Today’s AI agents are highly skilled task executors but they are not independently smart like humans. They are very good at following instructions, using digital tools, and automating repetitive work. However, they still lack deep reasoning and struggle when unexpected problems pop up. [1, 2, 3, 4, 5]
Think of them as super-fast assistants rather than independent thinkers. [1, 2]
What They Do Well
Use Tools: They can look up data, fill out forms, and connect different apps.
Write Code: Coding agents can find bugs and write software parts quickly.
Save Time: They compress hours of data sorting into just a few minutes.
Follow Routines: They excel at marketing, customer support routing, and basic finance tasks. [1, 2, 3, 4, 5, 6]
Where They Fail
Low First-Try Success: Many complex agents only succeed on their first try about 24% of the time.
Rigid Planning: They strictly follow human-made rules and prompt pipelines.
No Real Common Sense: If a step breaks, they often get stuck or make up errors.
Need Babysitting: Humans still have to watch them closely to fix their mistakes. [1, 2, 3, 4, 5]
Current Smart Scale
The tech world is currently moving from simple AI Agents to Agentic AI. [1]
Follows strict, human-made scripts to complete a set goal.
Widely available today.
Agentic AI
Can reflect, self-correct, and plan its own next steps.
Early stages and improving.
Right now, AI agents are incredibly powerful tools for boosting your daily productivity. But they are still software systems that need your guidance to stay on the right track. [1, 2, 3, 4]
AI, as a computer, can put into play a lot more variables than I can quickly. This helps with cleaning up a programming job. But like a child, I have to keep it on course. Like a hammer, I have to swing it; it won’t swing itself.
The more I use Grok and Perplexity, the more I see the limitations of AI. It reads the news and other simple sources that I can read myself to double check the answers. Many times the answers aren’t that reliable. I didn’t realize that reddit comments are considered a reliable source of information. I don’t do reddit but have read on there in the past. AI can gather numbers like financial reports and makes nice graphs however. I agree with leap1 that one has to watch over AI results closely. It’s actually not thinking like we think but just processing a large amount of data quickly. The conclusions can be flawed IMO. It’s still useful. The more precise the question, the more information you get that may be related to what you are actually searching…doc
First there is pattern matching which compares the input from the senses against the vast amount of data already stored. It recognizes faces and voices in micro seconds, the more familiar the inbound data the faster the match against data already stored, what we call “memory.”
The second part, where AI is still weak, is in verifying, in making sense of the inbound data, what I call the boolean brain.
The current version of AI is like a grade school student while we expect Ph.D. level understanding. Give it time, realize that what were are using is very low level A Intelligence.
o o o o o o o o o o o o o o o o o o
If there is nothing new under the Sun how come we keep inventing new things? This quandary used to tickle my brain. The Science of Complexity gives the answer, the solution. The adjacent possible or the emergent property. Let’s see what AI can do with that thought.
o o o o o o o o o o o o o o o o o o
GoogeAI:
Innovations and new ideas exist in an ever-expanding web of possibility. Because simple components interact to create entirely new, unpredictable combinations, the universe builds upon itself. Each new invention acts as a springboard, creating fresh doors that never existed before. [1, 2, 3, 4, 5]
The Science of Complexity beautifully explains this phenomenon through two core concepts:
The Adjacent Possible: Coined by theoretical biologist Stuart Kauffman, this concept dictates that at any given moment, a system can only transform into specific, nearby configurations. When a new technology or concept is created, it fundamentally shifts our starting point, making new doors reachable that were once impossible. This process is continuous, expanding our options infinitely. [1, 2, 3]
Emergent Properties: This describes how complex systems—whether they are ecosystems, economies, or the human brain—produce entirely new behaviors or properties that cannot be predicted by studying the individual parts in isolation. It is the “whole is greater than the sum of its parts” principle. The simple interactions of basic elements spark brilliant, unexpected realities. [1, 2, 3]
For an in-depth dive into how these ideas drive cultural and technological evolution, you can explore Steven Johnson’s popularized concepts on the subject, or review Stuart Kauffman’s work detailing how living systems naturally explore this space
If Emergent Properties exist, intelligence is doomed to constant learning to catch up with all that emerges and impacts our lives.
Highly recommended, At Home in the Universe by Stuart Kauffman.
BTW, Kauffman calls it “Order for free!”
GoogeAI:
Stuart Kauffman uses “Order for free” to describe how complexity itself triggers spontaneous self-organization. Rather than relying solely on Darwinian natural selection, he argues that natural systems—like complex networks of molecules or gene regulatory networks—naturally crystallize into highly stable, patterned structures once a certain threshold is crossed. [1, 2, 3, 4, 5]
I suspect a lot of users don’t (or even care that much, come to that) … reddit, predatory “pay to publish” journals, pre publication abstracts, studies with results that have never been replicated, web pages that repeat the misinformation from primary sources. And more besides. The reason being, of course, as much as folk are given to stating that you do need to check (especially the AI tool that they don’t tend tovuse) there’s a really strong tendency to read the AI opinion/conclusion without going any further before glomming onto it or gifting us all with the Google AI overview as if it’s an authoritative voice.
I find it hardest on the topics where I have a clue or 6.5. Checking every single cited source in order to discard the dead wrong, the misleading, or just plain inconsequential is a chore not unlike sorting through a mound of dog’s vomit to see what they’ve been eating. Easy to see why it’s not a popular exercise.
Well, if you mean “using well honed critical thinking skills”, neither are we. Critical thinking doesn’t come naturally for most folk and is a skill set that has to be learned and frequently reinforced. I think this skill set is likely to rapidly decline …inversely proportional to the ability to craft even better questions to give AI vehicles a better chance to reinforce whatever biases we already hold.
I am doing this for marketing purposes. A lot of subreddits are kids one-upping each other. Where the adults and kids roam together, I add substantially, which makes me an authority on art.
As well as an authority on manipulation of domestic abuse evaluators (from the couple of responses along with mine on my post that so impressed Google AI overview a year or so back!)!!
Back when I first started ruminating on how the Wankspangle had achieved this, I was pretty much an ignoramus on the topic (but, at least, knew enough to be aware). My practical experience of the consequences has grown quite considerably since (matched by expenditure on legal fees to deal with the ongoing fallout) …plus my ability to ask the “right” questions in ChatGPT. Of course, this is a double edged sword in that it’s perfectly possible to appear authoritative on any crackpot idea that occurs…but also, a fundamental feature of critical thinking/Scientific Method thinking is actually being able to ask the right questions in order to get a meaningful answer (I hope high school science teachers are still emphasizing this point).
Seeing as those questions and answers are all presumably available to AI retrieval systems, it’s such a crap shoot as to whether the responses given are even worth the time to type the prompts into the search box.
A rapid sifting through all avaliable information sounds great…but it seems to me that, if it’s a big unknown just how much of this information wrong, it’s a worse deal than the old methods where running up and down library stairs to access Index Medicus (for but one example) for research information.
My experiment with ChatGPT mentioned upstream was useful in that it highlighted the importance of a few changes to the search question (to get a less biased answer rather than to game the response…more important for personal decision making vs internet arguments) did change the response and fair bit.
I also mentioned back in my first post on this topic that, if the judge/court was unaware of this danger back then (Nov 2024) that might well change in the future. I think that prophesy has come to fruition given the plethora of examples from the legal world that still sometimes feature in my searches. I confess that I did think it was a moot point back then ..
..