How Dangerous is Ai for This Assignment?

One of the more misguided claims about the Ai Questioning side of the Ai debate (if it can be called that since one side so clearly trumps almost all other aspects of the conversation) is that we believe in sticking our heads in the sand when it comes to Ai. No one on the Questioning side has ever suggested that, so this is an obvious red herring response. In fact, the Questioning side wants people to understand the true state of Ai much more clearly than many do in order to understand the very real dangers it creates.

As much as I get tired of the “Ai is inevitable” discourse (because there are still many, many places where Ai is not part of anything), of course I recognize that it is here and that people will use it. Some people will even feel they have legitimate uses for it (even though I often disagree with them). Things like Ai usage charts can be a helpful tool in these instances.

One of the Ai usage charts that seems to be popular is from North Carolina, which first gained attention due to this EdWeek article last year. There have been updated versions since then. I had some concerns with the original one: using red for “No Ai usage” and assigning it 0 on the scale, while using green for “Full Ai Use with Human Oversight” and assigning it a 4 on the scale definitely gave the chart a major pro-Ai bias. The updated version removed some of the chart functionality that was helpful, added “Ai Resistant” to the 0 level, rebranded the highest level of 4 to “Empowered,” and added a rocket ship and an infinity symbol to the highest level. In other words, they doubled down on their pro-Ai bias. Plus, so many places you find any version of this chart has the table as an image, making it an accessibility nightmare.

So I decided to go back to the better first version, and re-frame it into an Ai harm chart. The colors were reversed, and a “Potential Ai Dangers” column was added.


How Dangerous is Ai for This Assignment?

Generative Ai Dangers and Disclosure Scale

Generative Ai refers to any of the thousands of Artificial intelligence tools in which the model generates new content (text, images, audio, video, code, etc.). This includes, but is not limited to, Large Language Models (LLMs) such as ChatGPT, Google Gemini, etc.; Image creators such as Dall-E3 and Adobe Firefly; and any tools with built in generative Ai capabilities such as Microsoft Copilot, Google Duet, Canva, etc., etc.)

Level of Ai Use Potential Ai Dangers Full Description Disclosure Requirements
0 No Ai Use No Ai generated bias or hate speech.

No made up Ai responses (aka “hallucinations”).

No water or electricity consumption by Ai data centers.

This assessment is completed entirely without Ai assistance.

Ai Must not be used at any point during the assessment.

This level ensured that student rely solely on their own knowledge, understanding, and skills.

No Ai disclosure required.

May require an academic honesty pledge that Ai was not used.

1 Ai-Assisted Idea Generation and Structuring Moderate to high chance of Ai generated bias or hate speech.

Moderate to high chance of made up Ai responses (aka “hallucinations”).

Water or electricity consumption by Ai data centers depends on amount of personal usage.

No Ai content is allowed in the final submission.

Ai can be used in the assessment for brainstorming, creating structures, and generating ideas for improving work.

Ai disclosure statement must be included disclosing how Ai was used.

Link(s) to Ai chat(s) must be submitted with final submission.

2 Ai-Assisted Editing Low to moderate chance of Ai generated bias or hate speech.

Low to moderate chance of made up Ai responses (aka “hallucinations”).

Water or electricity consumption by Ai data centers depends on amount of personal usage.

No new content can be created using Ai.

Ai can be used to make improvements to the clarity or quality of student created work to improve the final output.

Ai disclosure statement must be included disclosing how Ai was used.

Link(s) to Ai chat(s) must be submitted with final submission.

3 Ai for Specified Task Completion Chance of Ai generated bias or hate speech depends on usage levels.

Chance of made up Ai responses (aka “hallucinations”) depends on usage levels.

Water or electricity consumption by Ai data centers depends on amount of personal usage.

Ai is used to complete certain elements of the task, as specified by the teacher.

This level requires critical engagement with Ai generated content and evaluating its output.

You are responsible for providing human oversight and evaluation of all Ai generated content.

All Ai created content must be cited using proper MLA or APA citation.

Link(s) to Ai chat(s) must be submitted with final submission.

4 Full Ai Use with Human Oversight Excessive opportunities for bias and hate speech.

High possibility of made up Ai responses (aka “hallucinations”).

Largest consumption levels of water and electricity by Ai data centers.

You may use Ai throughout your assessment to support your own work in any way you deem necessary.

Ai should be a ‘co-pilot’ to enhance human creativity.

You are responsible for providing human oversight and evaluation of all Ai generated content.

You must cite the use of Ai using proper MLA or APA citation.

Link(s) to Ai chat(s) must be submitted with final submission.

Adapted by Matt Crosslin from the adaption by Vera Cubero for the North Carolina Department of Public Instruction (NCDPI) from the work of Dr. Leon Furze, Dr. Mike Perkins, Dr. Jasper Roe FHEA, & Dr. Jason Mcvaugh. Link to Original Work.

Creative Commons Licensed BY (attribution) NC (Non Commercial) SA

Creative Commons Licensed BY (attribution) NC (Non Commercial) SA (Share Alike)
To remix this for your use case, you may make an editable copy, using this TEMPLATE LINK.
Please maintain CC licensing and all attributions in all duplications, references, or remixing.


Some might argue with the column order in this chart, but I feel the potential harms need to be front and center in any usage, partially because they are so glazed over (at best) in most usage discussion, but mainly because we really should be more upfront about the dangers that can cause real harm. I kept the necessary language and attributions in place, except for updating the link to the original work to the correct link. The template link leads to the most recent version of the original chart, which wouldn’t be helpful for updating this version. But it is there.

Why Do Ai Cheerleaders Respond to Critics the Way They Do? (Part 2)

In part one of this blog post I started looking at some ways that “Ai Cheerleaders” (as I refer to them) respond to “Ai Questioners” (some would say “anti-Ai” or even terms meant to be derogatory like “activists”). I need to apologize to those that fall into the middle of the two camps – I know that there are many people that don’t accept all of the Ai hype but also think there can be good uses of Ai. So there are more than the two camps that I pointed out – my mistake. There really is a spectrum with total embrace of Ai on one side and totally rejection of Ai on the other, with most of us falling somewhere between the two extremes. But there does seem to be a large cluster of people on the “Ai positive” side of the spectrum where one would place the “Cheerleader” label, and another large cluster or people on the “Ai negative” side where one would place the “Questioner” label. I got in a hurry and just decided to look at the spectrum as if there was a hard line between the positive and negative sides, but I know there are many in the middle that feel they are neither positive nor negative.

However, the two sides I focused on do exist, and one of them (Cheerleaders) drives a lot of the narrative. The “Ai Cheerleader” side seems to be having the most influence on policy these days – up to the national level in the United States. It really should give some people pause when they realize that they are in agreement with Trump, Musk, and other people… and not in a “broken clock is right twice a day” kind of way. Agreeing with people that are wrong about everything means you are, well…

The frustrating thing is that the data on Ai harms and problems, as well as concerns of the real people being forced to use Ai every day, usually lines up with the Ai Questioning side. Not always, and not perfectly – but more often than not. So you have to wonder if the Ai Cheerleader side pays good attention to the Ai Questioning side, or even the Ai Neutral or Undecided side. This part 2 will dive into these concerns more broadly.

Watching “both sides” of an issue (or “all sides”) can definitely give one a full-ranging sense of the differing viewpoints surrounding controversial issues. Generally what one comes away with when doing this is that these issues contain a lot of complexity. However, that doesn’t mean that both sides recognize this complexity. For the issue of Ai, this complexity is often the source of the concerns for the Ai Questioning side. I also think this is where a good deal of the criticism of the Ai Questioners by Ai Cheerleaders misses the point. Or even mischaracterizes the Ai Questioners altogether.

So basically I want to go through several posts by various Ai Cheerleaders and discuss the concerning ways that they frame the Ai Questioning / Skeptical / Activist side:


There are many Ai Cheerleaders that echo some sentiments like Donald Clark stated recently:

“Once again I’d emphasise that you really do have to ‘try it to get it’. An interesting paper just published showed that much of the scepticism on GenAi comes from those who have not tried it. Lesson – for any real critical thinking or analysis, some use is necessary.”

What paper? There is no link. What is getting labeled as skepticism? Are the questioners of Ai also the skeptics? Every time I speak to someone that it is skeptical of Ai, it is after they have used it. I was just on a phone call with a friend last night that was making fun of how bad an Ai program she used that day was. Most of the skepticism I read in articles comes from people that are using it.

For example, if you read Audrey Watters, it is obvious she knows Ai and even mentions when she tries it out. I mean, she has spent decades studying Ai. And she is a skeptic:

“But Ai – a large-language model or predictive algorithm or otherwise – is built on a corpus that is, quite literally, bound to the past. Education’s Ai has been trained on outmoded curriculum, exclusionary practices, and racist data; it is trained on YouTube videos and YouTube comments and Wikipedia entries; it is trained (mostly) on the English-language Internet – trained on a very small slice of knowledge and culture because not all knowledge and culture have been recorded, let alone digitized; and yet simultaneously trained on a disproportionately large slice of discrimination and violence, because that has been the experiences of Black students, poor students, students with disabilities, non-English-speaking students, undocumented students, and queer, nonbinary, and trans students.”

Did the paper that Clark mentions take into account skeptics like Watters? Or did it give the same weight to the opinion of random people who have never touched Ai (and would naturally be skeptical) as it did to those that have studied it for decades? Who knows. But there is a big difference in noting the general skepticism that always proceeds new technology (or new anything) and noting how much real skeptics know about what they question.

Also, the idea that “any real” critical thinking requires use is contrary to what we know about critical thinking.

To be clear about my concerns here:

  • Referring to a paper without linking to it or even citing it is concerning at best.
  • The implication that most skeptics haven’t tried Ai is inaccurate.
  • Implying that most criticism doesn’t come from real critical thinking is concerning.

Digging into the criticism of Ai Questioners is always hard because of there are also all of these constant inappropriate digs at questioners’ intelligence. For instance, Stephen Downes said:

“But it isn’t Ai that’s the problem, is it?…. The product isn’t the problem, the companies that control the product are. Educated people should be able to see the difference.”

This was after referring to Watters and Marc Watkins, two people who know this difference well and who mention the companies that control Ai all the time. In fact, I am having a hard time thinking of any critic that doesn’t note the difference. Most people questioning Ai are criticizing the companies that control it AND the product itself. It’s not a simple either/or. How do you separate the people that control Ai from what Ai does? Not very cleanly. But Ai has problems all on its own: very little ability to detect bad text / false results / biased or hateful responses, resource decimation, and so on.

To be clear about my concerns here:

  • Claiming there is a clean separation between the Ai products and the companies that control Ai products is not realistic.
  • Taking swipes at “educated people,” implying that people who disagree might not be well-educated on this topic, just isn’t helpful for the conversation.

Oh but wait, we are told that Ai is improving rapidly and consuming less recourses, so these problems are gone or disappearing, right? Except, when you dig into the actual proof… not so much. This claim of “consuming less resources” was made based on five charts. The first chart claims to show how “the cost of intelligence is plummeting,” but it is a chart with straight lines and some suspicious assumptions on the x-axis. I have never seen a resource consumption chart (for something like energy) that is a straight line – they always have bumps and dips. Flattening out data like this means either it is all predictive and not based on real world numbers, or you are trying to hide a trend that the bumpy line would show. However, taking the chart at face value, while it does show that “achieving GPT‑4–level performance now costs nearly 1,000 times less,” the point is never made that this “should be a relief to those with genuine concerns about the environment.” It is about saving CEOs money: “This significant drop in costs will enable companies to integrate Ai into the free tiers of their products, potentially unlocking Ai for over a billion new users.” Yeah, simple math would point out that dropping costs by 1,000 times just to add 1,000,000,000 new users means you are still using 1,000,000 times more resources. Yes, I know that would make the same math mistake that the chart itself does, but even using correct math would show these decreases in cost are just there to empower greater resource deprivation.

Oh, and where is the chart that is saying “Ai capability is improving rapidly”? Is it the chart that basically says “it is okay for Ai adoption to take time”? Is it the chart that shows consulting firms are helping with Ai adoption? Or is it the one that shows there are more parts to Ai that need to work than just the language model? Perhaps it is the really problematic one that claims “any task whose outcome can be evaluated can be learned through reinforcement learning”?

(No, this claim about reinforced learning is not true – there are many forms of evaluation that are complex and take much more than reinforcement learning. This chart is based on math courses, and many, many topics are much difference than math. So you can’t say “ANY task.” And of course we know that if learners repeat something over and over again it will stick in some way. But these learners could also end up hating learning and whatever topic you are bludgeoning them with. Seriously people – talk to an instructional designer before setting up these studies!)

To be clear about my concerns here:

  • There isn’t a chart that claims Ai is improving rapidly.
  • The chart that shows costs of Ai plummeting is suspect in different ways.
  • Showing a decrease in consumption is not a relief to those with environmental concerns, as the chart in question is used to justify an even larger increase in usage – not saving the environment. Plus, hello: look at how capitalism works.

Anyways, back to those unnecessary digs at intelligence. This one takes some set-up: Marc Watkins made the point that

“It will take years of trial and error to integrate Ai effectively in our disciplines. That’s assuming the technology will pause for a time. It won’t. Which leaves us in a constant state of trying to adapt. So, why are we investing millions in greater access to tools no one has the bandwidth or resources to learn or integrate?”

Watkins’ point was also proven by one of the five charts in the last example, so that sounds like an accurate assessment… right? Well, Stephen Downes strongly implies that this is “less” “smart thinking” and then responds:

“The assumption is that we (the instructors, the institution) must fully master the tool before it is useful to learners. But of course, that’s not true at all.”

However, that assumption is not in Watkins’ writing, because there is a difference between effectiveness and mastery. For example, I can learn how to effectively cook without mastering cooking. Most grading scales that use the term “mastery” have several levels (Does Not Meet Standards, Meet Standards, Masters Standards, etc). I can think of probably thousands of ways that effective is different than mastery. Watkins is making the point that the time it takes to learn how to use Ai effectively is too long. Sure, instructors are using Ai in courses, but many students and teachers are trying to point out it is not always very effective. And as an instructional designer, I can think of thousands of examples of instructors using technology when they really shouldn’t have. Just because they are using something, that doesn’t make said usage effective.

To be clear about my concerns here:

  • Exchanging words like “mastery” for “effective” distorts the original questioner’s point.
  • There is a very well established and known history of instructors using technology in courses, and this is obviously what Watkins is referring to.
  • Taking swipes at how “smart” thinking is or is not just because someone disagrees with you just isn’t helpful for the conversation.

Then there is the implication that skeptics are not asking the right question. When Audrey Watters elsewhere said:

“We cannot outsource thinking and compassion to a hierarchy-generating machine and expect the world to be anything other than automated emptiness and exploitation”

Stephen Downes replied:

“But that’s a different question from the one asking whether we need human teachers to teach. Now I’m not endorsing the idea “that automation will make learning better, faster, cheaper, more scalable, more ‘personalized.'” What I think technology does is to enable students to manage their own learning for themselves, so that they can be thinking and compassionate learners, and not mere automatons following a teacher’s instructions.”

Now, I can’t find where Watters was asking a question, or claimed it was the only question to ask, or that it was the same as asking whether we need human teachers. Not to mention that different questions can still have overlapping answers. So I’m confused by Downes’ response to her point. But when Downes says what “technology does is to enable students to manage their own learning for themselves, so that they can be thinking and compassionate learners, and not mere automatons following a teacher’s instructions”… he actually is claiming that technology can make learning “better” and “more ‘personalized'” (aka the opposite of automatons). He may not claim that technology is automation, but that is the goal of so many Ai programs. We also know that technology can sometimes help learners that so choose to use it that way to become more “thinking and compassionate learners” (aka “better” learners), but its the learner that does this not the technology (and sometimes in spite of what the technology forces on them). However… many learners do not get that from technology. Research shows this again and again – in fact, my own research into moving students into self-determined learning often showed that many learners actually wanted to follow the teacher’s instructions rather than choose for themselves (even if in other instances they do want to do the thinking). It’s complicated, but I rarely see technology itself making the change in the learner. Believe me, I tried to create some that did.

To be clear about my concerns here:

  • We shouldn’t be claiming someone is asking a question when they aren’t asking a question, and besides that there is room for many questions in this discussion.
  • Claiming one doesn’t think technology will make learning better or more personalized, and then turning around and saying that you believe technology can help make learning better and more personalized.

There are also concerning ways of looking at education expressed by Ai Cheerleaders. For example, Stephen Downes had this to say about knowledge:

“The knowledge is pretty much irrelevant. Not because we don’t need it, but there will always be knowledge, the knowledge always changes, and if we need to, we can always look it up.”

This is in response to a blog post by Vickie Davis that says we need a “fire that sparks the joy of learning.” My experience as a public school teacher was that most children like my son love learning – but I have to spend a lot of time pointing out when he learned something that is misinformation, when he learned an opinion as a fact, or even how to apply what he learned in real life. These are things you can rarely just look up without running into vastly opposing opinions. The idea that you can “always look it up” is problematic in today’s misinformation reality. One could respond to Downes that learning is pretty much irrelevant, because there will always be learning, it can always change, and if they need to, people can always do it on their own anyways.

We need knowledge to tell when something that we learned is misinformation. We need knowledge to know what to do with what we learned when it’s all correct but requires different uses in different contexts. Often knowledge requires knowledge of context. But even saying that would be problematic, because that all circles back to learning. We need to start realizing that knowledge, learning, application, discernment, etc are all complicated interrelated aspects that can’t be separated out and therefore shouldn’t be treated with such black and white absolutist thinking.

To be clear about my concerns here:

  • Knowledge is more complicated than saying there “will always be knowledge, the knowledge always changes, and if we need to, we can always look it up.”
  • Context of knowledge is important to knowing what to learn about said knowledge.

Anyways, while you can argue over technical details about whether Ai is destroying education or not – does any of that matter when the people in charge of Ai are working with the people that are dismantling public schools and public libraries to replace both with Ai? And they are starting to export all of this outside the United States? When many children will only get access to the Ai replacement for public schools and libraries, does it really matter if education still exists for the elite? When you find yourself promoting Ai on the same side with Trump and Musk, shouldn’t it cause you major concerns?

Because yes, while I know many believe thing like “the promise of Ai is that everybody will be able to understand other languages” and that people working with Ai will all be “well-educated high-paid professionals,” the people currently in charge of Ai companies have proven time and time again that they will charge for their products eventually, and that they will not pay people very well to work with it. There is no Ai utopia lining up in for the masses in the current MAGA era.

Of course, there are those that say that it feels “like parts of higher education face a reckoning that can’t be ignored. We need to communicate value (social and economic) to our learners and to society.” But is that really the problem? Institutions spend untold billions on advertising, recruitment, outreach, etc, etc, etc. This kind of shows that no matter how much power an institution has, if another aspect of society holds more power, and half or more of that group hates said institution and is actively working to dismantle it… no amount of value communication will overcome that power imbalance. Of course, higher ed is doing many things wrong, but I’m not sure communication of value is one of them.

I could continue citing problems with posts and articles about the Ai Questioners and how so many get us wrong. Critics of the Ai Questioning side need to do a better job of really reading what the Questioners are saying, and then need to respond without stooping to unnecessary attacks against intelligence. It would help to stop making claims about Ai that are not backed up by the papers and studies one is quoting. In part 3 (if I can find the time), I am going to look at a specific comment left on one of my posts. I think the commenter thought they would show me how good Ai is, but critique of my post that Ai came up with is so bad that it would be hilarious if it wasn’t meant in all seriousness.

Why Do Ai Cheerleaders Respond to Critics the Way They Do? (Part 1)

Responses to Ai tend to fall into two camps: those that cheerlead any new thing in the Ai world, and those that question whether we need to have Ai take over everything (or really anything). I am noticing a concerning trend in the Ai Cheerleader camp in regards to their responses to the Ai Questioning camp. As I documented those concerns in a post, it grew too long for one post. So I broke it into at least two parts. This first part will set up what concerns me in responses to two of my posts, and then part 2 will dive into where these problematic responses appear in many different sources.

So I thought I should take a moment to respond to the criticisms of some of my posts first. Sigh…. Look, it is hard to take the time to unpack things here and then read people who obviously didn’t spend time reading what I wrote and then get it all wrong. When I refer to someone not “reading” something (my posts or Audrey Watters newsletters or anything else), I am not saying that these writings were never actually read. I am using the colloquial meaning of the word to indicate that even though someone did physically places their eyes on every word, they were looking at the words to respond and not understand what was being said. The whole “listening to argue and not understand” thing in written form, in other words.

For example, Stephen Downes responded to my post about course costs should be zero. My post is mostly about concerns I have with David Wiley’s model, and George Siemens’ claim that this should make course design costs zero. After an odd and mischaracterized summary of my points, Downes states this:

“Now it doesn’t matter that Crosslin doesn’t follow my work (he says he doesn’t) but if he did he’d know about my history in instructional design, the developments Siemens and I introduced to the field, and our emphasis on personal learning support as a replacement for pre-planned formal instructional design and pedagogy, something that can demonstrably be supported using Ai. And which should reduce costs to effectively zero.”

Well, I used to work for George Siemens for years, and have often cited Siemens and Downes in multiple published papers, so yes I am familiar. Following someone’s work that consists of multiple daily posts is hard to do, but of course I read those that are sent to me or quoted by others. But anyways, contributing to a field doesn’t mean you are the field, and there are many, many people that also contributed to the field of instructional design that don’t follow Siemens or Downes work. This response is kind of close to a logical fallacy – people that contribute to a field can still misunderstand things and even get things wrong. People inside a field can (and often do) get things about their own field wrong. So that just feels like throwing weight around without actually disproving anything. In fact, it doesn’t actually address my points at all. Whether I was right or wrong or both is independent of whether or not I know anyone’s history or what said history is.

This response would be like if someone criticized Jar Jar Binks as a bad character and George Lucas responded with “don’t you know my contribution to the field of Science Fiction?” Well, yes, we do know those contributions, but that has nothing to with how bad Binks is as a character. Plus, that doesn’t automatically make you the expert on all Science Fiction, nor does it mean the entire Science Fiction field is centered on Star Wars (as Trekkies will enthusiastically attest).

I haven’t seen anything that suggests a definite possibility that Ai can support personal learning, and I even tried to do that myself at one time. It is possible, but there is a long way to go to say that it definitely can be demonstrated. Maybe if Downes would actually follow my work he would know how I do know Siemens and his work and how I have worked in this specific area. But anyways, supporting personal learning with Ai would in no way reduce course costs to effectively zero, as Ai still needs a lot more human support than many like to acknowledge, and there are still many ways to design courses that Ai can not support.

So, yes, if you claim that course design costs should be zero, you do need to learn a lot more about the field of instructional design. If you mean to say that design of a certain type of course should be zero (effectively or otherwise), then say that and give a real example of how Ai is actually doing that (without any prompt engineering or corrections, because that would be someone’s time which is effectively a cost).

Also – Downes says that I recommend that he should read Audrey Watters. Please read what I wrote Stephen. It was not you that this statement was directed at. And I know the person that it was directed at does read Watters, so it was obviously a tongue-in-cheek joke about how he relies too much on Ai summaries for his daily reading.

BTW – when I make statements like “clean separations of course content and learning activities are generally labeled as bad pedagogy in instructional design circles,” this is because I currently work on a team of instructional designers, teach instructional design courses, and daily interact with IDs across the globe on social media. We had many discussions about this idea in the days leading up to me publishing my post. The overwhelming consensus among my circle of IDs was that this is bad pedagogy.

So let me restate my disagreements clearly:

  • David Wiley made a case that learning content and learning activities should be cleanly separated in the design phase, comparing it to HTML and CSS. This is not really how HTML and CSS work in the design phase, and it has generally been considered bad pedagogy to separate content and activities when designing a course. This was a criticism of SCORM, xAPI, and other attempts to do the same in the past. Me saying this is nothing new. The general idea of OELMs could still have merit regardless of this misunderstanding, but would probably be better off if not based on this idea.
  • George Siemens stated that OELMs are an example of why course design costs should be zero, and Stephen Downes said he agrees. However, Wiley’s model does not allow for zero costs in any way (effectively or otherwise). Wiley even stated that OELMs don’t replace textbooks or courseware – it just supplements them. Neither Siemens nor Downes has given a practical example of how course design would effectively be zero (Wiley was clear that his model would have costs), unless there is the belief that you can enter a prompt correctly the first time into Ai and get a complete course design out of it with no prompt engineering or clean-up needed.

I’m also concerned about how Downes contrasts “pre-planned formal instructional design and pedagogy” with “personal learning support” as if the two haven’t easily co-existed for decades or that one needs to replace the other. The fact is that most research into personal learning support has indicated that you need a lot more instructional design support to accomplish good personal learning – and Ai has yet to demonstrate the ability to provide the level of complexity and creativity needed to do so.

Later, Downes responds to my post about AGi becoming a thing someday:

“He first crtiticizes (sic) one of the core points: “a system made up of many individuals has more intellectual capacity than individuals? Well… duh. Hasn’t it been that way since the beginning of human intellect? Adding more people gives you more capacity.” I don’t think this captures the nuance between simply stacking capacity (ie, adding volume) as opposed to combining capacity (ie, adding complexity).”

Human systems since the beginning of time have obviously added both volume and complexity, so while that nuance is important, it doesn’t change my point. Adding more people gives you more volume, and adding more people gives you more complexity. It is both, not just one or the other. Anyone that has to work on a team knows this. Like I said later, not all systems do this well (in the example of MAGA), and that was me referring to this nuance.

Then there is this statement:

“Crosslin responds, “None of this is artificial.” But I don’t think we’ve seen a naturally occurring organization or institution through all of history.”

Well… what happened to nuance? Sure, everything humans do is artificial, but people that make natural chocolate or beer or whatever don’t call their products “artificial” (even though technically they are). Siemens was making a point about “modern companies, governments, and social structures,” so obviously I was also making the point there have been no naturally occurring organizations through history.

“But Crosslin’s main point is in the title: “I am pretty skeptical that AGi will ever happen – but people keep trying to see it where it doesn’t exist.” I’d defer to his deep knowledge here, but if he makes basic errors, I think it might just be an opinion based on an intuition, and nothing more.”

Well, I did say “I am pretty skeptical…” so obviously it is my opinion, but not one based on intuition. Scientists have analyzed the data from 8590 predictions of AGi through the years. While the number of experts that think AGi will never happen is constantly decreasing, I think the data shows that the consensus is not “very soon.” For me, if you dig into what is being said by the various predictions, those that say it will never happen have more solid reasoning and concrete proof. Those that think it is happening very soon seen to have a more vested financial interest in that being true (see Sam Altman, for example).

But it’s not like I am the only person outside the Ai industry that thinks AGi may never happen. Take Ai critic Ed Zitron for example:

There is no reason to humour anyone talking about AGI or give someone credit for doing so, it is not an “opinion” to “debate” as there is no actual proof it’s possible. “Extrapolation” is another word for making shit up

— Ed Zitron (@edzitron.com) April 4, 2025 at 4:54 PM

I didn’t add this to say that Zitron is the most knowledgeable person out there on the topic. It just shows others are coming to this conclusion as well. Many of the replies to Zitron’s post also indicate I am not alone:

  • “Chatbots could get exponentionally (sic) better each year for ten thousand years and still never be conscious. AGi is a religious concept, not a techincal (sic) or scientific one.” – ‪Paul Bloem
  • “it’s just bro-coded manifesting” – @james.claims
  • “AGi is currently as real as negative mass, in that you can build a mathematical framework for it, but there are ZERO real world processes for achieving it.” – @‪Meowdelyn

But again, to be clear about my disagreements with the article in question:

  • Stating that “many labs” think AGi is happening “very soon” flattens out a lot of nuance, as well as the fact that many labs also think “never.” This seems to be a problematic way to keep readers informed of the reality of Ai. Also, why only focus on labs and not individual scientists, researchers, and writers?
  • Related to the last point, but a problem that I didn’t dive into in the post: Why were the words “very soon” in quotes? Was that a quote from the posts quoted later? Who said it? Why is this left so vague?
  • Comparing “modern companies, governments, and social structures” to a “type of AGi” is very problematic, and seems to ignore the history of how human structures made of many parts have always held “more intellectual capability than individuals.” It is not a new thing that Ai is suddenly bringing to the table.
  • Saying that all of this aligns with “thinking on connectivism… over the years” appears to directly contradict the writings that were made about connectivism “over the years.” If that thinking has changed, then great – but acknowledge that you have changed and don’t make it seem like this is the same idea you have always had. It feels like trying too hard to make “fetch” a thing.

Anyways, when Downes says “I’d defer to his deep knowledge here, but if he makes basic errors, I think it might just be an opinion based on an intuition, and nothing more” – you can see an unnecessary attack on intelligence in the deep sarcasm within that statement. That is why I generally don’t follow any Ai Cheerleader, unless a fellow ID says there is something I need to read. As we will see in part 2, the constant digs at the intelligence of the Ai Questioning side are pretty rampant in the responses of the Ai Cheerleader side.