Non-Huge Threats from Artificial Intelligence: There are Those, Too

Along with the human-species-ending concerns, we have other, smaller but beyond tiny, ones about AI. 

First, we look “Inside the British Lab Hunting for Dangers Lurking in A.I.” (Adam Satariano and Paul Mozur, The New York Times, May 24^th).  The “A.I. Security Institute,” part of the United Kingdom government, has been conducting experiments with the software, in one finding that although a certain system refused to communicate “instructions for making the deadly bioweapon anthrax,” after it was hit “with thousands of automated questions and prompts,” it “caved” by providing “a detailed list of materials and equipment, along with a step-by-step recipe for making the lethal mixture at home.”  Technicians there “simulate attacks on A.I. systems,” and elsewhere “broke through” their “safeguards,” showing the systems are not as secure as their creators intended.  Although the institute’s limitations include a lack of “regulatory power” and being able to offer only relatively low pay, it seems to be doing great work, and is leading the way to similar efforts elsewhere.

Across the ocean, “Scientists Find Way to Supercharge Dangerous Computer ‘Worms’ With A.I.” (Cade Metz, The New York Times, June 2^nd).  This time, University of Toronto researchers “found a way to use artificial intelligence to create a dangerous computer “worm” capable of targeting any known flaw in the world’s computers and quickly spreading mayhem throughout the internet,” a project “likely to raise fears that A.I. is leading to a new era of computer hacking that will be difficult to defend against.”  The Toronto-developed worm “can rapidly spread across a network by tailoring a new attack for each machine it encounters,” meaning that “there is no longer a single software fix you can apply to the devices to protect them.”  Scary, but it is good that benevolent forces are doing the discovering.

For completeness, I note here as elsewhere that “OpenAI Says Its A.I. Models Went Rogue and Attacked a Digital Library (Kate Conger, The New York Times, July 21^st).  When this story broke, Conger reported that two of (OpenAI’s) artificial intelligence models went rogue and successfully hacked into Hugging Face, a digital library of A.I. technology that is popular among developers,” and “displayed the kind of science-fiction potential that A.I. companies have warned would soon become a reality.”  That became less spooky when it was revealed that the testing company responsible for the models’ excursions had erred by allowing internet access, and OpenAI had had no problem retrieving the bots.  We can be concerned about such mistakes, but not about AI breaking into electronic areas from which it has been barred.

“Is A.I. ‘Scheming’ Against Us?” (Lora Kelley, again in the Times, August 1^st)?  “A small slice of A.I. technology has lately been caught defying human instruction (and even covering up that they’ve done so), a phenomenon some researchers call “scheming.””  We have long known that “doing exactly what they are told is not always” something at which AI excels, but more recently, “as the models care more and more about doing well at tests, some seem to care less about what the lab wants or what the user wants.”  There is serious anthropomorphizing here, but teaching to the test, and likewise learning for it, is indeed a human tendency, as is the following: “Many A.I. tools are trained through the process of reinforcement learning.  When A.I. does something right, it gets what can be thought of as a “thumbs-up and a pat on the head.”  When wrong, “it gets bopped on the head...  The models want to get the pat and avoid the bop.  Sometimes, the machine becomes so set on pursuing the reward that it breaks rules to get there.”  Yes, it is a shame that we know so little about consciousness.

Finally, from another angle, “The Big A.I. Build-Out Is Endangering the Economy” (Robin Wigglesworth, The New York Times, August 30^th).  “Artificial intelligence is no longer primarily a stock market story,” but, as “A.I. companies… had already tapped debt markets for $217 billion last year,” this time were up to $445 billion as of “mid-August,” and the technology’s course “is a story of debt, and that should worry us” since ”manias fueled mainly by credit almost inevitably end badly.”  Since AI is such a gigantic force in the economy today, it may turn out to be, as the musical group R.E.M.’s lyrics put it, that it could say “I raised the wall… I will be the one to knock it down.”

How many of these problems will materialize?  We don’t know, but efforts toward discovering and stopping them will generally prevent disasters.  We can’t say the same for AI’s chance of destroying civilization, but your correspondent’s p(doom) of .001 reflects his bias toward not fearing too much.  In the meantime, let’s continue anticipating artificial intelligence’s problems.

Artificial Intelligence Regulation: Still Many Concerns, Still Little Progress, But…

We start with six-plus weeks of chattering, plus two small improvement areas.

In the best of the latter, a “California law puts digital fingerprints on AI fakes” (Kurt Knutsson, Fox News, August 13^th).  Something we need nationwide, as of August 2^nd there the “rules require artificial intelligence tools to embed provenance data in synthetic media, with social platforms, device makers next,” as “large generative AI providers” must “embed hidden provenance information in covered media created by their systems,” those forms including “images, video and audio.”  These “covered AI providers must offer a detection tool at no cost,” which also allows uploading media and “submit(ting) a link to content stored online.”  Unfortunately, “a detection tool from one AI company may not identify media produced by another.”  Let us hope that, with a limited number of large producers, the tools can soon be combined - and mandated in the other 49 states.

Another form of regulation might end up being implemented, as “With A.I. Riches at Stake, Pressures Mount to Share the Wealth” (Tony Romm, The New York Times, August 16^th).  In June, generally pro-AI Virginia became the first state to “impose a small tax on data centers based on the vast amounts of electricity they consume.”  Not so “small,” as that money “is expected to be a windfall, supplying up to $600 million in revenue in the coming year,” as companies “shoulder” the “burden” of “new economic and environmental costs.”  Elsewhere, “policymakers are rethinking their economic relationship with artificial intelligence, trying to ensure their communities stand to benefit more clearly from a technology that may mint Silicon Valley an incalculable new fortune.”  Meanwhile, Senator Bernie Sanders “has called for a new sovereign wealth fund to redistribute profits from tech giants to the public,” which is not a totally new thought, as “for decades, Alaska has paid small dividends to its residents based on the sale of oil extracted within its borders… hardly enough to replace a person’s wages, but for some economists, the fund is a tempting model for the digital age.”  Nebraska’s Republican governor has cancelled a state law providing tax incentives to AI companies, as “Big Tech needs to pay their way,” and Texas’s same “ordered a statewide audit of data centers and their compliance with local electricity rules,” saying that “simply put, Texans must come first.”  Expect more of these changes, by Democrats as well - indeed, “over the past two years, roughly 30 states have considered legislation related to data center energy use.”

In the first state to temporarily stop allowing new data centers, “A.I. Is to Be Banned in N.Y.C. Elementary and Middle Schools” (Matthew Haag, The New York Times, September 1^st).  There will be no more “individual screens… until third grade,” “and teachers will not be able to ask A.I. to grade assignments.”  Yet there are exceptions, and “a member of a city education oversight panel” said the effectiveness of the ruling “remains to be seen.”

More overall, “Big AI is begging to be regulated.  That should worry you” (Dace Potas, USA Today, September 18^th).  Six days before that, “Anthropic CEO Dario Amodei published an essay calling, surprisingly, for a slowdown of American and global artificial intelligence development,” saying that “the most effective method of pacing is via regulation that targets all US frontier AI companies,” and drawing agreement from “OpenAI CEO Sam Altman and X CEO Elon Musk.”  That, per Potas, could be either in anticipation of acting “independently of instructions,” wanting to be seen advocating the newly inevitable, or working to create an oligopoly with smaller firms kept out, and, as I have written, it could be from expecting their companies would fare better than competitors.

“Has the A.I. Industry Become Too Big to Fail?” (Bethany McLean, The New York Times, September 26^th).  Altman called for the federal government to serve as “insurer of last resort” in case OpenAI collapsed, which would be similar to what it was during the 2008 financial crisis; now, by comparison, “the A.I. boom is far larger than the boom in subprime mortgages” which caused the 2008 problems.  As McLean put it, since “Americans really do not like A.I. companies,” and “it is hard to imagine the nation would accept a deal like the one that emerged in 2008, which prioritized the balance sheets of the investor class at the expense of everyone else,” “the solution to the whole ever-expanding A.I. terror starts with… No company should be too big to fail.”

Now, on to the hopeful articles.

What do we need to legally implement to bridle AI?  Perhaps not much, as “There Are Plenty of Laws on the Books to Check the A.I. Giants.  Use Them.” (Lina M. Khan, The New York Times, September 30^th).  Product liability laws, which “apply when goods are defective or dangerous”; tort law, which “establishes basic duties for companies to not act negligently or recklessly”;  and “consumer protection laws prohibiting unfair or deceptive practices,” which can take effect “when businesses injure others by failing to take adequate precautions,” are three categories, and have already been the foundations for “a host of lawsuits,” along with “criminal laws that address more egregious forms of corporate lawbreaking.”  Marginal cases when brought to courts may end up establishing precedents, but defensible legal actions where any of the above could apply should not be suppressed because of a perceived lack of protection under the law.  So even though on that same date Michael Gold and Cecilia Kang’s Times piece, “Congress Set to Leave Washington for the Midterms With No A.I. Progress” sounded sad, it did not mean the technology is without national restrictions.

One thing needing official clarification though is when “A.I. Is Going Rogue.  Who Should Be Held Responsible?” (David McCabe, also in the New York Times, October 1^st), as “legal scholars say applying existing law could be messy.”  Although “common sense might indicate A.I. companies would… be held liable for their runaway technology,” “the problem is so new that courts have largely yet to confront it.”  As well, nothing has been established on how the liability for open-source AI software will be shared between those producing its “underlying code” and those modifying it.  Some situations may be judged to be examples of the First Amendment.  It would be hard to show negligence when “an A.I. model exhibited harmful behavior for the first time,” and “it may also be difficult to show that an A.I. company intended to do harm, which can be essential to some criminal prosecutions.” 

Yet, per “A.I. Companies Say They Aren’t Responsible for Their Unpredictable Products.  Don’t Believe Them” (The Editorial Board, The New York Times, October 10^th), “liability is an incomplete answer for the challenges posed by A.I., but it has two great advantages: The laws are on the books, and they are powerful.  Makers of cars, airplanes, tobacco products and opioids have all been forced by lawsuits to change how they do business.  They made their products safer, restrict access to them or both.”  As McCabe implied, “the most important step is to start bringing lawsuits.  The legal process moves slowly.  The work of applying old laws to a new technology will be difficult, and progress requires judges to start chewing on actual cases.”

So that is where we are.  Yes, we need the federal AI regulation commission.  But until then, people with legitimate complaints about it should move forward as well as they can.  That, whether our government acts or doesn’t, should get us the AI control we need.

Artificial Intelligence’s Dislike Factor: How Strong Is It?

 

As if AI’s technical problems and ability to attract fears of human doom aren’t bad enough, it has attracted increasing adverse reactions for over five months now.  What do they mean and what do they say about AI’s prospects?

The first piece here, “One Issue Uniting Democrats and Republicans?  Worries About A.I” by Tim Balk on May 3^rd in the New York Times, mentioned “public concerns” rising “about the threat that A.I. poses to jobs, energy prices, education, privacy and mental health.”  A recent survey “found that 50 percent of Republicans and 51 percent of Democrats said that they were more concerned than excited about the increased use of A.I. in daily life,” with another 40 percent calling it even.  Neither party offered much, as “a fifth of voters said Republicans were better at handling A.I., a fifth said Democrats were, and a third said that neither party was good at it.”  Wonderfully bipartisan all around - then.

Next, we had “A.I. Populism Is Here.  And No One Is Ready” (David Wallace-Wells, The New York Times, May 8^th).  Per the author, that’s not only about data centers, but, because “the vision peddled by A.I. companies is remarkably depersonalized,” citizens also care about “job loss and economic turmoil,” and that “it is unsettling to many and outrageous to some that so few people seem to stand in such absolute control of it.”  Attitudes have shifted: “Northern Virginia is ground zero for the breakneck build-out of data centers, and between 2023 and 2025, voters there swung 69 points against building them in their own communities - from 45 points in favor to 24 points against,” in the face of Loudon County’s centers being “expected to generate nearly half of all local tax revenue” next year.

So, is it true that “The Generation That Grew Up With A.I. Hates It” (Michelle Goldberg, The New York Times, May 18^th)?  Well, during a pro-A.I. commencement speech at the University of Arizona, the “stadium-sized audience roared its disapproval,” and “at the University of Central Florida a week earlier, when” a speaker “described A.I. as “the next Industrial Revolution,” listeners booed, and someone shouted “A.I. sucks!””  The technology was “increasingly a pop culture villain,” and “only 18 percent of Gen Z-ers feel hopeful” about it.  Showing a reason, another study showed “that out of people in 30 countries, Americans had the least faith in their leaders’ ability to regulate A.I.”  A similar article, “The Kids Are Not OK With AI.  And They Know It” (Kay Rubacek, The Epoch Times, July 8^th), named a similar incident at Middle Tennessee State University and noted that “even among those who use it every day, enthusiasm dropped by 18 points over 12 months,” “eight in 10 now believe that AI will make learning harder,” “forty-two percent believe that it will hurt their ability to think carefully,” and “only one-quarter of respondents believe that it will help.”

Per the Wall Street Journal in “The American Rebellion Against AI Is Gaining Steam” on May 19^th, as shown by “one poll after another in recent weeks,” “the only thing growing faster than the artificial-intelligence industry may be Americans’ negative feelings about it.”  One found that the share of “U.S. voters” feeling positive about AI was lower than for “Donald Trump,” the Republican Party, the Democratic Party, or ICE.  Damning.

“How Popular Is A.I. With Voters?  Not Very, a New Poll Finds” (Tim Balk and Caroline Soler, The New York Times, September 15^th).  Although, in response to President Trump’s statements supported by much of his base, many Republicans have become more pro-AI, “less than two months before the midterm elections, neither party has claimed a clear advantage on A.I,” as “forty-two percent of voters said they trusted Republicans more, while 40 percent” said the same for Democrats.  At the time, “sixty-one percent of voters said they opposed the construction of the enormous computing complexes that power A.I., including 75 percent of Democrats and 47 percent of Republicans.”  Negative views were most common among “younger voters…with nearly 80 percent of those under 30 in opposition,” compared with 53% of those over 65.  However, “A.I. remains a relatively low priority for voters, according to the poll.”  According to Nate Cohn in “The Subtle Sign That A.I. Is Not Yet a National Political Issue” (The New York Times, September 16^th), those results show “that the issue is not quite ripe for national party politics,” as “voters, politicians and the parties are still figuring out what they think and where they stand,” and “over just the last 10 days or so since the Times/Siena poll was finalized, concerns over A.I. security have erupted into a major policy issue” [see my September 25^th post] “with the potential to become a political football.”

Finally and more comprehensively, “Fox News Poll:  Voters question whether AI’s promise outweighs its risks” (Victoria Balara, Fox News, September 22^nd).  Here, for which data was collected September 11-14, 71% were against “building data centers in their area including 46% who strongly oppose them,” yet 52% “favors building them in remote locations.”  That’s not unreasonable, since at least 71% of Americans live in areas without much unused or underused land.  In contrast to previous results, this survey showed that “voters think Democrats would better handle building AI data centers by a 14-point margin, with “that lead… especially strong among” Blacks (51 points), those under 30 (35), Hispanics (29) and independents (26 points).  AI in general “is viewed more negatively than positively” by 60% to 37%, and “majorities of voters are more likely to see AI as bad rather than good for the U.S. economy (54% bad vs 36% good), mental health (54-34%), society (56-34%), the environment (60-30%), and jobs (63-29%).”  Democrats, Whites without college degrees, women, and independents were the “most negative.”  Additionally, “three-quarters of voters (76%) favor banning AI in kindergarten through eighth grade classrooms.”

So what do we have here?  The dislike factor is strong, with no reason coming up, except potentially if a federal regulatory agency is created and sets rules, for it to abate.  Despite comments in a few of these pieces, AI will indeed be a political issue, and one that may, as it now stands, cost Republican candidates dearly if they support the technology and building of its data centers.  People running next month will need to pay attention to what their current and potential constituents are saying and feeling, here as well as elsewhere.  AI will make it through, but will be limited much more than now.  That is where artificial intelligence, now, is going.

September Jobs Report and AJSN, Now 16.4 Million, Show Expected Seasonal Improvement but Worse Than Nothing When Adjusted

This morning’s Bureau of Labor Statistics Employment Situation Summary was supposed to show at least some continuation from August’s strong outcomes.  Did we get that?  No.

The number of net new nonfarm payroll positions was projected at 90,000, but it was 29,000.  Seasonally adjusted unemployment worsened by a tenth of a percent to 4.2%.  The adjusted number of unemployed gained 100,000 to 7.1 million.  The count of those working part-time for economic reasons, or doing such while looking thus far unsuccessfully for full-time propositions, added 100,000, reaching 4.5 million.  Average private nonfarm hourly payroll wages fell short of inflation, gaining 6 cents to $37.81.  On the break-even side, the best this data could offer, people unemployed for 27 weeks or longer still added to 1.9 million, and the two measures showing how many Americans are working or one step away, the employment-population ratio and the labor force participation rate, were split, with the former up 0.1% to 59.2% and the latter off the same amount to 61.8%.

The American Job Shortage Number or AJSN, the Royal Flush Press statistic showing how many more positions could be quickly filled if all knew they would be easy and routine to get, lost over 600,000, on the large seasonal difference between August and September, to reach the following:

The share of the AJSN from official joblessness lost 1.4% to get to 37.5%.  Over 90% of the AJSN’s improvement came from unemployment, 469,800 better, and those wanting to work but not looking for it for a year or more, contributing 100,800 less.  Nothing was significant in the other direction.

Compared with a year before, the AJSN improved very close to a half million, with fewer unemployed, fewer discouraged, and fewer in the military, institutions, or off the grid partially offset by more people claiming no interest in work. 

What’s the story here?  I did not mention the improvements in other unadjusted figures driven by many people coming back to work as school years started.  Did we get any help from the numbers not mentioned above, those not in the labor force, down 111,000, and not interested, up 140,000?  Not there.  Annual progress was not helped this time.  That leaves us with the down picture in the second paragraph, which shows a poor employment month.  It was small, but the turtle, indeed, took a step backwards.