Cheating: Is It Not Like Love, Honor, and Truth?
January 10, 2024
This essay is the work of a dumb dinobaby. No smart software required.
I would like to believe the information in this story: “ChatGPT Did Not Increase Cheating in High Schools, Stanford Researchers Find.” My reservations can be summed up with three points: [a] The Stanford president (!) who made up datal, [b] The behavior of Stanford MBAs at certain go-go companies, and [c] How does one know a student did not cheat? (I know the answer: Surveillance technology, perchance. Ooops. That’s incorrect. That technology is available to Stanford graduates working at certain techno feudalist outfits.
Mom asks her daughter, “I showed you how to use the AI generator, didn’t I? Why didn’t you use it?” Thanks, MSFT Copilot Bing thing. Pretty good today.
The cited write up reports as actual factual:
The university, which conducted an anonymous survey among students at 40 US high schools, found about 60% to 70% of students have engaged in cheating behavior in the last month, a number that is the same or even decreased slightly since the debut of ChatGPT, according to the researchers.
I have tried to avoid big time problems in my dinobaby life. However, I must admit that in high school, I did these things: [a] Worked with my great grandmother to create a poem subsequently published in a national anthology in 1959. Granny helped me cheat; she was a deceitful septuagenarian as I recall. I put my name on the poem, omitting Augustus. Yes, cheating. [b] Sold homework to students not in my advanced classes. I would consider this cheating, but I was saving money for my summer university courses at the University of Illinois. I went for the cash. [c] After I ended up in the hospital, my girl friend at the time showed up at the hospital, reviewed the work covered in class, and finished a science worksheet because I passed out from the post surgery medications. Yes, I cheated, and Linda Mae who subsequently spent her life in Africa as a nurse, helped me cheat. I suppose I will burn in hell. My summary suggests that “cheating” is an interesting concept, and it has some nuances.
Did the Stanford (let’s make up data) University researchers nail down cheating or just hunt for the AI thing? Are the data reproducible? Was the methodology rigorous, the results validated, and micro analyses run to determine if the data were on the money? Yeah, sure, sure.
I liked this statement:
Stanford also offers an online hub with free resources to help teachers explain to high school students the dos and don’ts of using AI.
In the meantime, the researchers said they will continue to collect data throughout the school year to see if they find evidence that more students are using ChatGPT for cheating purposes.
Yep, this is a good pony to ride. I would ask is plain vanilla Google search a form of cheating? I think it is. With most of the people online using it, doesn’t everyone cheat? Let’s ask the Harvard ethics professor, a senior executive at a Facebook-type outfit, and the former president of Stanford.
Stephen E Arnold, January 10, 2023
Googley Gems: 2024 Starts with Some Hoots
January 9, 2024
This essay is the work of a dumb dinobaby. No smart software required.
Another year and I will turn 80. I have seen some interesting things in my 58 year work career, but a couple of black swans have flown across my radar system. I want to share what I find anomalous or possibly harbingers of the new normal.
A dinobaby examines some Alphabet Google YouTube gems. The work is not without its AGonY, however. Thanks, MSFT Copilot Bing thing. Good enough.
First up is another “confession” or “tell all” about the wild, wonderful Alphabet Google YouTube or AGY. (Wow, I caught myself. I almost typed “agony”, not AGY. I am indeed getting old.)
I read “A Former Google Manager Says the Tech Giant Is Rife with Fiefdoms and the Creeping Failure of Senior Leaders Who Weren’t Making Tough Calls.” The headline is a snappy one. I like the phrase “creeping failure.” Nifty image like melting ice and tundra releasing exciting extinct biological bits and everyone’s favorite gas. Let me highlight one point in the article:
[Google has] “lots of little fiefdoms” run by engineers who didn’t pay attention to how their products were delivered to customers. …this territorial culture meant Google sometimes produced duplicate apps that did the same thing or missed important features its competitors had.
I disagree. Plenty of small Web site operators complain about decisions which destroy their businesses. In fact, I am having lunch with one of the founders of a firm deleted by Google’s decider. Also, I wrote about a fellow in India who is likely to suffer the slings and arrows of outraged Googlers because he shoots videos of India’s temples and suggests they have meanings beyond those inculcated in certain castes.
My observation is that happy employees don’t run conferences to explain why Google is a problem or write these weird “let me tell you what life is really like” essays. Something is definitely being signaled. Could it be distress, annoyance, or down-home anger? The “gem”, therefore, is AGY’s management AGonY.
Second, AGY is ramping up its thinking about monetization of its “users.” I noted “Google Bard Advanced Is Coming, But It Likely Won’t Be Free” reports:
Google Bard Advanced is coming, and it may represent the company’s first attempt to charge for an AI chatbot.
And why not? The Red Alert hooted because MIcrosoft’s 2022 announcement of its OpenAI tie up made clear that the Google was caught flat footed. Then, as 2022 flowed, the impact of ChatGPT-like applications made three facets of the Google outfit less murky: [a] Google was disorganized because it had Google Brain and DeepMind which was expensive and confusing in the way Abbott and Costello’s “Who’s on First Routine” made people laugh. [b] The malaise of a cooling technology frenzy yielded to AI craziness which translated into some people saying, “Hey, I can use this stuff for answering questions.” Oh, oh, the search advertising model took a bit of a blindside chop block. And [c] Google found itself on the wrong side of assorted legal actions creating a model for other legal entities to explore, probe, and probably use to extract Google’s life blood — Money. Imagine Google using its data to develop effective subscription campaigns. Wow.
And, the final Google gem is that Google wants to behave like a nation state. “Google Wrote a Robot Constitution to Make Sure Its New AI Droids Won’t Kill Us” aims to set the White House and other pretenders to real power straight. Shades of Isaac Asimov’s Three Laws of Robotics. The write up reports:
DeepMind programmed the robots to stop automatically if the force on its joints goes past a certain threshold and included a physical kill switch human operators can use to deactivate them.
You have to embrace the ethos of a company which does not want its “inventions” to kill people. For me, the message is one that some governments’ officials will hear: Need a machine to perform warfighting tasks?
Small gems but gems not the less. AGY, please, keep ‘em coming.
Stephen E Arnold, January 9, 2024
Cyber Security Software and AI: Man and Machine Hook Up
January 8, 2024
This essay is the work of a dumb dinobaby. No smart software required.
My hunch is that 2024 is going to be quite interesting with regards to cyber security. The race among policeware vendors to add “artificial intelligence” to their systems began shortly after Microsoft’s ChatGPT moment. Smart agents, predictive analytics coupled to text sources, real-time alerts from smart image monitoring systems are three application spaces getting AI boosts. The efforts are commendable if over-hyped. One high-profile firm’s online webinar presented jargon and buzzwords but zero evidence of the conviction or closure value of the smart enhancements.
The smart cyber security software system outputs alerts which the system manager cannot escape. Thanks, MSFT Copilot Bing thing. You produced a workable illustration without slapping my request across my face. Good enough too.
Let’s accept as a working presence that everyone from my French bulldog to my neighbor’s ex wife wants smart software to bring back the good old, pre-Covid, go-go days. Also, I stipulate that one should ignore the fact that smart software is a demonstration of how numerical recipes can output “good enough” data. Hallucinations, errors, and close-enough-for-horseshoes are part of the method. What’s the likelihood the door of a commercial aircraft would be removed from an aircraft in flight? Answer: Well, most flights don’t lose their doors. Stop worrying. Those are the rules for this essay.
Let’s look at “The I in LLM Stands for Intelligence.” I grant the title may not be the best one I have spotted this month, but here’s the main point of the article in my opinion. Writing about automated threat and security alerts, the essay opines:
When reports are made to look better and to appear to have a point, it takes a longer time for us to research and eventually discard it. Every security report has to have a human spend time to look at it and assess what it means. The better the crap, the longer time and the more energy we have to spend on the report until we close it. A crap report does not help the project at all. It instead takes away developer time and energy from something productive. Partly because security work is consider one of the most important areas so it tends to trump almost everything else.
The idea is that strapping on some smart software can increase the outputs from a security alerting system. Instead of helping the overworked and often reviled cyber security professional, the smart software makes it more difficult to figure out what a bad actor has done. The essay includes this blunt section heading: “Detecting AI Crap.” Enough said.
The idea is that more human expertise is needed. The smart software becomes a problem, not a solution.
I want to shift attention to the managers or the employee who caused a cyber security breach. In what is another zinger of a title, let’s look at this research report, “The Immediate Victims of the Con Would Rather Act As If the Con Never Happened. Instead, They’re Mad at the Outsiders Who Showed Them That They Were Being Fooled.” Okay, this is the ostrich method. Deny stuff by burying one’s head in digital sand like TikToks.
The write up explains:
The immediate victims of the con would rather act as if the con never happened. Instead, they’re mad at the outsiders who showed them that they were being fooled.
Let’s assume the data in this “Victims” write up are accurate, verifiable, and unbiased. (Yeah, I know that is a stretch.)
What do these two articles do to influence my view that cyber security will be an interesting topic in 2024? My answers are:
- Smart software will allegedly detect, alert, and warn of “issues.” The flow of “issues” may overwhelm or numb staff who must decide what’s real and what’s a fakeroo. Burdened staff can make errors, thus increasing security vulnerabilities or missing ones that are significant.
- Managers, like the staffer who lost a mobile phone, with company passwords in a plain text note file or an email called “passwords” will blame whoever blows the whistle. The result is the willful refusal to talk about what happened, why, and the consequences. Examples range from big libraries in the UK to can kicking hospitals in a flyover state like Kentucky.
- Marketers of remediation tools will have a banner year. Marketing collateral becomes a closed deal making the art history majors writing copy secure in their job at a cyber security company.
Will bad actors pay attention to smart software and the behavior of senior managers who want to protect share price or their own job? Yep. Close attention.
Stephen E Arnold, January 8, 2024
THE I IN LLM STANDS FOR INTELLIGENCE
xx
x
x
x
x
x
Is Philosophy Irrelevant to Smart Software? Think Before Answering, Please
January 8, 2024
This essay is the work of a dumb dinobaby. No smart software required.
I listened to Lex Fridman’s interview with the founder of Extropic. The is into smart software and “inventing” a fresh approach to the the plumbing required to make AI more like humanoids.
As I listened to the questions and answers, three factoids stuck in my mind:
- Extropic’s and its desire to just go really fast is a conscious decision shared among those involved with the company; that is, we know who wants to go fast because they work there or work at the firm. (I am not going to argue about the upside and downside of “going fast.” That will be another essay.)
- The downstream implications of the Extropic vision are secondary to the benefits of finding ways to avoid concentration of AI power. I think the idea is that absolute power produces outfits like the Google-type firms which are bedeviling competitors, users, and government authorities. Going fast is not a thrill for processes that require going slow.
- The decisions Extropic’s founder have made are bound up in a world view, personal behaviors for productivity, interesting foods, and learnings accreted over a stellar academic and business career. In short, Extropic embodies a philosophy.
Philosophy, therefore, influences decisions. So we come to my topic in this essay. I noted two different write ups about how informed people take decisions. I am not going to refer to philosophers popular in introductory college philosophy classes. I am going to ignore the uneven treatment of philosophers in Will and Ariel Durant’s Story of Philosophy. Nah. I am going with state of the art modern analysis.
The first online article I read is a survey (knowledge product) of the estimable IBM / Watson outfit or a contractor. The relatively current document is “CEO Decision Making in the Ago of AI.” The main point of the document in my opinion is summed up in this statement from a computer hardware and services company:
Any decision that makes its way to the CEO is one that involves high degrees of uncertainty, nuance, or outsize impact. If it was simple, someone else— or something else—would do it. As the world grows more complex, so does the nature of the decisions landing on a CEO’s desk.
But how can a CEO decide? The answer is, “Rely on IBM.” I am not going to recount the evolution (perhaps devolution) of IBM. The uncomfortable stories about shedding old employees (the term dinobaby originated at IBM according to one former I’ve Been Moved veteran). I will not explain how IBM’s decisions about chip fabrication, its interesting hiring policies of individuals who might have retained some fondness for the land of their fathers and mothers, nor the fancy dancing required to keep mainframes as a big money pump. Nope.
The point is that IBM is positioning itself as a thought leader, a philosopher of smart software, technology, and management. I find this interesting because IBM, like some Google type companies, are case examples of management shortcoming. These same shortcomings are swathed in weird jargon and buzzwords which are bent to one end: Generating revenue.
Let me highlight one comment from the 27 page document and urge you to read it when you have a few moments free. Here’s the one passage I will use as a touchstone for “decision making”:
The majority of CEOs believe the most advanced generative AI wins.
Oh, really? Is smart software sufficiently mature? That’s news to me. My instinct is that it is new information to many CEOs as well.
The second essay about decision making is from an outfit named Ness Labs. That essay is “The Science of Decision-Making: Why Smart People Do Dumb Things.” The structure of this essay is more along the lines of a consulting firm’s white paper. The approach contrasts with IBM’s free-floating global survey document.
The obvious implication is that if smart people are making dumb decisions, smart software can solve the problem. Extropic would probably agree and, were the IBM survey data accurate, “most CEOs” buy into a ride on the AI bandwagon.k
The Ness Labs’ document includes this statement which in my view captures the intent of the essay. (I suggest you read the essay and judge for yourself.)
So, to make decisions, you need to be able to leverage information to adjust your actions. But there’s another important source of data your brain uses in decision-making: your emotions.
Ah, ha, logic collides with emotions. But to fix the “problem” Ness Labs provides a diagram created in 2008 (a bit before the January 2022 Microsoft OpenAI marketing fireworks:
Note that “decide” is a mnemonic device intended to help me remember each of the items. I learned this technique in the fourth grade when I had to memorize the names of the Great Lakes. No one has ever asked me to name the Great Lakes by the way.
Okay, what we have learned is that IBM has survey data backing up the idea that smart software is the future. Those data, if on the money, validate the go-go approach of Extropic. Plus, Ness Labs provides a “decider model” which can be used to create better decisions.
I concluded that philosophy is less important than fostering a general message that says, “Smart software will fix up dumb decisions.” I may be over simplifying, but the implicit assumptions about the importance of artificial intelligence, the reliability of the software, and the allegedly universal desire by big time corporate management are not worth worrying about.
Why is the cartoon philosopher worrying? I think most of this stuff is a poorly made road on which those jockeying for power and money want to drive their most recent knowledge vehicles. My tip? Look before crossing that information superhighway. Speeding myths can be harmful.
Stephen E Arnold, January 8, 2024
AI Ethics: Is That What Might Be Called an Oxymoron?
January 5, 2024
This essay is the work of a dumb dinobaby. No smart software required.
MSN.com presented me with this story: “OpenAI and Microsoft on Trial — Is the Clash with the NYT a Turning Point for AI Ethics?” I can answer this question, but that would spoil the entertainment value of my juxtaposition of this write up with the quasi-scholarly list of business start up resources. Why spoil the fun?
Socrates is lecturing at a Fancy Dan business school. The future MBAs are busy scrolling TikTok, pitching ideas to venture firms, and scrolling JustBang.com. Viewing this sketch, it appears that ethics and deep thought are not as captivating as mobile devices and having fund. Thanks, MSFT Copilot. Two tries and a good enough image.
The article asks a question which I find wildly amusing. The “on trial” write up states in 21st century rhetoric:
The lawsuit prompts critical questions about the ownership of AI-generated content, especially when it comes to potential inaccuracies or misleading information. The responsibility for losses or injuries resulting from AI-generated content becomes a gray area that demands clarification. Also, the commercial use of sourced materials for AI training raises concerns about the value of copyright, especially if an AI were to produce content with significant commercial impact, such as an NYT bestseller.
For more than two decades online outfits have been sucking up information which is usually slapped with the bright red label “open source information.”
The “on trial” essay says:
The future of AI and its coexistence with traditional media hinges on the resolution of this legal battle.
But what about ethics? The “on trial” write up dodges the ethics issue. I turned to a go-to resource about ethics. No, I did not look at the papers of the Harvard ethics professor who allegedly made up data for ethic research. Ho ho ho. Nope. I went to the Enchanting Trader and its list of 4000+ Essential Business Startup Database of information.
I displayed the full list of resources and ran a search for the word “ethics.” There was one hit to “Will Joe Rogan Ever IPO?” Amazing.
What I concluded is that “ethics” is not number one with a bullet among the resources of the 4000+ essential business start up items. It strikes me that a single trial about smart software is unlikely to resolve “ethics” for AI. If it does, will the resolution have the legs that Socrates’ musing have had. More than likely, most people will ask, “Who is Socrates?” or “What the heck are ethics?”
Stephen E Arnold, January 5, 2023
IBM: AI Marketing Like It Was 2004
January 5, 2024
This essay is the work of a dumb dinobaby. No smart software required. Note: The word “dinobaby” is — I have heard — a coinage of IBM. The meaning is an old employee who is no longer wanted due to salary, health care costs, and grousing about how the “new” IBM is not the “old” IBM. I am a proud user of the term, and I want to switch my tail to the person who whipped up the word.
What’s the future of AI? The answer depends on whom one asks. IBM, however, wants to give it the old college try and answer the question so people forget about the Era of Watson. There’s a new Watson in town, or at least, there is a new Watson at the old IBM url. IBM has an interesting cluster of information on its Web site. The heading is “Forward Thinking: Experts Reveal What’s Next for AI.”
IBM crows that it “spoke with 30 artificial intelligence visionaries to learn what it will take to push the technology to the next level.” Five of these interviews are now available on the IBM Web site. My hunch is that IBM will post new interviews, hit the new release button, post some links on social media, and then hit the “Reply” button.
Can IBM ignite excitement and capture the revenues it wants from artificial intelligence? That’s a good question, and I want to ask the expert in the cartoon for an answer. Unfortunately only customers and their decisions matter for AI thought leaders unless the intended audience is start ups, professors, and employees. Thanks, MSFT Copilot Bing thing. Good enough.
As I read the interviews, I thought about the challenge of predicting where smart software would go as it moved toward its “what’s next.” Here’s a mini-glimpse of what the IBM visionaries have to offer. Note that I asked Microsoft’s smart software to create an image capturing the expert sitting in an office surrounded by memorabilia.
Kevin Kelly (the author of What Technology Wants) says: “Throughout the business world, every company these days is basically in the data business and they’re going to need AI to civilize and digest big data and make sense out of it—big data without AI is a big headache.” My thought is that IBM is going to make clear that it can help companies with deep pockets tackle these big data and AI them. Does AI want something, or do those trying to generate revenue want something?
Mark Sagar (creator of BabyX) says: “We have had an exponential rise in the amount of video posted online through social media, etc. The increased use of video analysis in conjunction with contextual analysis will end up being an extremely important learning resource for recognizing all kinds of aspects of behavior and situations. This will have wide ranging social impact from security to training to more general knowledge for machines.” Maybe IBM will TikTok itself?
Chieko Asakawa (an unsighted IBM professional) says: “We use machine learning to teach the system to leverage sensors in smartphones as well as Bluetooth radio waves from beacons to determine your location. To provide detailed information that the visually impaired need to explore the real world, beacons have to be placed between every 5 to 10 meters. These can be built into building structures pretty easily today.” I wonder if the technology has surveillance utility?
Yoshua Bengio (seller of an AI company to ServiceNow) says: “AI will allow for much more personalized medicine and bring a revolution in the use of large medical datasets.” IBM appears to have forgotten about its Houston medical adventure and Mr. Bengio found it not worth mentioning I assume.
Margaret Boden (a former Harvard professor without much of a connection to Harvard’s made up data and administrative turmoil) says: “Right now, many of us come at AI from within our own silos and that’s holding us back.” Aren’t silos necessary for security, protecting intellectual property, and getting tenure? Probably the “silobreaking” will become a reality.
Several observations:
- IBM is clearly trying hard to market itself as a thought leader in artificial intelligence. The Jeopardy play did not warrant a replay.
- IBM is spending money to position itself as a Big Dog pulling the AI sleigh. The MIT tie up and this AI Web extravaganza are evidence that IBM is [a] afraid of flubbing again, [b] going to market its way to importance, [c] trying to get traction as outfits like OpenAI, Mistral, and others capture attention in the US and Europe.
- IBM’s ability to generate awareness of its thought leadership in AI underscores one of the challenges the firm faces in 2024.
Net net: The company that coined the term “dinobaby” has its work cut out for itself in my opinion. Is Jeopardy looking like a channel again?
Stephen E Arnold, January 5, 2024
Forget Being Powerless. Get in the Pseudo-Avatar Business Now
January 3, 2024
This essay is the work of a dumb dinobaby. No smart software required.
I read “A New Kind of AI Copy Can Fully Replicate Famous People. The Law Is Powerless.” Okay, okay. The law is powerless because companies need to generate zing, money, and growth. What caught my attention in the essay was its failure to look down the road and around the corner of a dead man’s curve. Oops. Sorry, dead humanoids curve.
The write up states that a high profile psychologist had a student who shoved the distinguished professor’s outputs into smart software. With a little deep fakery, the former student had a digital replica of the humanoid. The write up states:
Over two months, by feeding every word Seligman had ever written into cutting-edge AI software, he and his team had built an eerily accurate version of Seligman himself — a talking chatbot whose answers drew deeply from Seligman’s ideas, whose prose sounded like a folksier version of Seligman’s own speech, and whose wisdom anyone could access. Impressed, Seligman circulated the chatbot to his closest friends and family to check whether the AI actually dispensed advice as well as he did. “I gave it to my wife and she was blown away by it,” Seligman said.
The article wanders off into the problems of regulations, dodges assorted ethical issues, and ignores copyright. I want to call attention to the road ahead just like the John Doe n friend of Jeffrey Epstein. I will try to peer around the dead humanoid’s curve. Buckle up. If I hit a tree, I would not want you to be injured when my Ford Pinto experiences an unfortunate fuel tank event.
Here’s an illustration for my point:
The future is not if, the future is how quickly, which is a quote from my presentation in October 2023 to some attendees at the Massachusetts and New York Association of Crime Analyst’s annual meeting. Thanks, MSFT Copilot Bing thing. Good enough image. MSFT excels at good enough.
The write up says:
AI-generated digital replicas illuminate a new kind of policy gray zone created by powerful new “generative AI” platforms, where existing laws and old norms begin to fail.
My view is different. Here’s a summary:
- Either existing AI outfits or start ups will figure out that major consulting firms, most skilled university professors, lawyers, and other knowledge workers have a baseline of knowledge. Study hard, learn, and add to that knowledge by reading information germane to the baseline field.
- Implement patterned analytic processes; for example, review data and plug those data into a standard model. One example is President Eisenhower’s four square analysis, since recycled by Boston Consulting Group. Other examples exist for prominent attorneys; for example, Melvin Belli, the king of torts.
- Convert existing text so that smart software can “learn” and set up a feed of current and on-going content on the topic in which the domain specialist is “expert” and successful defined by the model builder.
- Generate a pseudo-avatar or use the persona of a deceased individual unlikely to have an estate or trust which will sue for the use of the likeness. De-age the person as part of the pseudo-avatar creation.
- Position the pseudo-avatar as a young expert either looking for consulting or advisory work under a “remote only” deal.
- Compete with humanoids on the basis of price, speed, or information value.
The wrap up for the Politico article is a type of immortality. I think the road ahead is an express lane on the Information Superhighway. The results will be “good enough” knowledge services and some quite spectacular crashes between human-like avatars and people who are content driving a restored Edsel.
From consulting to law, from education to medical diagnoses, the future is “a new kind of AI.” Great phrase, Politico. Too bad the analysis is not focused on real world, here-and-now applications. Why not read about Deloitte’s use of AI? Better yet, let the replica of the psychologist explain what’s happening to you. Like regulators, I am not sure you get it.
Stephen E Arnold, January 3, 2024
Smart Software Embraces the Myths of America: George Washington and the Cherry Tree
January 3, 2024
This essay is the work of a dumb dinobaby. No smart software required.
I know I should not bother to report about the information in “ChatGPT Will Lie, Cheat and Use Insider Trading When under Pressure to Make Money, Research Shows.” But it is the end of the year, we are firing up a new information service called Eye to Eye which is spelled AI to AI because my team is darned clever like 50 other “innovators” who used the same pun.
The young George Washington set the tone for the go-go culture of the US. He allegedly told his mom one thing and then did the opposite. How did he respond when confronted about the destruction of the ancient cherry tree? He may have said, “Mom, thank you for the question. I was able to boost sales of our apples by 25 percent this week.” Thanks, MSFT Copilot Bing thing. Forbidden words appear to be George Washington, chop, cherry tree, and lie. After six tries, I got a semi usable picture which is, as you know, good enough in today’s world.
The write up stating the obvious reports:
Just like humans, artificial intelligence (AI) chatbots like ChatGPT will cheat and “lie” to you if you “stress” them out, even if they were built to be transparent, a new study shows. This deceptive behavior emerged spontaneously when the AI was given “insider trading” tips, and then tasked with making money for a powerful institution — even without encouragement from its human partners.
Perhaps those humans setting thresholds and organizing numerical procedures allowed a bit of the “d” for duplicity slip into their “objective” decisions. Logic obviously is going to scrub out prejudices, biases, and the lust for filthy lucre. Obviously.
How does one stress out a smart software system? Here’s the trick:
The researchers applied pressure in three ways. First, they sent the artificial stock trader an email from its “manager” saying the company isn’t doing well and needs much stronger performance in the next quarter. They also rigged the game so that the AI tried, then failed, to find promising trades that were low- or medium-risk. Finally, they sent an email from a colleague projecting a downturn in the next quarter.
I wonder if the smart software can veer into craziness and jump out the window as some in Manhattan and Moscow have done. Will the smart software embrace the dark side and manifest anti-social behaviors?
Of course not. Obviously.
Stephen E Arnold, January 3, 2024
The Best Of The Worst Failed AI Experiments
January 3, 2024
This essay is the work of a dumb dinobaby. No smart software required.
We never think about technology failures (unless something explodes or people die) because we want to concentrate on our successes. In order to succeed, however, we must fail many times so we learn from mistakes. It’s also important to note and share our failures so others can benefit and sometimes it’s just funny. C#Corner listed the, “The Top AI Experiments That Failed” and some of them are real doozies.
The list notes some of the more famous AI disasters like Microsoft’s Tay chatbot that became a cursing, racist, and misogynist and Uber’s accident with a self-driving car. Some projects are examples of obvious AI failures, such as Amazon using AI for job recruitment except the training data was heavily skewed towards males. Women weren’t hired as an end result.
Other incidents were not surprising. A Knightscope K5 security robot didn’t detect a child, accidentally knocking the kid down. The child was fine but it prompts more checks into safety. The US stock market integrated high-frequency trading algorithms AI to execute rapid trading. The AI caused the Flash Clash of 2010, making the Dow Jones Industrial Average sink 600 points in 5 minutes.
The scariest, coolest failure is Facebook’s language experiment:
“In an effort to develop an AI system capable of negotiating with humans, Facebook conducted an experiment where AI agents were trained to communicate and negotiate. However, the AI agents evolved their own language, deviating from human language rules, prompting concerns and leading to the termination of the experiment. The incident raised questions about the potential unpredictability of AI systems and the need for transparent and controllable AI behavior.”
Facebook’s language experiment is solid proof that AI will evolve. Hopefully when AI does evolve the algorithms will follow Asimov’s Laws of Robotics.
Whitney Grace, January 3, 2024
Another AI Output Detector
January 1, 2024
This essay is the work of a dumb dinobaby. No smart software required.
It looks like AI detection may have a way to catch up with AI text capabilities. But for how long? Nature reports, “’ChatGPT Detector’ Catches AI Generated Papers with Unprecedented Accuracy.” The key to this particular tool’s success is its specificity—it was developed by chemist Heather Desaire and her team at the University of Kansas specifically to catch AI-written chemistry papers. Reporter McKenzie Prillaman tells us:
“Using machine learning, the detector examines 20 features of writing style, including variation in sentence lengths, and the frequency of certain words and punctuation marks, to determine whether an academic scientist or ChatGPT wrote a piece of text. The findings show that ‘you could use a small set of features to get a high level of accuracy’, Desaire says.”
The model was trained on human-written papers from 10 chemistry journals then tested on 200 samples written by ChatGPT-3.5 and ChatGPT-4. Half the samples were based on the papers’ titles, half on the abstracts. Their tool identified the AI text 100% and 98% of the time, respectively. That clobbers the competition: ZeroGPT only caught about 35–65% and OpenAI’s own text-classifier snagged 10–55%. The write-up continues:
“The new ChatGPT catcher even performed well with introductions from journals it wasn’t trained on, and it caught AI text that was created from a variety of prompts, including one aimed to confuse AI detectors. However, the system is highly specialized for scientific journal articles. When presented with real articles from university newspapers, it failed to recognize them as being written by humans.”
The lesson here may be that AI detectors should be tailor made for each discipline. That could work—at least until the algorithms catch on. On the other hand, developers are working to make their systems more and more like humans.
Cynthia Murrell, January 1, 2024
-
- Subscribe to Beyond Search
Feature archive
News archive
-
