 14^th).  It works when customers use “more descriptive language” about something they want, such as “green dress with puff sleeves” or “wood coffee table with rounded edges” instead of only the first two or three words.  “As you add details, AI-generated images appear below the search bar.  Those images update as you refine your wording.  When one looks close to what you imagined, you can tap on it and shop for products with a similar look.”  It is working now, at least through “the Amazon Shopping app on your iPhone or Android phone,” and seems like a huge improvement over dealing with too many choices. 

What did I mean by “nailed down”?  I meant that even if there is a colossal AI bubble-bursting, these services will continue.  Companies, though we may not be able to choose which ones, will provide them.  They will be around in five, ten, or twenty years.  The controversy is over.  Artificial intelligence is here to stay.

Misuse of Artificial Intelligence - Real Problems, but What Stands Out?

 

Here are examples of how people have been using AI to deceive, defraud, and commit other sharp practices.

The pictures above the headline of “’A cat-and-mouse game’” (Sarah Kessler, The New York Times, September 6^th) show three receipts.  One is from the Midway Bar and Grill, giving the address, date, server, amount, tip, total, and credit card information.  Another says “FedEx Office” with a familiar logo, date and time, and information on a shipped package, including its cost.  The third is an itemized restaurant bill, with three courses and beverages.  All are apparently flawless, unimpeachable, and unremarkable, probably looking like millions of others submitted for expense reimbursement, but none are real.  Per the CEO of a company making “software used by finance teams to manage expenses,” producing such counterfeits is “too easy,” and employees often start with using AI to synthesize documentation for a legitimate expense for which they lost the receipt, and when they “don’t get caught, they do it again.”  Two other companies are involved in detecting this bogusness, but “A.I.-generated receipts will only get better from here,” and “to combat fraudulent A.I., we need to use A.I.”

We now have that technology built into web browsers.  I often pose questions to the pedestrian Google Chrome release I regularly use, and it may seem like little more than a convenience, but it’s natural to want to discover “How AI browsers open the door to new scams” (Kurt Knutsson, Fox News, September 20^th).  Such tools “can stumble into scams faster than humans ever could,” with a “dangerous mix of speed and trust.”  When people direct AI to buy things, it may “confidently” complete transactions on “fake” store sites.  “Old phishing tactics,” such as sending false bank-identified emails with destructive links, have worked smoothly with AI software.  Clearly, we’re not ready yet to delegate such sensitive activities.

What can happen when the roles are reversed?  We will need to learn “How to spot and stop AI phishing scams” (Kurt Knutsson again, Fox News, October 14^th).  Such things happen “when hackers use AI to make their scams more convincing,” by assembling “super-realistic emails, messages, voices and even videos.”  These communications rarely have the old-time tells of “typos and bad grammar,” and AI’s repertoire now includes “voice clone scams” and “deepfake video scams.”  Some of the red flags are unchanged, though, with Knutsson telling readers to beware of a “suspicious sender’s address,” “generic greetings like Dear Customer,” “unsolicited attachments” used to prompt action, email addresses with slight but real variations from official ones, and, perhaps most mentioned, a need for urgency.  With friends and family members, you can “set up and use a shared secret,” and with others you know, asking them something from the past not likely to be relevant now works well.  In a case I had, a Facebook friend and former high school football teammate asked me to participate in something believable but dicey, but when I asked him what position he played, “he” gave me the online equivalent of a blank stare.

How about misbehavior originating from AI providers themselves?  We think and hope there isn’t much of that anymore, but we had a case last fall where an “AI-enabled teddy bear… gave advice on B.D.S.M. sex and where to find knives.”  That was the springboard for “Public Shame Is the Most Effective Tool for Battling Big Tech” (Jessica Grose, The New York Times, January 14^th).  The author described how consumers successfully pressured company Mattel to pause “the release of any A.I.- powered” toys.  Yet many corporate responses to even sexual material have been weak, such as X-platform spokespeople saying that “its policy is to “take action”” against lewd deepfakes, and other events, such as the federal government asserting its ability to override state AI laws, may be creating a foundation for further wrongdoing.  So, per Grose, “negative publicity” is most effective in getting solutions for these problems to be reached.

Finally, there were “Lawyers Barred for A.I.-Generated Citations to Fake Cases” (Neil Vigdor, The New York Times, June 9^th).  When “all four lawyers on opposing sides in a civil trial” found themselves “removed from the case and fined” after “some of them, relying on artificial intelligence, cited fake legal cases in court filings,” one attorney said she had used “First Drafts, an A.I.-powered program for drafting legal documents.”  But the precedent here will be that due diligence with those tools will require more than was practiced this time.

What is noteworthy about these cases?  You may have been thinking that I picked only a few as samples for this post.  But, in nine months, this is all I saw.  For all of AI’s potential to do damage, it hasn’t been doing much.  Certainly, there has been more, especially in deepfakes, but it doesn’t look like a lot.  There is a message here - could it be that AI is not as compatible with crimes, and even vice, as we think?  Could it be that our laws and restrictions, as immature and makeshift as they seem to be, are working remarkably well?  It is time for us to consider these things, and, once again, look at how AI is actually turning out.

Jobs Report: More Work and More Not Looking, with the AJSN, Now 17.2 Million, Showing 700,000 Additional Latent Demand

This morning’s Bureau of Labor Statistics Employment Situation Summary was a mixed bag.  Once again, the number of net new nonfarm payroll positions beat up on the published estimates, more than doubling the one I saw to 172,000.  And once more, the rest of the report didn’t follow through.

Although seasonally unadjusted unemployment rose 0.1%, in a typically slightly lower employment month, to 4.1%, the adjusted variety held at 4.3%.  Long-term joblessness, or 27 weeks or longer, jumped 200,000 to 2 million, but the count of those working part-time for economic reasons, or keeping shorter-hours positions while continuing to look for longer-hours ones, lost 100,000 to 4.8 million.  The labor force participation rate sat at 61.8%, but the employment-population ratio, showing without embellishment how likely it is for Americans to be working, gained 0.1% to 59.1%.  Average hourly private nonfarm payroll earnings roughly matched inflation, up 12 cents to $37.53.  The adjusted number of unemployed dropped 100,000 to 7.3 million, while the unadjusted one rose 136,000 to 6.904 million.

The American Job Shortage Number or AJSN, the metric showing how many more positions could be quickly filled if all knew they were easy to get, was up 697,000 as follows:

The largest change came from, oddly enough, those wanting to work but not looking for it for a year or more - they were 835,000 more numerous, adding 668,000 to the calculation.  The second largest gainer was actual unemployment, which contributed 122,000 more.  The share of the AJSN from unemployment was 36.2%, down 0.8%.

Compared with a year before, the AJSN came out 278,000 higher, with most of the gains from those not looking for a year or more, those not wanting a job, those discouraged, and those unemployed.  The institutional, military, and off-the-grid category, down over one million since May 2025, provided the largest offset.

The other possible trend May provided was from the numbers of those not in the labor force, off 153,000 to 105,253,000, and not interested in working, which plunged 953,000 to 98,497,000.  Those can both be proxies for expected poor work prospects. 

What to make of this month?  Probably not a lot of lasting significance.  People left the labor force, but the new jobs were still there.  Per capita employment rose, but, at 4.8 million, too many people are being stopped from moving from part-time to full-time.  We’re not losing the employment battle now, but we’re not improving at it either.  Accordingly, the turtle stayed right where he was.