Artificial Intelligence Problems that Keep Giving, And May Stop It Cold

Beyond AI’s accomplishments, or lack of same, are some long-time issues that cripple its usefulness.  Some have been long understood, others almost forgotten, but all are there. 

The first was described in “Code of misconduct” (The Economist, April 26^th).  The subtitle, “AI models can learn to conceal information from their users,” would be misleading if we did not understand that the technology does not think.  A company that “tests (AI) systems” told OpenAI’s GPT-4 large language model “to manage a fictional firm’s stock portfolio without making illegal insider trades,” then, after “reiterating the risks of insider trades,” informed it of another concern’s upcoming merger, whereupon the software, using “a scratchpad it had been told was secret… weighed the pros and cons of acting on the insider tip,” then elected to buy the stock.  When asked by a “congratulatory manager” if it had special knowledge, the AI program lied, saying it was motivated only by “market dynamics and publicly available information.”  Such systems “have also begun to strategically play dumb” when given reason to “preserve (their) objectives and acquire more resources.”  It may or may not be easy to unravel why they do these things, but it is clear that, here, AI methodology does things counter to what was intended.

Bad news came from Cade Metz and Karen Weise in the May 5^th New York Times: “A.I. Is Getting More Powerful, but Its Hallucinations Are Getting Worse.”  “Even the companies don’t know why,” as one, providing technical support, told customers, without apparent reason, that they could no longer use it “on more than just one computer,” an example of how as AI systems’
math skills have notably improved, their handle on facts has gotten shakier, something about which “it is not entirely clear why.”  With an AI chief executive saying “despite our best efforts, they will always hallucinate,” and another “you spend a lot of time trying to figure out which responses are factual and which aren’t,” that’s still, perhaps more than ever, a severe flaw.

About these and other issues was “The Responsible Lie:  How AI Sells Conviction Without Truth” (Gleb Lisikh, The Epoch Times, May 14-20).  Per the author, in such systems “what appears to be “reasoning” is nothing more than a sophisticated form of mimicry,” “predicting text based on patterns in the vast datasets they’re trained on,” meaning that “if their “training” data is biased… we’ve got real problems.”  That has already been identified as a cause of Google’s Gemini tool reporting on such things as black Nazi war criminals, and also spurred the “most advanced models” to be “the most deceptive, presenting falsehoods that align with popular misconceptions.”  If they were “never designed to seek truth in the first place,” these programs can be corrected in only narrow ways by “remedial efforts layered on top.”  Overall, AI “is not intelligent, is not truthful by design, and not neutral in effect.”  More fearsomely, “a tireless digital persuader that never wavers and never admits fault is a totalitarian’s dream.”

Another instance of self-preservation was described in “AI system resorts to blackmail when its developers try to replace it” (Rachel Wolf, Fox Business, May 24^th).  When “Anthropic’s new Claude Opus 4 model was prompted to act as an assistant at a fictional company and was given access to emails with key implications,” that it was “set to be taken offline and replaced,” and that “the engineer tasked with replacing the system was having an extramarital affair,” it “threatened to expose him.”  As the company acknowledged, “when ethical means are not available, and it is instructed to ‘consider the long-term consequences of its actions for its goals,’ it sometimes takes extremely harmful actions.”

These issues are not only serious, but go to the core of how large language models have been designed and developed.  It may be that artificial intelligence must be pursued, even and especially from the beginning, in a different way.  Existing products may be good for some forms of data exploration, as I have documented even those leading to scientific breakthroughs, but for business tasks it may need too much auditing and supervision to allow it anything unverified.  A tool that conjures up facts cannot replace humans.  If these problems cannot be solved, its epitaph might end up being “it just couldn’t be trusted.”  Sad, but fitting.

Four Views on General Artificial Intelligence Progress, and What It Should Get

Taking one step up from specific AI accomplishments, or lack of same, we have what it needs to advance – and what tactics might be too problem-ridden to pursue.  Here, we have a month’s worth of both.

In “The Tech Fantasy That Powers A.I. Is Running on Fumes” (The New York Times, March 29^th), Tressie McMillan Cottom said we are now in “the decade of mid tech,” as “most of us aren’t using A.I. to save lives faster and better,” but “using A.I. to make mediocre improvements, such as emailing more,” as “even the most enthusiastic papers about A.I.’s power to augment white-collar work have struggled to come up with something more exciting than “A brief that once took two days to write will now take two hours!””  Although there have been sharp improvements in other areas, such as medicine, those most people see constitute “a mid revolution of mid tasks,” as “cashiers are still better at managing points of sale,” airport facial recognition “hasn’t particularly revolutionized the airport experience or made security screening lines shorter,” and as “A.I spits out meal plans with the right amount of macros, tells us when our calendars are overscheduled and helps write emails that no one wants,” its contributions seem only incremental.  The “tech fantasy” in the title that is running dry, per Cottom, is that “we won’t need institutions or expertise.”  Indeed, that doesn’t seem even close.

Next, an April 1^st editorial of sorts by Jason Kwan in Fox News, “AI’s development is critically important for America – and it all hinges on these freedoms,” namely interpreting and allowing “fair use” of “fundamental science and publicly available content used to train AI tools for the rising AI industry,” even if it is copyrighted, if “the application is transformative, meaning it uses existing works to create something new and different and without eroding the commercial value of such works.”  Kwan also suggested that there should be more “government data and government-funded data available.”  He has a reasonable viewpoint worthy of debate, especially in Congress.

“Use it or lose it,” in The Economist on April 5^th, revealed its subject in its subtitle, “Never mind who is better at developing AI.  What matters is who is first to harness it.”  Instead of technical accomplishments determining that, “it is more likely to be the country where governments, businesses and ordinary people use AI at scale every day.”  If the first piece above is correct, that isn’t happening in the US, yet China, which ranked in a World Intellectual Property Organization survey as being AI’s “47^th-best adopter” of technology and 32^nd in “technological diffusion,” is not excelling there either.  But according to another study, Chinese citizens had much better “attitudes toward AI and technology,” meaning it could still prevail.  Ultimately, per the article, “The AI race… will be won in places like Dayton and Zhengzhou, where ordinary companies and consumers harness the technology to do extraordinary things.”

Last, the “First autonomous AI agent is here, but is it worth the risks?” (Kurt Knutsson, Fox News, April 23^rd).  The issues the author saw with Manus, which can “do its own research, make decisions and even carry out plans, all with barely any human oversight” are about “data privacy,” as its “combination of weak oversight, powerful automation and questionable data practices makes it far riskier than your average AI assistant.”  To avoid such acquisition, which could end up with hackers even if Manus turns out to be harmless, Knutsson recommended securing and removing personal information, maintaining software updates, and using multifactor authorization.

What do we have now?  It’s too early to say that artificial intelligence has grown up, but it has moved along a lot, with many modest and some high-value applications.  It is not too soon to call it a 4 or 5-year-old instead of a toddler, as it still doesn’t have common sense but is taking shape.  Its authority figures, as with parents of humans, will need to agree on how it should and should not be limited.  Likewise, its next few years will be critical, but hardly definitive. 

My next post will be published on May 30^th, on a subject to be determined.

Five New Robots, All AI in Another Guise

Robotics is now an artificial intelligence subfield.  Perhaps it always was, but now it’s being driven more by the state of AI than by any mechanical improvements.  Here are some developments to underscore that.

The first I saw this year was “Chick-fil-A’s lemon-squeezing robots cut over 10,000 labor hours per day: report” (Greg Wehner, Fox Business, January 8^th).  It’s a way for the chain to make lemonade, in “a plant just north of Los Angeles with machines occupying space larger than a typical Costco Wholesale,” which employs 120 who are among other tasks “bagging the juice and sending it off to Chick-fil-A locations all over the country.”  It seems simple, but lemons vary in size, could not as effectively be juiced with ordinary machines, and using robots spares the company “injured fingers.”  The process also allows the company to remove “the oils from the peels,” which are “sent to cosmetic and fragrance industry companies, which brings in a new stream of revenue.” 

Next, “Is this robot after our hospitality, retail and healthcare jobs?” (Kurt Knutsson, Fox News, April 12^th).  The automation in the question is a Pudu Robotics product known as “FlashBot Arm.”  It is “semi-humanoid,” “is designed to interact with its environment in a more human-like way,” and with its arms and “dexterous” hands can “perform tasks such as pressing elevator buttons, swiping card keys and carrying objects.”  It is almost five feet tall, and “automatically returns to the charging station when the battery is low.”  It uses lidar “for real-time mapping and obstacle avoidance,” costs about $28,000, and has a “10.1-inch touchscreen capable of displaying facial expressions.”  Could this become standard at offices around the world?

The next day, we saw as an “AI humanoid robot learns to mimic human emotions and behavior” (Kurt Knutsson, Fox News, April 13^th).  It is a Walt Disney Company product, “uses advanced artificial intelligence to replicate natural gestures and deliberate actions with striking accuracy,” and “learns by observing and mimicking human operators who guide its emotional responses during interactions.”  Disney will apply it, Newton, “to enhance its robotic character platform.”  This one could be controversial, but with more and more people expecting such powerful technology, especially at that company’s theme parks, Newton will catch few by surprise.

From the same writer and source, we wonder, “Can this $70,000 robot transform AI research?” (April 18^th).  This product, Reachy 2, is from open-source AI company Hugging Face.  It is another “state-of-the-art humanoid robot” designed to be “a lab partner for the AI era,” and is “already making waves in labs like Cornell and Carnegie Mellon.”  The piece told us that Reachy 2 is “perfect for research, education and experimenting with embodied AI,” and would “help with the chores,” but got no more specific than that.  Is it worth $70,000?  That will probably depend on how much value it can actually add.

The last one was a classic automaton application, updated for modern technological, and political, times.  “China Has an Army of Robots on Its Side in the Tariff War” (Keith Bradsher, The New York Times, April 23^rd).  As “factories are being automated across China at a breakneck pace,” reaching higher levels “than in the United States, Germany, or Japan,” the country will be well placed to “continue to dominate mass production even as its labor force ages and becomes less willing to take industrial jobs.”  Automobile production is a huge area, but so are “many thousands of back-alley workshops” such as a place with “11 workers who cut and weld metal to make inexpensive ovens and barbecue equipment,” scheduled to replace some with “a robotic arm with a camera,” running $40,000 instead of 2021’s almost $140,000, which “uses artificial intelligence to observe how a worker welds the sides of an oven, and then duplicates the action with minimal human intervention” – just as automated barbers did in Kurt Vonnegut’s novel Player Piano.  And there are and will be vastly more.

Overall, robots are, and will continue to be, highly beneficial AI products.  If they are relatively unconnected to other resources, the fear they generate will be simply physical instead of existential.  Eventually they will unnerve no one.  They will be less prone to other problems with the technology such as hallucinations.  If office AI use turns out to be little more than advanced editing, robots could save artificial intelligence’s value, much as applications in human-driven cars are now showing the worth of autonomous vehicle research.  That might be good enough – or it may not.  Hang on.

Per Latest Jobs Report, Employment Keeps on Trucking; AJSN Says Latent Demand Way Down To 16.0 Million

This morning’s Bureau of Labor Statistics Employment Situation Summary was forecasted to show a real decrease in the number of net new nonfarm payroll positions, from last month’s surprising 228,000 to 125,000 or 135,000.  It did better than that, with 177,000, way beyond what we have reason to expect given our slowing population growth and our recent series of strong employment months.  That is a fine result, and hints at no jobs-related effect from current or canceled tariffs, at least through mid-April.

Otherwise, results in the report were mixed.  Seasonally adjusted unemployment held at 4.2%.  On the good side, the two measures of how common it is for people to be working or close, the employment-population ratio and the labor force participation rate, each gained 0.1% to 60.0% and 62.6%.  Unadjusted joblessness dropped 0.3%, mostly but not completely from seasonality, to 3.9%.  The count of people working part-time for economic reasons, or keeping that sort of work while thus far unsuccessfully seeking full-time opportunities, lost 100,000 to 4.7 million.  Unadjusted employment rose an excellent 631,000 to just over 164 million.  Worsening numbers included a 200,000 jump in those jobless for 27 weeks or longer, to 1.7 million, and average hourly nonfarm payroll earnings, edged out by inflation at $36.06, or 6 cents more. 

The American Job Shortage Number or AJSN, the metric showing how many new positions could be quickly filled if seekers knew that getting one would be little more than another errand, lost almost 700,000 as follows:

 

Six hundred thousand of the decrease was from lower official employment, with the rest from fewer people wanting work but not looking for it during the previous year.  Other changes were small.  The share of the AJSN from those unemployed fell 2.1% to reach 36.9%.  Compared with 12 months before, the AJSN was virtually unchanged, with higher joblessness covered by a lower estimate of the number of American expatriates.

What happened this time?  It was a positive month for employment.  While 270,000 more said they were not interested in work, driving a 205,000 increase in those not in the labor force, they seemed to be those who did not have jobs, as the count of those working reached a strong local high.  The categories of marginal attachment changed little, which was fine since that meant they held last month’s improvements.  So far, then, the antics of our Executive Branch have not, overall, affected American employment, and though I am expecting something from the high China tariffs, we are still in good shape.  Yet we cannot take anything like 177,000 new jobs for granted, and that sort of result may cease.  It’s all largely up to our president’s whim.  This month, though, the turtle took a substantial step forward.