Thursday, July 11, 2024

A Directionless Month for AI – What Do Its Scattered Reports Say?

Artificial intelligence seems to be moving into another phase, or subphase.  There have never been so many concerns, not about the technology causing a major disaster, but about it being adequate and worthwhile at all.  At the same time, there are new perceptions and more progress in related employment, and, if I am perceiving it correctly, growing acceptance that where AI is now might be where it is for months or even years, so stakeholders are realizing they cannot just wait for the next big eruption. 

A look at one chronic AI problem is the subject of “How Game Theory Can Make AI More Reliable” (Steve Nadis, Wired.com, June 9th).  The author bemoaned the tendency of large language models to give different answers, not all correct, to generative (open-ended) and discriminative (choice of options) questions.  Researchers have invented a game of sorts where “two modes” of LLMs are asked to agree on answers.  Excellence at games is a long-time AI strength, so engineers have reason to be optimistic such inconsistencies will go away.

Top AI workers are scarce, and many, soon after starting work with an organization, move on to another for more pay.  That is why, on one level, “Retention is all you need” (The Economist, June 15th).  As, per a market analyst, “20,000 companies in the West are hiring AI experts,” “AI talent, previously hoarded at tech giants, is becoming more distributed.”  Indeed, AI hiring at Amazon, Apple, Google, Meta, and Microsoft has only about broken even since the beginning of 2023.  Advertisements for many more software positions now mention AI, as do those for math, scientific research and development, and information design.

Defining another clear problem, Sherin Shibu’s “How Can AI Help Small Businesses?  It’s A Matter of Trust, According to a New Report” (Entrepreneur, June 17th), tells us that “just 7% of U.S. desk workers see AI answers as completely trustworthy,” the share of American businesses as of June was only 5% although 96% of “surveyed executives felt pressure to bring AI into their business,” and “privacy” and “data quality” were the other two top issues.  These tell AI companies where they should focus, perhaps above all other considerations.

Unsurprisingly, more universities and technical schools are preparing students for jobs using the technology.  A Brock Dumas June 17th Fox Business piece, “Want an AI career?  These colleges offer degrees for the best chance” provides a current look.  “A new study by software development firm Vention shows which U.S. colleges are offering bachelor’s degrees with the best chance of landing a job fresh out of school – and the potential for the biggest paychecks.”  The top five were California Polytechnic State University, Wake Forest University in North Carolina, Trinity University in Texas, Clarkson University in New York state, and Knox College in Illinois.  There are more, and this list will change, but these may be the ones currently most worth investigating.

Also, for now, “We’re Still Waiting for the Next Big Leap in AI” (Will Knight, Wired.com, June 20th).  Knight wrote that “the world is still waiting for another AI leap forward… akin to that delivered by GPT-4” 16 months ago.  He described possible contenders, but then concluded “it’s unclear how long the world must wait for that next big leap in AI.  OpenAI has said it has started training its next big model.  In the meantime, we will need to figure out new ways to measure how useful the technology really is.”

On June 27th, Fox News released “US tops world ratings for AI preparedness:  China, Russia and Iran lag in key measures, report finds” (Peter Aitken).  The rankings were on countries’ “ability to immediately adopt artificial intelligence… into their economies.”  The American “value of preparedness” tied the Netherlands, and was followed by Finland, Estonia, New Zealand, Germany, Sweden, Australia, Japan, and Israel.  That may or may not prove to be valuable information.

There must be something noteworthy about a source called Futurism.com having the most consistently negative AI articles of anyone.  Here are two more.  The first is “There’s a Small Problem with the AI Industry:  It’s Making Absolutely No Money” (July 4th).  Author Sharon Adarlo noted that “Goldman Sachs analysts have concluded… that AI just isn’t making any serious money yet,” and that “Goldman found that companies that hoped to profit from using AI to boost productivity – ranging from H&R Block to Walmart – have seen their shares vastly underperform the broader stock market since the tail end of 2022.”  She also found that “the only companies making much actual revenue off AI are the ones selling the hardware it needs, like Nvidia.”  While “companies using it want to see major returns,” “so far, it sounds like they aren’t.  Maybe the real question is how much runway the AI industry has before business leaders move onto the next thing.”  This piece is consistent with my previous comments.

The second Futurism.com piece was “Expert Warns That AI Industry Due for Huge Collapse” (Victor Tangermann, July 9th).  He named a “founding partner” at a “macroeconomic research firm” calling AI “completely unproven,” saying that its hallucinations may never go away and that it was “too energy hungry,” along with a previous Stability AI CEO telling bankers that “this will be the biggest bubble of all time.”  Hardly universal views, but with merit.

Overall, is it true that, per New York Times columnist Thomas Friedman on July 9th, “the artificial intelligence revolution of the past four years is widely expected to slam into the white-collar job market in the next four like a Category 5 hurricane”?  We don’t know that.  As with actual storms, we will benefit from tracking them and projecting their courses and times of landfall, but this one is way too weak and distant to fear.  We don’t need to buy flashlights, batteries, and water, and may never need to.  The future of artificial intelligence is still up in the air – and it may never come down.

Friday, July 5, 2024

Jobs in June: Seasonal Worsenings Mostly Offset Elsewhere, with AJSN Showing Latent Demand Up Almost Half a Million

The published projections I saw for this morning's Bureau of Labor Statistics Employment Situation Summary was that it would be worse than for May’s data a month ago.  The two estimates of net new nonfarm payroll positions were 190,000 and 200,000, and unemployment might be going up again.  So what happened?

Employment as above, at plus 206,000, was quite close to the predictions.  Seasonally adjusted joblessness had its third straight 0.1% gain, to 4.1%, with the corresponding total of people up 200,000 to 6.8 million.  (We now can ignore when the BLS says something “changed little.”)  Long-term unemployed, for 27 weeks or longer, gained 100,000 to 1.5 million, up 36% from June 2023, with so many people joining the labor force that its participation rate increased, 0.1% to 62.6%.  The measure of how many Americans are actually working, the employment-population ratio, stayed at 60.1%.  The count of those working part-time for economic reasons, or holding onto shorter-hours positions while looking for full-time ones, shed 200,000 to get to 4.2 million.  Average private nonfarm payroll earnings rose 9 cents per hour, close to the inflation rate, to $35.00.

Since May and June have different employment characteristics, the seasonally unadjusted figures did not match the others.  Unemployment that way jumped 0.6% to 4.3%.  The count of those not interested in working lost 637,000 to 93,776,000.  Those employed rose 433,000 to 161,774,000.

The American Job Shortage Number or AJSN, the statistic showing how many additional positions could be quickly filled if all knew they would be easy and routine to get, was up 478,000, as follows:


The share of the AJSN from those officially jobless was 4.3% higher at 37.7%.  Compared with a year earlier, the AJSN grew 504,000, with almost 800,000 more from unemployment partially equalized by, among others, 174,000 from fewer people wanting work but not looking for it for a year or more, and 200,000 fewer from expatriates. 

What patterns can we get from this report?  The new jobs, once again plentiful and nothing to take for granted, went largely to people with statuses other than simple unemployment.  Many more people returned to the labor market, and enough were unsuccessful to bring overall joblessness up.  A goodly number of those without work are not finding it, even after six months away.  Latent demand is not only alive and well but increasing.  Still, June is a tougher month than May, and the smaller, marginal categories show that this was, overall, a good one.  The turtle took a moderate step forward.