Regulate Does Not Mean Regulate. Leave the EU Does Not Mean Leave the EU. Got That?
May 30, 2023
I wrote about Sam AI-man’s explanation that he wants regulation. I pointed out that his definition of regulate means leaving OpenAI free to do whatever it can to ace out the Google and a handful of other big outfits chasing the pot of gold at the end of the AI rainbow.
I just learned from the self-defined trusted news source (Thomson Reuters) that Mr. AI-man has no plans to leave Europe. I understand. “Leave” does not mean leave as in depart, say adios, or hit the road, Jack.
“ChatGPT Maker OpenAI Says Has No Plan to Leave Europe” reports:
OpenAI has no plans to leave Europe, CEO Sam Altman said on Friday, reversing a threat made earlier this week to leave the region if it becomes too hard to comply with upcoming laws on artificial intelligence.
I am not confused. Just as the company’s name OpenAI does not mean “open,” the AI-man’s communication skills are based on the probabilities of certain words following another word. Got it. The slippery fish with AI-man is that definition of the words in his mind do not regress to the mean. The words — like those of some other notable Silicon Valley high tech giants — reflect the deeper machinations of a machine-assisted superior intelligence.
Translated this means: Regulate means shaft our competitors. Leave means stay. Regulate means let those OpenAI sheep run through the drinking water of free range cattle
The trusted write up says:
Reacting to Altman’s tweet on Friday, Dutch MEP Kim van Sparrentak, who has worked closely on the AI draft rules, told Reuters she and her colleagues must stand up to pressure from tech companies…. Voluntary codes of conduct are not the European way.
What does this statement mean to AI-man?
I would suggest from my temporary office in clear thinking Washington, DC, not too much.
I look forward to the next hearing from AI-man. That will be equally easy to understand.
Stephen E Arnold, May 30, 2023
Probability: Who Wants to Dig into What Is Cooking Beneath the Outputs of Smart Software?
May 30, 2023
Note: This essay is the work of a real and still-alive dinobaby. No smart software involved, just a dumb humanoid.
The ChatGPT and smart software “revolution” depends on math only a few live and breathe. One drawer in the pigeon hole desk of mathematics is probability. You know the coin flip example. Most computer science types avoid advanced statistics. I know because my great uncle Vladimir Arnold (yeah, the guy who worked with a so so mathy type named Andrey Kolmogorov, who was pretty good at mathy stuff and liked hiking in the winter in what my great uncle described as “minimal clothing.”)
When it comes to using smart software, the plumbing is kept under the basement floor. What people see are interfaces and application programming interfaces. Watching how the sausage is produced is not what the smart software outfits do. What makes the math interesting is that the system and methods are not really new. What’s new is that memory, processing power, and content are available.
If one pries up a tile on the basement floor, the plumbing is complicated. Within each pipe or workflow process are the mathematics that bedevil many college students: Inferential statistics. Those who dabble in the Fancy Math of smart software are familiar with Markov chains and Martingales. There are garden variety maths as well; for example, the calculations beloved of stochastic parrots.
MidJourney’s idea of complex plumbing. Smart software’s guts are more intricate with many knobs for acolytes to turn and many levers to pull for “users.”
The little secret among the mathy folks who whack together smart software is that humanoids set thresholds, establish boundaries on certain operations, exercise controls like those on an old-fashioned steam engine, and find inspiration with a line of code or a process tweak that arrived in the morning gym routine.
In short, the outputs from the snazzy interface make it almost impossible to understand why certain responses cannot be explained. Who knows how the individual humanoid tweaks interact as values (probabilities, for instance) interact with other mathy stuff. Why explain this? Few understand.
To get a sense of how contentious certain statistical methods are, I suggest you take a look at “Statistical Modeling, Causal Inference, and Social Science.” I thought the paper should have been called, “Why No One at Facebook, Google, OpenAI, and other smart software outfits can explain why some output showed up and some did not, why one response looks reasonable and another one seems like a line ripped from Fantasy Magazine.
In a nutshell, the cited paper makes one point: Those teaching advanced classes in which probability and related operations are taught do not agree on what tools to use, how to apply the procedures, and what impact certain interactions produce.
Net net: Glib explanations are baloney. This mathy stuff is a serious problem, particularly when a major player like Google seeks to control training sets, off-the-shelf models, framing problems, and integrating the firm’s mental orientation to what’s okay and what’s not okay. Are you okay with that? I am too old to worry, but you, gentle reader, may have decades to understand what my great uncle and his sporty pal were doing. What Google type outfits are doing is less easily looked up, documented, and analyzed.
Stephen E Arnold, May 30, 2023
Smart Software Knows Right from Wrong
May 29, 2023
Note: This essay is the work of a real and still-alive dinobaby. No smart software involved, just a dumb humanoid.
The AI gold rush is underway. I am not sure if the gold is the stuff of the King’s crown or one of those NFT confections. I am not sure what company will own the mine or sell the miner’s pants with rivets. But gold rush days remind me of forced labor (human indexers), claim jumping (hiring experts from one company to advantage another), and hydraulic mining (ethical and moral world enhancement). Yes, I see some parallels.
I thought of claim jumping and morals after reading “OpenAI Competitor Says Its Chatbot Has a Rudimentary Conscience: A Fascinating Concept.” The following snippet from the article caught my attention:
Jared Kaplan, a former OpenAI research consultant who went on to found Anthropic with a group of his former coworkers, told Wired that Claude is, in essence, learning right from wrong because its training protocols are “basically reinforcing the behaviors that are more in accord with the constitution, and discourages behaviors that are problematic.”
Please, read the original.
I want to capture several thoughts which flitted through my humanoid mind:
- What is right? What is wrong?
- What yardstick will be used to determine “rightness” or “wrongness.”
- What is the context for each right or wrong determination; for example, at the National Criminal Justice Training Center, there is a concept called “sexploitation.” The moral compass of You.com prohibits searching for information related to this trendy criminal activity? How will the Anthropic approach address the issue of a user with a “right” intent from a user with a “wrong” intent?
Net net: Baloney. Services will do what’s necessary to generate revenue. I know from watching the trajectories of the Big Tech outfits that right, wrong, ethics, and associated dorm room discussions wobble around and focus on getting rich or just having a job.
The goal for some will be to get their fingers on the knobs and control levers. Right or wrong?
Stephen E Arnold, May 29, 2023
Shall We Train Smart Software on Scientific Papers? That Is an Outstanding Idea!
May 29, 2023
Note: This essay is the work of a real and still-alive dinobaby. No smart software involved, just a dumb humanoid.
I read “Fake Scientific Papers Are Alarmingly Common. But New Tools Show Promise in Tackling Growing Symptom of Academia’s Publish or Perish Culture.” New tools sounds great. Navigate to the cited document to get the “real” information.
MidJourney’s representation of a smart software system ingesting garbage and outputting garbage.
My purpose in mentioning this article is to ask a question:
In the last five years how many made up, distorted, or baloney filled journal articles have been produced?
The next question is,
How many of these sci-fi confections of scholarly research have been identified and discarded by the top smart software outfits like Facebook, Google, OpenAI, et al?
Let’s assume that 25 percent of the journal content is fakery.
A question I have is:
How does faked information impact the outputs of the smart software systems?
I can anticipate some answers; for example, “Well, there are a lot of papers so the flawed papers will represent a small portion of the intake data. The law of large numbers or some statistical jibber jabber will try to explain away erroneous information. Remember. Bad information is part of the human landscape. Does this mean smart software is a mirror of errors?
Do smart software outfits remove flawed information? If the peer review process cannot, what methods are the smart outfits using. Perhaps these companies should decide what’s correct and what’s incorrect? That sounds like a Googley-type idea, doesn’t it?
And finally, the third question about the impact of bad information on smart software “outputs” has an answer. No, it is not marketing jargon or a recycling of Google’s seven wonders of the AI world.
The answer, in my opinion, is garbage in and garbage out.
But you knew that, right?
Stephen E Arnold, Mary 29, 2023
The Return: IBM Watsonx!
May 26, 2023
![Vea4_thumb_thumb_thumb_thumb_thumb_t[1]](http://arnoldit.com/wordpress/wp-content/uploads/2023/05/Vea4_thumb_thumb_thumb_thumb_thumb_t1-5.gif "Vea4_thumb_thumb_thumb_thumb_thumb_t[1]")
Note: This essay is the work of a real and still-alive dinobaby. No smart software involved, just a dumb humanoid.
It is no surprise IBM’s entry into the recent generative AI hubbub is a version of Watson, the company’s longtime algorithmic representative. Techspot reports, “IBM Unleashes New AI Strategy with ‘watsonx’.” The new suite of tools was announced at the company’s recent Think conference. Note “watsonx” is not interchangeable with “Watson.” The older name with the capital letter and no trendy “x” is to be used for tools individuals rather than company-wide software. That won’t be confusing at all. Writer Bob O’Donnell describes the three components of watsonx:
“Watsonx.ai is the core AI toolset through which companies can build, train, validate and deploy foundation models. Notably, companies can use it to create original models or customize existing foundation models. Watsonx.data, is a datastore optimized for AI workloads that’s used to gather, organize, clean and feed data sources that go into those models. Finally, watsonx.governance is a tool for tracking the process of the model’s creation, providing an auditable record of all the data going into the model, how it’s created and more.Another part of IBM’s announcement was the debut of several of its own foundation models that can be used with the watsonx toolset or on their own. Not unlike others, IBM is initially unveiling a LLM-based offering for text-based applications, as well as a code generating and reviewing tool. In addition, the company previewed that it intends to create some additional industry and application-specific models, including ones for geospatial, chemistry, and IT operations applications among others. Critically, IBM said that companies can run these models in the cloud as a service, in a customer’s own data center, or in a hybrid model that leverages both. This is an interesting differentiation because, at the moment, most model providers are not yet letting organizations run their models on premises.”
Just to make things confusing, er, offer more options, each of these three applications will have three different model architectures. On top of that, each of these models will be available with varying numbers of parameters. The idea is not, as it might seem, to give companies decision paralysis but to provide flexibility in cost-performance tradeoffs and computing requirements. O’Donnell notes watsonx can also be used with open-source models, which is helpful since many organizations currently lack staff able build their own models.
The article notes that, despite the announcement’s strategic timing, it is clear watsonx marks a change in IBM’s approach to software that has been in the works for years: generative AI will be front and center for the foreseeable future. Kinda like society as a whole, apparently.
Cynthia Murrell, May 26, 2023
OpenAI Clarifies What “Regulate” Means to the Sillycon Valley Crowd
May 25, 2023
![Vea4_thumb_thumb_thumb_thumb_thumb_t[1]](http://arnoldit.com/wordpress/wp-content/uploads/2023/05/Vea4_thumb_thumb_thumb_thumb_thumb_t1-10.gif "Vea4_thumb_thumb_thumb_thumb_thumb_t[1]")
Note: This essay is the work of a real and still-alive dinobaby. No smart software involved, just a dumb humanoid.
Sam AI-man begged (at least he did not get on his hands and knees) the US Congress to regulate artificial intelligence (whatever that means). I just read “Sam Altman Says OpenAI Will Leave the EU if There’s Any Real AI Regulation.” I know I am old. I know I lose my car keys a couple of times every 24 hours. I do recall Mr. AI-man wanted regulation.
However, the write up reports:
Though unlike in the AI-friendly U.S., Altman has threatened to take his big tech toys to the other end of the sandbox if they’re not willing to play by his rules.
The vibes of the Zuckster zip through my mind. Facebook just chugs along, pays fines, and mostly ignores regulators. China seems to be an exception for Facebook, the Google, and some companies I don’t know about. China had a mobile death-mobile. A person accused and convicted would be executed in the mobile death van as soon as it arrived at the location where the convicted bad actor was. Re-education camps and mobile death-mobiles suggest that some US companies choose to exit China. Lawyers who have to arrive quickly or their client has been processed are not much good in some of China’s efficient state machines. Fines, however, are okay. Write a check and move on.
Mr. AI-man is making clear that the word “regulate” means one thing to Mr. AI-man and another thing to those who are not getting with the smart software program. The write up states:
Altman said he didn’t want any regulation that restricted users’ access to the tech. He told his London audience he didn’t want anything that could harm smaller companies or the open source AI movement (as a reminder, OpenAI is decidedly more closed off as a company than it’s ever been, citing “competition”). That’s not to mention any new regulation would inherently benefit OpenAI, so when things inevitably go wrong it can point to the law to say they were doing everything they needed to do.
I think “regulate” means what the declining US fast food outfit who told me “have it your way” meant. The burger joint put in a paper bag whatever the professionals behind the counter wanted to deliver. Mr. AI-man doesn’t want any “behind the counter” decision making by a regulatory cafeteria serving up its own version of lunch.
Mr. AI-man wants “regulate” to mean his way.
In the US, it seems, that is exactly what big tech and promising venture funded outfits are going to get; that is, whatever each company wants. Competition is good. See how well OpenAI and Microsoft are competing with Facebook and Google. Regulate appears to mean “let us do what we want to do.”
I am probably wrong. OpenAI, Google, and other leaders in smart software are at this very moment consuming the Harvard Library of books to read in search of information about ethical behavior. The “moral” learning comes later.
Net net: Now I understand the new denotation of “regulate.” Governments work for US high-tech firms. Thus, I think the French term laissez-faire nails it.
Stephen E Arnold, May 25, 2023
Top AI Tools for Academic Researchers
May 25, 2023
![Vea4_thumb_thumb_thumb_thumb_thumb_t[1]](http://arnoldit.com/wordpress/wp-content/uploads/2023/05/Vea4_thumb_thumb_thumb_thumb_thumb_t1-4.gif "Vea4_thumb_thumb_thumb_thumb_thumb_t[1]")
Note: This essay is the work of a real and still-alive dinobaby. No smart software involved, just a dumb humanoid.
Like it or not, AI is reshaping academia. Setting aside the thorny issue of cheating, we see the technology is also changing how academic research is performed. In fact, there are already many AI options for researchers to pick from. Euronews narrows down the choices in, “The Best AI Tools to Power your Academic Research.” Writer Camille Bello shares the top five options as chosen by Mushtaq Bilal, a researcher at the University of Southern Denmark. She introduces the list with an important caveat: one must approach these tools carefully for accurate results. She writes:
“[Bilal] believes that if used thoughtfully, AI language models could help democratise education and even give way to more knowledge. Many experts have pointed out that the accuracy and quality of the output produced by language models such as ChatGPT are not trustworthy. The generated text can sometimes be biased, limited or inaccurate. But Bilal says that understanding those limitations, paired with the right approach, can make language models ‘do a lot of quality labour for you,’ notably for academia.
Incremental prompting to create a ‘structure’
To create an academia-worthy structure, Bilal says it is fundamental to master incremental prompting, a technique traditionally used in behavioural therapy and special education. It involves breaking down complex tasks into smaller, more manageable steps and providing prompts or cues to help the individual complete each one successfully. The prompts then gradually become more complicated. In behavioural therapy, incremental prompting allows individuals to build their sense of confidence. In language models, it allows for ‘way more sophisticated answers’. In a Twitter thread, Bilal showed how he managed to get ChatGPT to provide a ‘brilliant outline’ for a journal article using incremental prompting.”
See the write-up for this example of incremental prompting as well as a description of each entry on the list. The tools include: Consensus, Elicit.org, Scite.ai, Research Rabbit, and ChatPDF. The write-up concludes with a quote from Bill Gates, who asserts AI is destined to be as fundamental “as the creation of the microprocessor, the personal computer, the Internet, and the mobile phone.” Researchers may do well to embrace the technology sooner rather than later.
Cynthia Murrell, May 25, 2023
Google AI Moves Slowly to Google Advertising. Soon, That Is. Soon.
May 24, 2023
![Vea4_thumb_thumb_thumb_thumb_thumb_t[1]](http://arnoldit.com/wordpress/wp-content/uploads/2023/05/Vea4_thumb_thumb_thumb_thumb_thumb_t1-9.gif "Vea4_thumb_thumb_thumb_thumb_thumb_t[1]")
Note: This essay is the work of a real and still-alive dinobaby. No smart software involved, just a dumb humanoid.
I read l ”Google Search Ads Will Soon Automatically Adapt to Queries Using Generative AI.” The idea of using smart software to sell ads is one that seems obvious to me. What surprised me about this article in TechCrunch is the use of the future tense and the indefinite “soon.” The Sundar Financial Times’ PR write up emphasized that Google has been doing smart software for a looooong time.
How could a company so dependent on ads be in the “will” and “soon” vaporware announcement business?
I noted this passage in the write up:
Google is going to start using generative AI to boost Search ads’ relevance based on the context of a query…
But why so slow in releasing obvious applications of generative software?
I don’t have answers to this quite Googley question, probably asked by those engaged in the internal discussions about who’s on first in the Google Brain versus DeepMind softball game, but I have some observations:
- Google had useful technology but lacked the administrative and managerial expertise to get something out the door and into the hands paying customers
- Google’s management processes simply do not work when the company is faced with strategic decisions. This signals the end of the go go mentality of the company’s Backrub to Google transformation. And it begs the question, “What else has the company lost over the last 25 years?”
- Google’s engineers cannot move from Ivory Tower quantum supremacy mental postures to common sense applications of technology to specific use cases.
In short, after 25 years Googzilla strikes me as arthritic when it comes to hot technology and a little more nimble when it tries to do PR. Except for Paris, of course.
Stephen E Arnold, May 24, 2023
Sam AI-man Begs for Regulation; China Takes Action for Structured Data LLM Integration
May 24, 2023
Note: This essay is the work of a real and still-alive dinobaby. No smart software involved, just a dumb humanoid.
Smart software is capturing attention from a number of countries’ researchers. The US smart software scene is crowded like productions on high school auditoria stages. Showing recently was OpenAI’s really sincere plea for regulation, oodles of new smart software applications and plug ins for browsers, and Microsoft’s assembly line of AI everywhere in Office 365. The venture capital contingent is chanting, “Who wants cash? Who wants cash?” Plus the Silicon Valley media are beside themselves with in-crowd interviews with the big Googler and breathless descriptions of how college professors fumble forward with students who may or may not let ChatGPT do that dumb essay.
US Silicon Valley deciders in action in public discuss the need for US companies to move slowly, carefully, judiciously when deploying AI. In private, these folks want to go as quickly as possible, lock up markets, and rake in the dough. China skips the pretending and just goes forward with certain guidelines to avoid a fun visit to a special training facility. The illustration was created by MidJourney, a service which I assume wants to be regulated at least sometimes.
In the midst of this vaudeville production, I noted “Researchers from China Propose StructGPT to Improve the Zero-Shot Reasoning Ability of LLMs over Structured Data.” On the surface, the write up seems fairly tame by Silicon Valley standards. In a nutshell, whiz kids from a university I never heard of figure out how to reformat data in a database table and make those data available to a ChatGPT type system. The idea is that ChatGPT has some useful qualities. Being highly accurate is not a core competency, however.
The good news is that the Chinese researchers have released some of their software and provided additional information on GitHub. Hopefully American researchers can take time out from nifty plug ins, begging regulators to regulate, and banking bundles of pre-default bucks in JPMorgan accounts.
For me, the article makes clear:
- Whatever the US does, China is unlikely to trim the jib of technologies which can generate revenue, give China an advantage, and provide some new capabilities to its military.
- US smart software vendors have no choice but go full speed ahead and damn the AI powered torpedoes from those unfriendly to the “move fast and break things” culture. What’s a regulator going to do? I know. Have a meeting.
- Smart software is many things. I perceive what can be accomplished with what I can download today and maybe some fiddling with the Renmin University of China, Beijing Key Laboratory of Big Data Management and Analysis Methods, and the University of Electronic Science and Technology of China method is a great big trampoline. Those jumping can do some amazing and hitherto unseen tricks.
Net net: Less talk and more action, please.
Stephen E Arnold, May 24, 2023
AI Builders and the Illusions they Promote
May 24, 2023
![Vea4_thumb_thumb_thumb_thumb_thumb_t[1]](http://arnoldit.com/wordpress/wp-content/uploads/2023/05/Vea4_thumb_thumb_thumb_thumb_thumb_t1-3.gif "Vea4_thumb_thumb_thumb_thumb_thumb_t[1]")
Note: This essay is the work of a real and still-alive dinobaby. No smart software involved, just a dumb humanoid.
Why do AI firms insist on calling algorithmic mistakes “hallucinations” instead of errors, malfunctions, or glitches? The Guardian‘s Naomi Klein believes AI advocates chose this very human and mystical term to perpetuate a fundamental myth: AI will be humanity’s salvation. And that stance, she insists, demonstrates that “AI Machines Aren’t ‘Hallucinating.’ But their Makers Are.”
It is true that, in a society built around citizens’ well-being and the Earth’s preservation, AI could help end poverty, eliminate disease, reverse climate change, and facilitate more meaningful lives. But that is not the world we live in. Instead, our systems are set up to exploit both resources and people for the benefit of the rich and powerful. AI is poised to help them do that even more efficiently than before.
The article discusses four specific hallucinations possessing AI proponents. First, the assertion AI will solve the climate crisis when it is likely to do just the opposite. Then there’s the hope AI will help politicians and bureaucrats make wiser choices, which assumes those in power base their decisions on the greater good in the first place. Which leads to hallucination number three, that we can trust tech giants “not to break the world.” Those paying attention saw that was a false hope long ago. Finally is the belief AI will eliminate drudgery. Not all work, mind you, just the “boring” stuff. Some go so far as to paint a classic leftist ideal, one where humans work not to survive but to pursue our passions. That might pan out if we were living in a humanist, Star Trek-like society, Klein notes, but instead we are subjects of rapacious capitalism. Those who lose their jobs to algorithms have no societal net to catch them.
So why are the makers of AI promoting these illusions? Kelin proposes:
“Here is one hypothesis: they are the powerful and enticing cover stories for what may turn out to be the largest and most consequential theft in human history. Because what we are witnessing is the wealthiest companies in history (Microsoft, Apple, Google, Meta, Amazon …) unilaterally seizing the sum total of human knowledge that exists in digital, scrapable form and walling it off inside proprietary products, many of which will take direct aim at the humans whose lifetime of labor trained the machines without giving permission or consent. This should not be legal. In the case of copyrighted material that we now know trained the models (including this newspaper), various lawsuits have been filed that will argue this was clearly illegal. Why, for instance, should a for-profit company be permitted to feed the paintings, drawings and photographs of living artists into a program like Stable Diffusion or Dall-E 2 so it can then be used to generate doppelganger versions of those very artists’ work, with the benefits flowing to everyone but the artists themselves?”
The answer, of course, is that this should not be permitted. But since innovation moves much faster than legislatures and courts, tech companies have been operating on a turbo-charged premise of seeking forgiveness instead of permission for years. (They call it “disruption,” Klein notes.) Operations like Google’s book-scanning project, Uber’s undermining the taxi industry, and Facebook’s mishandling of user data, just to name a few, got so far so fast regulators simply gave in. Now the same thing appears to be happening with generative AI and the data it feeds upon. But there is hope. A group of top experts on AI ethics specify measures regulators can take. Will they?
Cynthia Murrell, May 24, 2023
-
- Subscribe to Beyond Search
Feature archive
News archive
-
