Court Rulings and What Kind of Critiques Count in the Ai Debate

As expected, courts have been ruling in favor of Ai companies in various copyright lawsuits over the past month or so. The Ai sector has some big political names backing it up, so I would expect more rulings in this direction.

However, that doesn’t mean the rulings were correct. For example, one of the lawsuits looked at how transformational Claude is when it gives answers. Of course Claude is not transformational – it quite often just gives straight out plagiarized answers. Ai is just not generally transformative, unless you only look at the top 1-2% of it’s output. Your average Weird Al Yankovich song is more transformational than most Ai. I guess my past studies to be an art teacher gave me a greater sense of what transformational means verses copying and plagiarism. Rearranging a few words in order to fool plagiarism detection software is not all that is needed to be transformational.

Also as expected, the educational bloggers that are in favor of Ai also agreed with the ruling. We all knew they would before hand. But the justifications they give for agreeing with the rulings are kind of head-scratching. Take, for instance, this statement by Stephen Downes:

“In this I am in agreement with the court. You can’t make learning from what you read illegal, even if it’s a computer that is doing the learning.”

Technically it wasn’t the “learning” that the authors had a problem with, but the thousands (if not millions) of times that “learning” would be used to make someone money. Copyright law has always said that if you want to do that, you just have to pay for a specific right to do that. For example, video game makers can’t create a virtual world and place Time magazine in their world just because they might have bought a copy of Time in real life. If they wanted to work out an agreement with Time to do so they could. You generally can’t use intellectual property in computer programming without permission of the copyright holder – which usually includes different (typically larger) payment arrangements.

Not to mention that “Fair Use” generally does not cover using an entire copyrighted work – just portions of it. Ai typically takes an entire work in as input, so it is clearly not Fair Use.

All of that aside, Downes has written in the past about how scientists no longer see the “human brain is a computer” metaphor as valid. To say “a computer is doing the learning” only makes sense if you see the human brain as a computer. Computers don’t “learn.” You feed a book into Ai, it doesn’t “learn” that book. It stores the text and pattern matches the text based on an algorithm – which is often a more complex set of “if/then” statements. Those pattern matches are stored in a database more akin to a digital warehouse than to a brain. None of this is learning. Outputs from Ai prompts aren’t even complete thoughts like human would have. Ai output is just predicting what the next most likely word or pixel is based on stored pattern matches. This is also how we know the human brain does not work like a computer – learning and recall are all so much different than any of this. As far as we currently know (which isn’t as far as many realize).

(That also begs the question: if we don’t fully understand how humans learn, how can anybody confidently say that computers are doing the same thing?)

Current copyright law is based on the idea that the cost of a single book is fair compensation for the person that buys it taking that knowledge into their work and using it make money, or talking with their group of friends and sharing what they liked about it, or other limited scenarios. The occasional celebrity that reads and shares with millions is an outlier, less than 1% of the sales and probably easily made up for by the fact that this usage would increase sales anyways. Libraries and schools pay more for multiple usage books because more people will use them. Because of all of this, you can’t apply current copyright laws, concepts of transformation, or Fair Use to Ai even if you could make a case that it was “learning” – all because of the sheer scale of usage on the output side. Many things that were “learned” by Ai are used hundreds or even thousands of times for millions of people around the world.

You or I reading a book and then using that knowledge in life or work is in no way comparable to Ai using stored pattern recognition with millions of people.

But so many times when somebody in education is championing Ai, I wonder if they are really aware of the stakes of Ai?

That same week, Downes had this to say about a list of locally-run open source alternatives to ChatGPT:

it’s not just the big corporations, and it’s not going away. It’s just programming, and you can run it on your own computer. It’s not some conspiracy by techbros to take over society; it’s just math.

First of all, calling these tools “alternatives” to ChatGPT is doing a LOT of work. The ones on the list that I have tried are mimicking a small portion of ChatGPT, but there is no way you are going to be a true “alternative” to ChatGPT without massive computing power and data processing facilities. It’s literally what made ChatGPT into ChatGPT: massive scale of computational power.

Just to break down this statement, of course people know there are always smaller alternatives to the big companies. But that doesn’t change the fact that the big companies are dominating the field. If the big companies weren’t so dominant, wouldn’t it just be “locally run open source Ai options” rather that “alternatives to ChatGPT?” Yes, the big companies are dominating, and until they lose that dominance, it is in effect the big companies. If you don’t think Techbros aren’t trying to dominate society, I suggest you go read about people like Elon Musk, Jeff Bezos, and others like them. So many Techbros out there trying to emulate Musk alone discredits that statement.

And also… wait. If it (Ai) is “just programming”… then it can’t “learn.” Programming is not learning.

Also that same week, Downes said “Sure. In a world of rising costs, privatization of social spaces, militarization of police, and uncertainty about the war, sure, let’s blame Ai.” I mean, there can be more that one problem – its not either those OR Ai. Most are merely including Ai in the list. But rising costs of anything from real estate to insurance have been tied to private equity firms using Ai to rapidly find weak points of the market to exploit. Militarization of the police is being scaled up to all of society using Ai. The amount of death and destruction caused by war is increasing exponentially due to Ai guided drones. Misinformation online is exploding thanks to Ai. All of this existed before Ai, but the sheer acceleration of the volume in the past year or so due to designing Ai to do just that can’t be denied. Well, I guess it was denied in one over simplifying statement there, so anything is possible.

To circle back to my concern of whether or not Ai Cheerleaders are actually reading and listening to the Ai questioners, this post by George Siemens runs the line between claiming there isn’t effective pushback against Ai to possible gatekeeping those that do push back. Siemens starts off by talking about Timnit Gebru without naming her, and then claims she is “largely unheard now as if the researcher has completely faded from public dialogue other than in a small camp.” A quick search shows she is still speaking and being heard of in very large camps… but even if it was “small camps”… so what? Is criticism not valid if it is not uber popular?

George then says this:

“I’d love a rich, vigorous, counter conversation to AI hype. What we are starting to see, however, is more of an emotional reaction than a sustained and effective critique.”

A quick look shows that not only Gebru, but also Audrey Watters, Safiya Noble, Emily Bender, Chris Gilliard, and so many others have been offering sustained and effective critique. The fact that so many critics of Ai are women makes that “emotional” swipe concerning. Probably not meant that way, but I always caution anyone to NOT use “emotional” as a criticism. What is it about academics that makes so many of them constantly turn “emotional” into a refutation? Something being “emotional” does not make something automatically wrong. It is just a dismissive response that really says nothing.

Multiple times every week I talk to friends and loved ones that are seeing their jobs ruined by Ai. Their company invests big in an Ai product that ends up wasting so much of their time. When workers don’t use it, and the company has to justify their massive spending on creating an in-house Ai, they make it part of the yearly evaluation. Yes, people I know are being evaluated by how much they use an Ai solution that actually decreases productivity. And when that doesn’t work, some companies are actually making it an HR violation to not use Ai. Yes – this is actually happening.

So, yes, some of us might get “emotional” about the fact that so many people’s jobs are getting ruined by Ai being forced on them. The horror!

Not to mention that, yes, Ai is killing people. It is causing people to break from reality and get involuntarily institutionalized. It is telling people to kill themselves. It is still giving racist, sexist, homophobic, transphobic, ableist, ageist, etc responses. I guess when you are none of those things, you can just criticize anyone hurt by Ai as “emotional”?

But I guess there is some listening happening. Where I previously expressed concern over Siemens stating that course design should cost zero, I guess he has pivoted to a different statement:

“It’s the golden age for designers in education. Content is near zero in terms of cost now.”

So now it’s not course design that costs zero, but content creation. That is a start at least. But seeing that lots of people are still paying lots of money for content, I am not sure where this “golden age” is happening? The examples he gives of how great Ai-generated content is are… not so great: awkward moving jaws and other body parts, a meh (at best) level mockumentary, a site full of okay-ish Ai created content, and a caveman video with a few subtle racist caricatures. I doubt anyone sharing that video (including Siemens) really looked at that closely to even notice.

This seems to be an idea Siemens doubles down on quite often (“I’ve mentioned before that the cost of developing content has basically zero economic value.”). Why are so many companies putting economic value on it then? Was this meant as more of a prediction of the future than a statement of the current status quo? Was it an Ai-driven hallucinogenic summary? Not sure. But also… wasn’t this all the argument for why MOOCs would make teaching obsolete, that OERs would make content have no value, and so on? We are a couple of decades into content being free online. Trillions of webpages and videos have covered every topic imaginable for quite a while now. Before that, textbook companies where bundling entire classes with textbook adoptions. Why did none of that destroy the economic value of developing content?

Because people are individuals that want their own spin on content. Instructors rarely take textbook resources or OERs or YouTube videos and make entre classes out of that. A few do, but most don’t. People like to have their own spin on content. Not to mention that we know from decades of research that social presence and teacher presence help online learners immensely. You need the actual instructor in that content for students to feel connection, not some random Ai avatar.

Since we are at least two decades into free content abundance (and the field of content creation is still growing – at least for now), I doubt content creation is going any where fast. People are paying a LOT for something that Siemens claims “has little economic value.” It is pretty easy to say that “video is almost as easy to create as writing a paragraph” when you are just doing it for fun or to keep up with trends. Those of us that have actually tried using Ai for actually courses in real life? It falls short. Horribly. Sure it may be “cheap and impressively high quality media” to some… but the result you get is never what you imagined in your head. So you either have to settle for something different and inferior to what you wanted, or you do it yourself. Most people historically have decided to do it themselves, and I doubt Ai will change that once the shiny factor wears off.

Also: “cheap” by what metric? Small towns that are seeing their water supply destroyed by Ai-driven data centers would beg to differ on it being cheap. Those of us that have to make up the slack in our electricity bills so that local governments can subsidize data centers would beg to differ. Oh, and it turns out that Ai uses way, way, waaaaay more energy to “create ideas” than a human does. Getting your energy costs subsidized by grants and seed money is giving and impression of “cheap” that will go away someday soon.

Fighting Predicatability in Online Learning

Over the weekend I had a conversation with a friend about online learning.  It turns out he has taken several online courses over the past few years.  He had an interesting statement that I think many in online education need to pay attention to:

“I have gotten to where I don’t need to read the syllabus anymore in online learning.  All the courses are the same”

I asked him if he was referring to the cycle of “read this, answer a discussion question, respond to other students, take a quiz, rinse, repeat.” He said that was exactly the case.

I see this a lot in online education, but to be fair it started in face-to-face courses.  So many classes you only had to find out what the dates of tests were and the rest was easy to figure out.  Some call it laziness, but it probably actually had a huge helping of “if it ain’t broke, don’t fix it” throw in the mix.

Are discussion boards becoming the new scan-tron tests?  I realize that there is some active learning in having students answer discussion questions, but so many times the questions become so stale that what little “active” was in there gets washed away in staleness.

If you have an introductory online learning class (in other words – it is the first online course that students will take) – then I say give them something tried and true to help them get used to it.  To a degree.  For the rest of us – we need to infuse our online courses with personality and a little originality.  Try to think outside the box.  Try some new tools. or at least try to get your students to create assignments using tools they don’t normally use.

But above all, give them a reason to read the syllabus.  Oh – and then at least try to not put them to sleep while they are reading it.  Remember the K.I.S.S. method.

Style vs. Substance in Instructional Design

I’m pretty sure that if you care about actual learning, you have run in to the same problem I have: going to check out the latest award-winning course, program, idea, etc and then coming to the inevitable conclusion that it is pretty much junk.  Some educational awards and accolades do go to great projects… but it seems so many times the attention goes to the slickest, shiniest object in the room and  not necessarily the best.

Clark Quinn has a great blog post about his experiences in this area, which also examines the difficulty in teaching others the difference between what has true quality and what is really just whitewashing a dead, boring lesson to make it look (as we say in Texas) purty.

Sometimes it feels like people think that a good online course only involves the following steps:

  1. Take your lecture notes and edit them well to make them flow beautifully when read
  2. Transfer these edited notes to html and break it all up into decent sized chunks
  3. Slap on a discussion question and a multiple choice quiz at the end
  4. Lather, rinse, and repeat.

Never mind that all these skills (editing, copy/paste, chunking, pushing buttons on a website to create things) are all things that your average high school-er can do – this is now considered high level ed tech pedagogy, right?

Um… yeah….

I wish I was only describing some neophyte professor just creating their first online course, but sadly this list pretty much describes what I have seen labeled as “high quality” instructional design by many people with graduate degrees in this stuff.  I have even heard it labeled “active learning.” (!) (?)

I guess since the student does have to read and respond to discussion posts… that counts as active?  I guess as long as they don’t fall asleep….

Now, I realize that the format listed above can work in some situations, especially if a lot of thought is put in to it. But what usually happens is that it is treated like an online course design template used for every course with components rapidly plugged in. Which in some cases might not even be bad – just not the best option that exists out there.

Will The Internet Start Looking More Like the World, or the World Like the Internet?

I was pondering future trends last week while watching the evening weather forecast.  Forecasting while watching a forecast?  Anyways… We were in for a possible round of severe weather that week. The news anchor put up a map of “storm spotters” – a network of people that would call in from their homes and tell what is happening in their area.

In other words, forecasting the weather is starting to incorporate crowd sourcing.

We have seen a giant push to get websites to work intuitively… and to even start thinking for us.  So on one hand – the Internet is starting to look more like the real world.  But I think even more often we are starting to see the world around us looking more and more like the Internet.  The powers that be are starting to see that there is power in crowd sourcing and social networking.  I wonder what real-life social networks we will see spring up next?

The real question for us is – can we use these ideas in education?  What if we took this weather stations ideas and applied them to a class? What if, instead of one large class, we broke that class down into smaller units based on geographic location.  Each smaller group forms a study group of sorts that watches issues related to the class subject in their area.  The small groups are loosely tied to one another in a way to share what they are learning about the subject.  The small groups would study local events or places. In this situation, the LMS would become more like the newscast – aggregating all of the input in one spot for everyone to benefit.

What if time and location became irrelevant for synchronous classes? What if you were grouped with a small group of people that lived near you when you sign up for a class, and then that group decided what day and time to meet for class?  The instructor would then send out assignments each week or maybe record a video for the group to work through. Maybe the instructor even met with each group.  then the groups send in their work to the class and the instructor aggregates all of the information coming in from each group and summarizes them for the entire class (which would essentially include all small groups no matter where they meet in the world).

Potentially, you could ave hundreds of students all meeting in a synchronous fashion, but all still in a way that fits their schedule.  This is, of course, another area where there is technology to do this… but we need one that is more specifically geared for educators.