Artificial Intelligence

Is AI Generated Art Really Coming for Your Job?

You might have noticed this Twitter thread about improvements in AI-generated art work. Well, if you are still on Twitter that is. Here is the thread – well, at least, until You-Know-Who “MySpaces” Twitter out of service:

So let’s take a look at this claim that AI-generated artwork is coming to disrupt people’s jobs in the very near future. First of all, yes it is really cool to be able to enter a prompt like that and get results like this. There is obviously a lot of improvement in the AI. It actually looks useful now. But saying “a less capable technology is developing faster than a stable dominant technology (human illustration)”…?

[tweet 1596581772644569090 hide_thread=’true’]

Whoa, now. Time for a reality check. AI art is just now somewhat catching up with where human art has been for hundreds of years. AI was programmed by people that had thousands of years of artistic development easily available in a textbook. So saying that it is “developing faster”? With humans being able to create photo-realistic drawings as well as illustrate any idea that comes to their mind – where is there left to “develop” in art?

That is like a new car company saying they are “developing new cars faster than the stable industry.” Or someone saying that they have blazed new technology in travel because they can cross the country faster in a car than a horse and wagon did in the past. The art field had to blaze trails for thousands of years to get where it is, and the AI versions are just basically cheating to play catch up (and it is still not there yet).

The big question is: can this technology come up with a unique, truly creative piece of artwork on its own? The answer is still “no.” And beating the Lovelace Test is not proof that the answer is “yes,” because the Lovelace test is not really a true test of creativity.

[tweet 1596680005638975489 hide_thread=’true’]

Yes, all artists stand on the shoulders of others, but there is still an element to creativity that involves blending those influences into something else that transcends being strictly derivative of existing styles. Every single example of AI artwork so far has been very derivative of specific styles of art, usually on purpose because you have to pick an existing artistic style just to get your images in the first place.

But even the example above of an “otter making pizza in Ancient Rome” is NOT a “novel, interesting way” by the standards that true artists use. I am guessing that Mollick is referring to the Lovelace 2.0 Test, which the creator of said test stated that “I didn’t want to conflate intelligence with skill: The average human can play Pictionary but can’t produce a Picasso.”

Of course, the average artist can’t produce an original painting on the level of Picasso either (unless they are just literally re-painting a Picasso, which many artists do to learn their craft). The people working on this particular AI Art Generator have basically advanced the skill of their AI to where it can pass the Lovelace 2.0 Test without really becoming truly creative. And honestly, “Draw me a picture of a man holding a penguin” is a sad measure of artistic creativity – no matter how complex you make that prompt as the test goes along.

But Mollick’s claims in this thread is just an example of people not understanding the field that they say is going to be disrupted. For example, marveling over correct lighting and composition? We have had illustration software that could do this correctly for decades.

[tweet 1596621251627335680 hide_thread=’true’]

Artists will tell you that in a real world situations, the time consuming part of creating illustrations is figuring out what the human that wants to art… actually wants. “The otter just looks wrong – make it look right!” is VERY common feedback. The client probably also created several specific details about the otter, plane, positions of things, etc that has to be present in any artwork they want. Then there are all of the things they had in their head that they didn’t write down. Pulling those details that out of clients is what professional artists are trained to do.

This is where AI in art, education, etc always falls apart: programmers always have to start with the assumption that people actually know what they want out of the AI generator in the first place. The clients that professionals work with rarely ever want something as simple as “otter on a plane using wifi.” The reality is that they rarely even have that specific or defined idea of what they want in the beginning. There is a difficult skill of learning to figure out what people actually want that the experts in AGI/strong AI/etc tell us is probably never going to be possible from AI.

So, is this a cool development that will become a fun tool for many of us to play around with in the future? Sure. Will people use this in their work? Possibly. Will it disrupt artists across the board? Unlikely. There might be a few places where really generic artwork is the norm and the people that were paid very little to crank them out will be paid very little to input prompts. Look, PhotoShop and asset libraries made creating company logos very, very easy a long time ago. But people still don’t want to take the 30 minutes it takes to put one together, because thinking through all the options is not their thing. You still have to think through those options to enter an AI prompt. And people just want to leave that part to the artists. The same thing was true about the printing press. Hundreds of years of innovation has taught us that the hard part of the creation of art is the human coming up with the ideas, not the tools that create the art.

The Problem of Learning Analytics and AI: Empowering or Resistance in the Age of “AI”

So where to begin with this series I started on Learning Analytics and AI? The first post started with a basic and over-simplified view of the very basics. I guess the most logical place to jump to is… the leading edge of the AI hype? Well, not really… but there is an event in that area happening this week, so I need to go there anyways.

I was a bit surprised that the first post got some attention – thank you to those that read it. Since getting booted out of academia, I have been unsure of my place in the world of education. I haven’t really said much publicly or privately, but it has been a real struggle to break free from the toxic elements of academia and figure out who I am outside of that context. I was obviously surrounded by people that weren’t toxic, and I still adjunct at a university that I feel supports its faculty… but there were still other systemic elements that affect all of us that are hard to process once you are gone.

So, anyway, I just wasn’t sure if I could still write anything that made a decent point, and I wasn’t too confident I did that great of a job writing about such a complex topic in a (relatively) short blog post last time. Maybe I didn’t, but even a (potentially weak) post on the subject seems to resonate with some. Like I said in the last post, I am not the first to bring any of this up. In fact, if you know of any article or post that makes a better point than I do, please feel free to add it in the comments.

So, to the topic at hand: this week’s Empowering Learners in the Age of AI conference in Australia. My concern with this conference is not with who is there – it seems to be a great group of very knowledgeable people. I don’t know some of them, but many are big names in the field that know their stuff. What sticks out to me is who is not there, as well as how AI is being framed in the brief descriptions we get. But neither of those points is specific to this conference. In fact, I am not really looking at the conference as much as some parts of the field of AI, with the conference just serving as proof that the things I am looking at are out there.

So first of all, to address the name of the conference. I know that “empowering learners” is a common thing to say not just in AI, but education in general. But it is also a very controversial and problematic concept as well. This is one concern that I hang on all of education and even myself as I like the term “empower” as well. No matter what my intentions (or anyone else’s), the term still places the institution and the faculty as the center of the power in the learning process – there to decide whether the learners get to be empowered or not. One of the best posts on this topic is by Maha Bali: The Other Side of Student Empowerment in a Digital World. At the end of the post, she gets to some questions that I want to ask of the AI field, including these key ones:

“In what ways might it reproduce inequality? How participatory has the process been? How much have actual teachers and learners, especially minorities, on the ground been involved in or consulted on the design, implementation, and assessment of these tools and pedagogies?”

I’ll circle back to those throughout the post.

Additionally, I think we should all question the “Age of AI” and “AI Society” part. It is kind of complicated to get into what AI is and isn’t, but the most likely form of AI we will see emerge first is what is commonly called “Artificial General Intelligence” (AGI), which a is deceptive way of saying “pretending to act like humans but not really be intelligent like we are.” AGI is really a focus on creating something that “does” the same tasks humans can, which is not what most people would attribute to an “Age of AI” or “AI Society.” This article on Forbes looks at what this means, and how experts are predicting that we are 10-40 years away from AGI.

Just as an FYI, I remember reading in the 1990s that we were 20-40 years away from AGI then as well.

So we aren’t near an Age of AI, probably not in many of our lifetimes, and even the expert options may not end up being true. The Forbes articles fails to mention that there were many problems with the work that claimed to be able to determine sexuality from images. In fact, there is a lot to be said about differentiating AI from BS that rarely gets brought up by the AI researchers themselves. Tristan Greene best sums it up in his article about “How to tell the difference between AI and BS“:

“Where we find AI that isn’t BS, almost always, is when it’s performing a task that is so boring that, despite there being value in that task, it would be a waste of time for a human to do it.”

I think it would have been more accurate to say you are “bracing learners for the age of algorithms” than empowering for an age of AI (that is at least decades off but may never actually happen according to some). But that is me, and I know there are those that disagree. So I can’t blame people for being hopeful that something will happen in their own field sooner than it might in reality..

Still, the most concerning thing about the field of AI is who is not there in the conversations, and the Empowering Learners conference follows the field – at least from what I can see on their website. First of all, where are the learners? Is it really empowering for learners when you can’t really find them on the schedule or in the list of speakers and panelists? Why is their voice not up front and center?

Even bigger than that is the problem that has been highlighted this week – but one that has been there all along:

The specific groups she is referring to are BIPOC, LGBTQA, and Disabilities. We know that AI has discrimination coded into it. Any conference that wants to examine “empowerment” will have to make justice front and center because of long existing inequalities in the larger field. Of course, we know that different people have different views of justice, but “empowerment” would also mean each person that faces discrimination gets to determine what that means. Its really not fair to hold a single conference accountable for issues that long existed before the conference did, but by using the term “empowerment” you are setting yourself up to a pretty big standard.

And yes, “empowerment” is in quotes because it is a problematic concept here, but it is the term the field of AI and really a lot of the world of education uses. The conference web page does ask “who needs empowering, why, and to do what?” But do they mean inequality? And if so, why not say it? There are hardly any more mentions of this question after it is brought up, much less anything connecting the question to inequality, in most of the rest of the program. Maybe it will be covered in conference – it is just not very prominent at all as the schedule stands. I will give them the benefit of the doubt until after the conference happens, but if they do ask the harder questions, then they should have highlighted that more on the website.

So in light of the lack of direct reference to equity and justice, the concept of “empowerment” feels like it is taking on the role of “equality” in those diagrams that compare “equality” with “equity” and “justice”:

Equality vs equity vs justice diagram
(This adaption of the original Interaction Institute for Social Change image by Angus Macguire was found on the Agents of Good website. Thank you Alan Levine for helping me find the attribution.)

If you aren’t going to ask who is facing inequalities (and I say this looking at the fields of AI, Learning Analytics, Instructional Design, Education, all of us), then you are just handing out empowerment the same to all. Just asking “who needs empowering, why, and to do what?” doesn’t get to critically examining inequality.

In fact, the assumption is being made by so many people in education that you have no choice but to utilize AI. One of the best responses to the “Equality vs Equity vs Justice” diagrams has come from Bali and others: what if the kids don’t want to play soccer (or eat an apple or catch a fish or whatever else is on the other side of the fence in various versions)?

Resistance is a necessary aspect of equity and justice. To me, you are not “empowering learners” unless you are teaching them how to resist AI itself first and foremost. But resistance should be taught to all learners – even those that “feel they are safe” from AI. This is because 1) they need to stand in solidarity with those that are the most vulnerable, to make sure the message is received, and 2) they aren’t as safe as they think.

There are many risks in AI, but are we really taking the discrimination seriously? In the linked article, Princeton computer science professor Olga Russakovsky said

“A.I. researchers are primarily people who are male, who come from certain racial demographics, who grew up in high socioeconomic areas, primarily people without disabilities. We’re a fairly homogeneous population, so it’s a challenge to think broadly about world issues.”

Additionally, (now former) Google researcher Timnit Gebru said that scientists like herself are

“some of the most dangerous people in the world, because we have this illusion of objectivity.”

Looking through the Empowering Learner event, I don’t see that many Black and Aboriginal voices represented. There are some People of Color, but not near enough considering they would be the ones most affected by discrimination that would impede any true “empowerment.” And where are the experts on harm caused by these tools, like Safiya Noble, Chris Gilliard, and many others? The event seems weighted towards those voices that would mostly praise AI, and it is a very heavily white set of voices as well. This is the way many conferences are, including those looking at education in general.

Also, considering that this is in Australia, where are the Aboriginal voices? Its hard to tell on the schedule itself. I did see on Twitter that the conference will start with an Aboriginal perspective. But when is that? In the 15 minute introductory session? That is no where near enough time for that. Maybe they are elsewhere on the schedule and just not noted well enough to tell. But why not make that a prominent part of the event rather than part of a 15 minute intro (if that is what it is)?

There are some other things I want to comment on about the future of AI in general:

  • The field of AI is constantly making references to how AI is affecting and improving areas such as medicine. I would refer you back to the “How to tell the difference between AI and BS” article for much  of that. But something that worries me about the entire AI field talking this way is that the are attributing “artificial intelligence” to things that boil down to advanced pattern recognition mainly using human intelligence. Let’s take, for example, recognizing tumors in scans. Humans program the AI to recognize patterns in images that look like tumors. Everything that the AI knows to look for comes directly from human intelligence. Just because you can then get the algorithm to repeat what the humans programmed it to thousands of times per hour, that doesn’t make it intelligence. It is human intelligence pattern recognition that has been digitized, automated, and repeated rapidly. This is generally what is happening with AI in education, defense, healthcare, etc.
  • Many leaders in education in general like to say that “institutions are ill-prepared for AI” – but how about how ill-prepared AI is for the equity and reality?
  • There is also often talk in the AI community about building trust between humans and machines that we see examples of at the conference as well: “can AI truly become a teammate in group learning or a co-author of a ground-breaking scientific discovery?” I don’t know what the speaker plans to say, but the answer is no. No we shouldn’t build trust and no we shouldn’t anthropomorphize AI. We should always be questioning it. But we also need to be clear, again, that AI is not the one that is writing (or creating music or paintings). This is the weirdest area of AI – they feed a bunch of artistic or music or literary patterns into AI, tell it how to assemble the patterns, and when something comes out it is attributed to AI rather than the human intelligence that put it all together. Again, the machine being able to repeat and even refine what the human put there in the first place is not the machine creating it. Take, for example, these different AI generated music websites. People always send these to me and say “look how well the machine put together ambient or grindcore music or whatever.” Then I  listen… and it is a mess. They take grindcore music and chop it up in to bits and then run those bits through pattern recognition and spit out this random mix – that generally doesn’t sound like very good grindcore. Ambient music works the best to uninitiated ears, but to fans of the music it still doesn’t work that great.
  • I should also point out about the conference that there is a session on the second day that asks “Who are these built for? Who benefits? Who has the control?” and then mentions “data responsibility, privacy, duty of care for learners” – which is a good starting point. Hopefully the session will address equity, justice, and resistance specifically. The session, like much of the field of AI, rests on the assumption that AI is coming and there is nothing you can do to resist it. Yes the algorithms are here, and it is hard to resist – but you still can. Besides, experts are still saying 10-40 years for the really boring stuff to emerge as I examined above.
  • I also hope the conference will discuss the meltdown that is happening in AI-driven proctoring surveillance software.
  • I haven’t gotten much into surveillance yet, but yes all of this relies on surveillance to work. See the first post. Watch the Against Surveillance Teach-In Recording.
  • I was about to hit publish on this when I saw an article about a Deepfake AI Santa that you can make say whatever you want. The article says “It’s not nearly as disturbing as you might think”… but yes, it is. Again, people saying something made by AI is good and realistic when it is not. The Santa moves and talks like a robot with zero emotion. Here again, they used footage of a human actor and human voice samples and the “AI” is an algorithm that chops it up into the parts that makes your custom message. How could this possibly be misused?
  • One of the areas of AI that many in the field like to hype are “conversational agents” aka chatbots. I want to address that as well since that is an area that I have (tried) to research. The problem with researching agents/bots is that learners just don’t seem to be impressed with them – it’s just another thing to them. But I really want to question how these count as AI after having created some myself. The process for making a chatbot is that you first organize a body of information into chunks of answers or statements that you want to send as responses. You then start “training” the AI to connect what users type into the agent (aka “bot”) with specific statements or chunks of information. The AI makes a connection and sends the statement or information or next question or video or whatever it may be back to the user. But the problem is, the “training” is you guessing dozens of ways that the person might ask a question or make a statement (including typos or misunderstandings) that matches with the chunk of information you want to send back. You literally do a lot of the work for the AI by telling it all the ways someone might type something into the agent that matches each chunk of content. They want at least 20 or more. What this means is that most of the time, when you are using a chatbot, it gives you the right answer because you typed in one of the most likely questions that a human guessed and added to the “training” session. In the rare cases where some types something a human didn’t guess, then the Natural Language Processing kicks in to try and guess the best match. But even then it could be a percentage of similar words more than “intelligence.” So, again, it is human intelligence that is automated and re-used thousands of times a minute – not something artificial that has a form of intelligence. Now, this might be useful in a scenario when you have a large body of information (like an FAQ bot for the course syllabus) that could use something better than a search function. Or maybe a branching scenarios lesson. But it takes time to create a good chatbot. There is still a lot of work and skill to creating the questions and responses well. But to use chatbots for a class of 30, 50, 100? You probably will spend so much time making it that it would be easier to just talk to your students.
  • Finally, please know that I realize that what I am talking about still requires a lot of work and intelligence to create. I’m not doubting the abilities of the engineers and researchers and others that put their time into developing AI. I’m trying to get at the pervasive idea that we are in an Age of AI that can’t be avoided. Its a pervasive idea that was even made in a documentary web series a year ago. I also question whether “artificial intelligence” is the right term for all of this, rather than something more accurate like “automation algorithms.”

Again, everything I touch on here is not as much about this conference, as it is about the field of AI since this conference is really just a lot of what is in the AI field concentrated into two days and one website. The speakers and organizers might have already planned to address everything I brought up here a long time ago, and they just didn’t get it all on the website. We will see – there are some sessions with no description and just a bio. But still, at the core of my point, I think that educators need to take a different approach to AI than we have so far (maybe by not calling it that when it rarely is anything near intelligent) by taking justice issues seriously. If the machine is harming some learners more than others, the first step is to teach resistance, and to be successful in that all learners and educators need to join in the resistance.

The Problem of Learning Analytics and AI

For some time now, I have been wanting to write about some of the problems I observed during my time in the Learning Analytics world (which also crosses over into Artificial Intelligence, Personalization, Sentiment Analysis, and many other areas as well). I’m hesitant to do so because I know the pitchforks will come out, so I guess I should point out that all fields have problems. Even my main field of instructional design is far from perfect. Examining issues with in a field (should be) a healthy part of the growth of a field. So this will probably be a series of blog posts as I look at publications, conferences, videos, and other aspects of the LA/PA/ML/AI etc world that are in need of a critical examination. I am not the first or only person to do this, but I have noticed a resistance by some in the field to consider these viewpoints, so hopefully adding more voices to the critical side will bring more attention to these issues.

But first I want to step back and start with the basics. At the core of all analytics, machine learning, AI, etc are two things: surveillance and algorithms. Most people wouldn’t put it this way, but let’s face it: that is how it works. Programs collect artifacts of human behavior by looking for them, and then process those through algorithms. Therefore, the core of all of this is surveillance and algorithms.

At the most basic level, the surveillance part is a process of downloading a copy of data from a database that was intentionally recording data. That data is often a combination of click-stream data, assignment and test submissions, discussion forum comments, and demographic data. All of this is surveillance, and in many cases this is as far as it goes. A LOT of the learning analytics world is based on click stream data, especially with an extreme focus on predictive analytics. But in a growing number of examples, there are also more invasive forms of surveillance added that rely on video recordings, eye and motion detection, bio-metric scans, and health monitoring devices. The surveillance is getting more invasive.

I would also point out that none of this is accidental. People in the LA and AI fields like to say that digital things “generate” data, as if it is some unintentional by-product of being digital: “We turned on this computer, and to our surprise, all this data magically appeared!”

Data has to be intentionally created, extracted, and stored to exist in the first place. In fact, there usually is no data in any program until programmers decide they need it. They will then create a variable to store that data for use within the program. And at this moment is where bias is introduced. The reason why certain data – like names, for example – are collected and others aren’t has to do with a bias towards controlling who has access and who doesn’t. Then that variable is given a name – it could be “XD4503” for all the program cares. But to make it easier for programmers to work together, they create variables names that can be understood by everyone on the team: “firstName,” “lastName,” etc.

Of course, this designation process introduces more bias. What about cultures that have one name, or four names? What about those that have two-part names, like the “al” that is common in the Arabic names, but isn’t really used for alphabetizing purposes? What about cultures that use their surname as their first name? What about random outliers? When I taught eighth grade, I had two students that were twins, and their parents gave them both nearly identical sets of five names. The only difference between the two was that the third name was “Jevon” for one and “Devon” for the other. So much of the data that is created – as well as how it is named, categorized, stored, and sorted – is biased towards certain cultures over others.

Also note here that there is usually nothing that causes this data to leave the program utilizing it. In order for some outside process or person to see this data, programmers have to create a method for displaying and / or storing that data in database. Additionally, any click stream, video, or bio-metric data that is stored has to be specifically and intentionally captured in ways that can be stored. For example, a click in itself is really just an action that makes a website execute some function. It disappears after that function happens – unless someone creates a mechanism for recording what was clicked on, when it was clicked, what user was logged in to do the click, and so on.

All of this to say that none of this is coincidental, accidental, or unplanned. There is a specific plan and purpose for every piece of data that is created and collected outside of the program utilizing them. None of the data had to be collected just because it was magically “there” when the digitials were turned on. The choice was made to create the data through surveillance, and then store it in a way that it could be used – perpetually if needed.

Therefore, different choices could be made to not create and collect data if the people in control wanted it that way. It is not inevitable that data has to be generated and collected.

Of course, most of the few people that will read this blog already know all of this. The reason I state this all here is for anybody that might still be thinking that the problems with analytics and AI is created during the design of the end user products. For example, some believe that the problems that AI proctoring has with prejudice and discrimination started when the proctoring software was created… but really this part is only the continuation of problems that started when the data that these AI systems utilized was intentionally created and stored.

I think that the basic fundamental lens or mindset or whatever you want to call it for publishing research or presenting at conferences about anything from Learning Analytics to AI has to be a critical one rooted in justice. We know that surveillance and algorithms can be racist, sexist, ablest, transphobic, and the list of prejudices goes on. Where people are asking the hard questions about these issues, that is great. Where the hard questions seem to be missing, or people are not digging deep enough to see the underlying biases as well, I want to blog about it. I have also noted that the implementation of LA/ML/AI tools in education too often lacks input from the instructional design / learning sciences / etc fields – so that will probably be in the posts as well.

While this series of posts is not connected to the Teach-In Against Surveillance, I was inspired to get started on this project based on reflecting on why I am against surveillance. Hopefully you will join the Teach-In tomorrow, and hopefully I will get the next post on the Empowering Learners for the Age of AI conference written in this lifetime. :)

People Don’t Like Online Proctoring. Are Institutional Admins Getting Why?

You might have noticed a recent increase in the complaints and issues being leveled against online proctoring companies. From making students feeling uncomfortable and/or violated, to data breaches and CEOs possibly sharing private conversations online, to a growing number of student and faculty/staff petitions against the tools, to lawsuits being leveled against dissenters for no good reason, the news has not been kind to the world of Big Surveillance. I hear the world’s tiniest violin playing somewhere.

It seems that the leadership at Washington State University decided to listen to concerns… uhhh… double down and defend their position to use proctoring technology during the pandemic. While there are great threads detailing different problems with the letter, I do want to focus in on a few statements specifically. Not to specifically pick on this one school, but because WSU’s response is typical of what you hear from too many Higher Ed administrations. For example, when they say…

violations of academic integrity call into question the meaningfulness of course grades

That is actually a true statement… but not in the way it was intended. The intention was to say that cheating hurts academic integrity because it messes up the grade structures, but it could also be taken to say that cheating calls into highlights the problem with the meaningfulness of grades because cheating really doesn’t affect anyone else.

Think about it: someone else cheats, and it casts doubt on the meaning of my grade if I don’t cheat? How does that work exactly? Of course, this is a nonsense statement that reals highlights how cheating doesn’t change the meaning of grades for anyone else. Its like the leaders at this institution are right there, but don’t see the forest for the trees: what exactly does a grade mean if the cheaters that get away with it don’t end up hurting anyone but themselves? Or does cheating only cause problems for non-cheaters when the cheaters get caught? How does that one work?

But let’s focus here: grades are the core problem. Yes, many people feel they are arbitrary and even meaningless. Still others say they are unfair, while some look at them as abusive. At the very least, you really should realize grades are problematic. Students can guess and get a higher grade than what they really actually know. Tests can be gamed. Questions have bias and discrimination built in too many times. And so on. Online proctoring is just an attempted fix for a problem that existed long before “online” was even an option.

But let’s see if the writers of the letter explain exactly how one person cheating harms someone else… because maybe I am missing something:

when some students violate academic integrity, it’s unfair for the rest. Not only will honest students’ hard work not be properly reflected…. Proctoring levels the playing field so that students who follow the rules are not penalized in the long run by those who don’t.

As someone that didn’t cheat in school, I am confused as to how this exactly works. I really never spent a single minute caring about other students’ cheating. You knew it happened, but it didn’t affect you, so it was their loss and not yours. In fact, you never lost anything in the short or long run from other student’s cheating. I have no clue how my hard work was not “properly reflected” by other students’ cheating.

(I would also note that this “level the playing field” means that they assume proctoring services catch all “cheaters” online, just like have instructors in the classroom on campus meant that all of the “cheaters” in those classes. But we all know that is not the case.)

I have never heard a good answer for how does supposed “penalization” works. Most of the penalization I know of from classes are systemic issues against BIPoC students that happens in ways that proctoring never deals with. You sometimes wish institutions would put as much money into fighting that as they would spying through student cameras….

But what about the specific concerns with how these services operate?

Per WSU’s contract, the recorded session is managed by an artificial intelligence “bot” and no human is on the other end at ProctorU watching the student. Only the WSU instructor can review the recorded session.

A huge portion of the concern about proctoring has been about the AI bots – which are here presented as an “it’s all okay because” solution…? Much of the real concern many have expressed is with the algorithms themselves and how they are usually found to be based on racist, sexist, and ableist norms. Additionally, the other main concern is what the instructor might see when they do review a recording of a student’s private room. No part of the letter in question addresses any of the real concerns with the bigger picture.

(It is probably also confusing to people whether or not someone is watching on the other side of the camera when there are so many complaints online from students that have had issues with human proctors, especially ones that were “insulting me by calling my skin too dark” as one complaint states.)

The response then goes on to talk about getting computers that will work with proctoring service to students that need them, or having students come in to campus for in-person proctoring if they just refuse to use the online tool. None of this addresses the concerns of AI bias, home privacy, or safety during a pandemic.

The moral of the point I am making here is this: if you are going to respond to concerns that your faculty and staff have, make sure you are responding to the actual concerns and not some imaginary set of concerns that few have expressed. There is a bigger picture as to why people are objecting to these services, which – yes – may start with feeling like they are being spied on by people and/or machines. But just saying “look – no people! (kind of)” is not really addressing the core concerns.

So What Do You Want From Learning Analytics?

If you haven’t noticed lately, there is a growing area of concern surrounding the field of learning analytics (also sometimes combined with artificial intelligence). Of course, there has always been some backlash against analytics in general, but I definitely noticed at the recent Learning Analytics and Knowledge (LAK) conference that it was more than just a random concern raised here and there that you usually get at any conference. There were several voices loudly pointing out problems both online and in the back channel, as well as during in-person conversations at the conference. Many of those questioning what they saw were people with deep backgrounds in learning theory, psychology, and the history of learning research. But its not just people pointing out how these aspects are missing from so much of the Learning Analytics field – it is also people like Dr. Maha Bali questioning the logic of how the whole idea is supposed to work in blog posts like Tell Me, Learning Analytics…

I have been known to level many of the current concerns at the Learning Analytics (LA) field myself, so I probably should spell out what exactly it is that I want from this field as far as improvement goes. There are many areas to touch on, so I will cover them in no particular order. This is just what comes to mind off the top of my head (probably formed by my own particular bias, of course):

  • Mandatory training for all LA researchers in the history of educational research, learning theory, educational psychology, learning science, and curriculum & instruction. Most of the concerns I heard voiced at any LAK I have attended was that these areas are sorely missing in several papers and posters. Some papers were even noticed as “discovering” basic educational ideas, like students that spend more time in a class perform better. We have known this from research for decades, so… why was this researched in the first place? And why was none of this earlier research cited? But you see this more than you should in papers and posters in the LA field – little to no theoretical backing, very little practical applications, no connection to psychology, and so on. This is a huge concern, because the LAK Conference Proceedings is in the Top 10 Educational Technology journals as ranked by Google. But so many of the articles published there would not even go beyond peer review in many of the other journals in the Top 10 because of their lack of connection to theory, history, and practice. This is not to say these papers are lacking rigor for what they include – it is just that most journals in Ed-Tech require deep connections to past research and existing theory to even be considered. Other fields do not require that, so it is important to note this. Also, as many have pointed out, this is probably because of the Computer Science connection in LA. But we can’t forego a core part of what makes human education, well… human… just because people came from a background where those aspects aren’t as important. They are important to what makes education work, so just like a computer engineer that wants to get into psychology would have to learn the core facets of psychology to publish in that area, we should require LA researchers to study the core educational topics that the rest of us had to study as well. This is, of course, something that could be required to change many areas in Education itself as well – just having an education background doesn’t mean one knows a whole lot about theory and/or educational research. But I have discussed that aspect of the Educational world in many places in the past, so now I am just focusing on the LA field.
  • Mandatory training for all LA researchers in structural inequalities and the role of tech and algorithms in creating and enforcing those inequalities. We have heard the stories about facial recognition software not recognizing black faces. We know that algorithms often contain the biases of their creators. We know that even the prefect algorithms have to ingest imperfect data that will contain the biases of those that generated it. But its time to stop treating equality problems as an after thought, to be fixed only when they get public attention. LA researchers need to be trained in recognizing bias by the people that have been working to fight the biases themselves. Having a white male instructor mention the possibility of bias here and there in LA courses is not enough.
  • Require all LA research projects to include instructional designers, learning theorists, educational psychologists, actual instructors, real students, people trained in dealing with structural inequalities, etc as part of the research team from the very beginning. Getting trained in all of the fields I mentioned above does not make one an expert. I have had several courses on educational psychology as part of my instructional design training, but that does not make me an expert in educational psychology. We need a working knowledge of other fields to inform our work, but we also need to collaborate with experts as well. People with experience in these fields should be a required part of all LA projects. These don’t all have to separate people, though. A person that teaches instructional design would possibly have experience in several areas (practical instruction, learning theory, structural inequality, etc). But you know who’s voice is incredibly rare in the LA research? Students. Their data traces DO NOT count as their voice. Don’t make me come to a conference with a marker and strike that off your poster for you.
  • Be honest about the limitations and bias of LA. I read all kinds of ideas for what data we need in analytics – from the idea that we need more data to capture complex ways learning manifests itself after a course ends, to the idea that analytics can make sense of the word around us. The only way to get more (or better) data is to increase surveillance in some way or form. The only way to make more sense is to get more data, which means… more surveillance. We should be careful not to turn our entire lives into one mass of endless data points. Because even if we did, we wouldn’t be capturing enough to really make sense of the world. For example, we know that click stream data is a very limited way to determine activity in a course. A click in an online course could mean hundreds of different things. We can’t say that this data tells us what learners are doing or watching or learning – only just what they are clicking on. Every data point is just that – a click or contact or location or activity with very little context and very little real meaning by itself. Each data point is limited, and each data point has some type of bias attached to it. Getting more data points will not overcome limitations or bias – it will collect and amplify them. So be realistic and honest with those limitations, and expose the bias that exists.
  • Commit to creating realistic practical applications for instructors and students. So many LA projects are really just ways to create better reports for upper level admin. Either that, or ways to try and decrease drop-outs (or increase persistence across courses as the new terminology goes). The admin needs their reports and charts, so you can keep doing that. But educators need more than drop-out/persistence stuff. Look, we already have a decent to good idea what causes those issues and what we can do to improve them. Those solutions take money, and throwing more data at them is not going to decrease the need for funding once a more data-driven problem (which usually look just like the old problems) is identified. Please: don’t make “data-driven” become a synonymy for “ignore past research and re-invent the wheel” in educators eyes. Look for practical ways to address practical issues (within the limitations of data and under the guiding principle of privacy). Talk to students, teachers, learning theorists, psychologists, etc while you are just starting to dig into the data. See what they say would be a good, practical way to do something with the data. Listen to their concerns. Stop pushing for more data when they say stop pushing.
  • Make protecting privacy your guiding principle. Period. So much could be said here. Explain clearly what you are doing with the data. Opt-in instead of opt-out. Stop looking for ways to squeeze every bit of data out of every thing humans do and say (its getting kind of gross). Remember that while the data is incomplete and biased, it is still a part of someone else’s self-identity. Treat it that way. If the data you want to collect was actual physical parts of a person in real life – would you walk around grabbing it off of them the way you are collecting data digitally now? Treat it that way, then. Or think of it this way: if data was the hair on our heads, are you trying to rip or cut it off of peoples’ heads without permission? Are you getting permission to collect the parts that fall to the floor during a haircut, or are you sneaking in to hair cutting places to try and steal the stuff on the floor when no one is looking? Or even worse – are you digging through the trash behind the hair salon to find your hair clippings? Also – even when you have permission – are you assuming that just because the person who got the hair cut is gone, that this means the identity of each hair clipping is protected… or do you realize that there are machines that can identify DNA from those hair clippings still?
  • Openness. All of what I have covered here will require openness – with the people you collect data from, with the people you report the analytical results to, with the general public about the goals and results, etc. If you can’t easily explain the way the algorithms are working because they are so complex, then don’t just leave it there, Spend the time to make the algorithms make sense, or change the algorithm.

There are probably more that I am missing, or ways that I failed to explain the ones I covered correctly. If you are reading this and can think of additions or corrections, please let me know in the comments. Note: the first bullet point was updated due to misunderstandings about the educational journal publishing system. Also see the comments below for good feedback from Dr. Bali.

Updating Types of Interactions in Online Education to Reflect AI and Systemic Influence

One of the foundation concepts in instructional design and other parts of the field of education are the types of interaction that occur in the educational process online. In 1989, Michael G. Moore first categorized three types of interaction in education: student-teacher, student-student, and student-content. Then, in 1994, Hillman, Willis, and Gunawardena expanded on this model, adding student-interface interactions. Four years later, Anderson & Garrison (1998) added three more interaction types to account for advances in technology: teacher-teacher, teacher-content, and content-content. Since social constructivist theory did not quite fit into these seven types of interaction, Dron decided to propose four more types of interaction in 2007: group-content, group-group, learner-group, and teacher-group. Some would argue that “student-student” and “student-content” still cover these newer additions, and to some degree that is true. But it also helps to look at the differences between these various terms as technology has advanced and changed interactions online – so I think the new terms are helpful. More recently, proponents of connectivism have proposed acknowledging patterns of “interactions with and learning from sets of people or objects [which] form yet another mode of interaction” (Wang, Chen, & Anderson, 2014, p. 125). I would call that networked with sets of people and/or objects.

The instructional designer within me likes to replace “student” with “learner” and “content” with “design” to more accurately describe the complexity of learners that are not students and learning designs that are not content. However, as we rely more and more on machine learning and algorithms, especially at the systemic level, we are creating new things that learners will increasingly be interacting with for the foreseeable future. I am wondering if it is time to expand this list of interactions to reflect that? Or is it long enough as it is?

So the existing ones I would keep, with “learner” exchanged for “student” and “design” exchanged for “content”:

  • learner-teacher (ex: instructivist lecture, learner teaching the teacher, or learner networking with teacher)
  • learner-learner (ex: learner mentorship, one-on-one study groups, or learner teaching another learner)
  • learner-design (ex: reading a textbook, watching a video, listening to audio, completing a project, or reading a website)
  • learner-interface (ex: web-browsing, connectivist online interactions, gaming, or computerized learning tools)
  • teacher-teacher (ex: collaborative teaching, cross-course alignment, or professional development)
  • teacher-design (ex: teacher-authored textbooks or websites, teacher blogs, or professional study)
  • group-design (ex: constructivist group work, connectivist resource sharing, or group readings)
  • group-group (ex: debate teams, group presentations, or academic group competitions)
  • learner-group (ex: individual work presented to group for debate, learner as the teacher exercises)
  • teacher-group (ex: teacher contribution to group work, group presentation to teacher)
  • networked with sets of people or objects (ex: Connectivism, Wikipedia, crowdsourced learning, or online collaborative note-taking)

The new ones I would consider adding include:

  • algorithm-learner (ex: learner data being sent to algorithms; algorithms sending communication back to learners as emails, chatbot messages, etc)
  • algorithm-teacher (ex: algorithms communicating aggregate or individual learner data on retention, plagiarism, etc)
  • algorithm-design (ex: algorithms that determine new or remedial content; machine learning/artificial intelligence)
  • algorithm-interface (ex: algorithms that reformat interfaces based on input from learners, responses sent to chatbots, etc)
  • algorithm-group (ex: algorithms that determine how learners are grouped in courses, programs, etc)
  • algorithm-system (ex: algorithms that report aggregate or individual learner data to upper level admin)
  • system-learner (ex: system-wide initiatives that attempt to “solve” retention, plagiarism, etc)
  • system-teacher (ex: cross-curricular implementation, standardized teaching approaches)
  • system-design (ex: degree programs, required standardized testing, and other systemic requirements)

Well… that gets too long. But I suspect that a lot of the new additions listed would fall under the job category of what many call “learning engineering” maybe? You might have noticed that it appears as if I removed “content-content” – but that was renamed “algorithm-design,” as that is mainly what I think of for “content-content.” But I could be wrong. I also left out “algorithm-algorithm,” as algorithms already interface with themselves and other algorithms by design. That is implied in “algorithm-design,” kind of in the same way I didn’t include learners interacting with themselves in self-reflection as that is implied in “learner-learner.” But I could be swayed by arguments for including those as well. I am also not sure how much “system-interface” interaction we have, as most systems interact with interfaces through other actors like learners, teachers, groups, etc. So I left that off. I also couldn’t think of anything for “system-group” that was different from anything else already listed as examples elsewhere. And I am not sure we have much real “system-system” interaction outside of a few random conversations at upper administrative levels that rarely trickle down into education without being vastly filtered through systemic norms first. Does it count as “system-system” interaction in a way that affects learning if the receiving system is going to mix it with their existing standards before approving and disseminating it first? I’m not sure.

While many people may not even see the need for the new ones covered here, please understand that these interactions are heavily utilized in surveillance-focused Ed-Tech. Of course, all education utilizes some form of surveillance, but to those Ed-Tech sectors that make it their business to promote and sell surveillance as a feature, these are interactions that we need to be aware of. I would even contend that these types of interaction are more important behind the scenes of all kinds of tech than many of us realize. So even if you disagree with this list, please understand that these interactions are a reality.

So – that is 20 types of interaction, with some more that maybe should have been included or not depending on your viewpoint (and I am still not sure we have advanced enough with “algorithm-interface” yet to give it it’s own category, but I think we will pretty soon). Someone may have done this already and I just couldn’t find it in a search – so I apologize if I missed others’ work. None of this is to say that any of these types of interactions are automatically good for learners just because I list them here – they just are the ones that are happening more and more as we automate more and more and/or take a systems approach to education. In fact, these new levels could be helpful in informing critical dialogue about our growing reliance on automation and surveillance in education as well.

Artificial Intelligence and Knowing What Learners Know Once They Have “Learned”

One of the side effects – good or bad – of our increasing utilization of Artificial Intelligence in education is that it brings to light all of the problems we have with knowing how a learner has “learned” something. This specific problem has been discussed and debated in Instructional Design courses for decades – some of my favorite class meetings in grad school revolved around digging into these problems. So it is good to see these issues being brought to a larger conversation about education, even if it is in the context of our inevitable extinction at the hands of our future robot overlords.

Dave Cornier wrote a very good post about the questions to ask about AI in learning. I will use that post to direct some responses mostly back to the AI community as well as those utilizing AI in education. Dave ends up questioning a scenario that is basically the popular “Netflix for Education” approach to Educational AI: the AI perceives what the learners choose as their favorite learning resource by likes, view counts, etc, and then proposes new resources to specific learners to help them learn more, in the way Netflix recommends new shows to watch based on the popularity of other shows (which were connected to each other by popularity metrics as well).

This, of course, leads to the problem that Dave points out: “If they value knowledge that is popular, then knowledge slowly drifts towards knowledge that is popular.” Popular, as we all learn at some point, does not always equal good, helpful, correct, etc. However, people in the AI field will point out that they can build a system that relies on the expertise of experts and teachers in the field rather than likes, and I get that. Some have done that. But there is a bigger problem here.

Let’s back up to the part from Dave’s post about how AI accomplishes recommendations by simplifying the learners down to a few choices, much in the same way Netflix simplifies viewing choices down to a small list of genres. This is often true. However, this is true not because programmers wanted it that way – this is the model they inherited from education itself. Sure, it is true that in an ideal learning environment, the teacher talks to all learners and gets to make personal teaching choices for each one because of that. But in reality, most classes design one pathway for all learners to take: read this book, listen to these lectures, take this test, answer this discussion question while responding to two peers, wash, rinse, repeat.

AI developers know this, and to their credit, they are offering personalized learning solutions that at least expand on this. Many examinations of the problems with AI skip over this part and just look at ideal classrooms where learners and instructors have time to dig into individual learner complexities. But in the real world? Everyone follows the one path. So adding 7 or 10 or more options to the one that now to exists (for most)? Its at least a step in right direction, right?

Depends on who you ask. But that is another topic for anther day.

This is kind of where a lot of what is now called “personalized education” is at. I compare this state to all of those personalized gift websites, where you can go buy a gift like a mouse pad and get a custom message or name printed on it. Sure, the mouse pad is “personalized” with my name… but what if I didn’t need a mouse pad in the first place? You might say “well, there were only a certain set of gifts available and that was the best one out of the choices that were there.”

Sure, it might be a better gift than some plain mouse pad from Walmart to the person that needed a mouse pad. But for everyone else – not so much.

Like Dave and many have pointed out – someone is choosing those options and limiting the number of them. But to someone going from the linear choice of local TV stations to Netflix, at first that choice seems awesome. However, soon you start noticing the limitations of only watching something on Netflix. Then it starts getting weird. If I liked Stranger Things, I would probably like Tidying Up with Marie Kondo? Really?

The reality is, while people in the AI field will tell you that AI “perceives the learner and knowledge in a field,” it is more accurate to say that the AI “records choices that the learner makes about knowledge objects and then analyzes those choices to find patterns between the learner and knowledge object choices in ways that are designed to be predictive in some way for future learners.” If you just look at all that as “perceiving,” then you probably will end up with the Netflix model and all the problems that brings. But if you take a more nuanced look at what happens (it’s not “perceiving” as much as “recording choices” for example), and connect it with a better way of looking at the learner process, you will end up with better models and ideas.

So back to how we really don’t have that great of an idea of how learning actually happens in the brain. There are many good theories, and Stephen Downes usually highlights the best in emerging research in how we really understand the actual process of learning in the brain. But since there is still so much we either a) don’t know, or b) don’t know how to quantify and measure externally from the brain – then we can’t actually measure “learning” itself.

As a side note: this is, quite frankly, where most of the conversation on grading goes wrong. Grades are not a way to measure learning. We can’t stick a probe on people’s heads and measure a “learning” level in human brains. So we have to have some kind of external way to figure out if learning happens. As Dr. Scott Warren puts it: its like we are looking at this brick wall with a few random windows that really aren’t in the right spot and are trying to figure out what is happening on the other side of the wall.

Some people are clinging to the outmoded idea that brains are like computers: input knowledge/skills, output learning. Our brains don’t work like that. But unfortunately, that is often the way many look at the educational process. Instructors design some type of input – lectures, books, training, videos, etc – and then we measure the output with grades as way to say if “learning happened” or not.

The reality is, we technically just point learners towards something that they can use in their learning process (lectures, books, videos, games, discussions, etc), they “do” the learning, and then we have to figure out what they learned. Grades are a way to see how learners can apply what they learned to a novel artifact – a test, a paper, a project, a skill demonstration, etc. Grades in no way measure what students have learned, but rather how students can apply what they learned to some situation or context determined by someone else. That way – if they apply it incorrectly by, say, getting the question wrong – we assume they haven’t learned it well enough. Of course, an “F” on a test could mean the test was a poor way to apply the knowledge as much as it could say that the learner didn’t learn. Or that the learner got sidetracked while taking the test. Or, so on….

The learning that happens in between the choosing of the content/context/etc and the application of the knowledge gained on a test or paper or other external measurement is totally up to the learner.

So that is what AI is really analyzing in many designs – it is looking at what choices were made before the learning and what the learner was able to do with their learning on the other side of the learning based on some external application of knowledge/skills/etc. We have to look at AI something that affects and/or measures the bookends to the actual learning.

Rather than the Netflix approach to recommendations, I would say a better model to look to is the Amazon model of “people also bought this.” Amazon looks at each thing they sell as an individual object that people will connect in various ways to other individual objects – some connects that make sense, others that don’t. Sometimes people look at one item and buy other similar items instead, sometimes people buy items that work together, and sometimes people “in the know” buy random things that seem disconnected to newbies. The Amazon system is not perfect, but it does allow for greater individuality in purchasing decisions, and doesn’t assume that “because you bought this phone, you might also want to buy this phone as well because it is a phone, too.”

In other words, the Amazon model can see the common connections as well as the uncommon connections (even across their predefined categories), and let you the consumer decide which connections work for you or not. The Netflix model looks for the popular/common connections within their predefined categories.

I would submit that learners need ways to learn that can look at common learning pathways as well as uncommon pathways – especially across any categories we would define for them.

Of course, Amazon can collect data in ways that would be illegal (for good reason) in education, and the fact that they have millions of transactions each day means that they get detailed data about even obscure products in ways that would be impossible at a smaller scale in education. In no way should this come across as me proposing something inappropriate like “Amazon for Education!” The point I am getting at here is that we need a better way to look at AI in education:

  • Individuals are complex, and all systems need to account for complexity instead of simplifying for the most popular groups based on analytics.
  • AI should not be seen as something that perceives the learner or their knowledge or learning, but one that collects incomplete data on learners choices.
  • The goal of this collection should not just be to perceive learners and content, but to understand complex patterns made by complex people.
  • The categories and patterns selected by the creators of AI applications should not become limitations on the learners within that application.
  • While we have good models for how we learn, the actual act of “learning” should still be treated as a mysterious process (until that changes – if ever).
  • AI, like all education, does not measure learning, but how learning that occurred mysteriously in the learner was applied to an external context or artifact. This will be a flawed process, so the results of any AI application should be viewed within the bias and flaws created by the process.
  • The learners perception of what they learned and how well they were able to apply it to external context/artifact is mostly ignored or discarded as irrelevant self-reported data, and that should stop.