Get Ready To Be Replaced By AI

[spork]

Prepare to be assimilated:

https://www.nytimes.com/2019/12/18/education/artificial-intelligence-tutors-teachers.html.

[polly_mer]

Ideas the article neglected:

* Having even less human contact means people who weren't all that thrilled about learning math in the first place get even less reinforcement in the fun and positive aspects of math once one gets the basics mastered.

* All the math that's easy for a program to grade by doing is the part that the average human will plug into a computer when encountered in the wild.  The hard part of algebra is changing the paragraph of English into a solvable equation set. Solving algebraic equations is the easy part because there are rules to follow and in fact are programmed into the AI.  What the AI is doing is identifying patterns in how students are failing to apply the known rules.

* The math that's really hard to teach, hard to learn, and increasingly vital to our society has nothing to do with the applications of AI in the article.  The AI in the article is a better form of practice for the basics; it doesn't at all address the STEM pipeline problems I see every day in my professional life.

[downer]

Am I going to be replaced or assimilated?

I don't think the technology is there yet for most humanities classes. Maybe it never will be. But I will be happy to hand over paper and exam grading to the computers.

Why will people need to learn the stuff that AI can teach? Won't their jobs also be replaced by AI too?

[polly_mer]

Why will people need to learn the stuff that AI can teach? Won't their jobs also be replaced by AI too?

It depends.  Using AI for the baseline practice can be a good use of resources.  If, however, people never build enough on the baseline to go beyond what the AI can do, then yeah, those jobs are gone.  For example, one cannot do partial differential equations without algebra and we're nowhere near replacing people who can look at the world and write the partial differential equations to describe it. 

However, as I've said elsewhere, the pages and pages of algebra and even some parts of calculus are no longer part of the job I have; if one just needs the answer, then one types it into a program and gets the answer.  The math parts we need are writing the big equations that need hours/months on a supercomputer and new ways to solve those equations faster.  We need the humans who can figure out where those equations are inadequate, what it means that when we get inconsistent results, and how to fix the programs.

However, we're already well into the replacement for the hordes of clerks for simple data entry and processing like timecards.  We have software that can write other software. Anything that a human can do by following a list of concrete instructions is something we can program into a computer.  We only need the humans for the parts we can't program and can't figure out how to program to have the machine teach itself.

[spork]

I remember the junior high days of doing an algebra problem set, giving it to the teacher at the beginning of the next class, and then sitting there for the whole class period while the teacher reviewed how to solve the problems that students had just turned in. It wasn't the most engaging, and the method assumed that all children in the class actually did the problem set. An AI system (a.k.a. adaptive learning) could very well speed up the process and permit a flipped classroom approach, as the article hints at.

But I'm thinking more along the lines of teaching and learning something like writing. I am on the verge of requiring that students run every single piece of writing they submit to me through a grammar/spelling checker or suffer the consequences. These are really really basic mistakes that students ought to have learned to avoid in elementary and high school and fundamentally indicate a lack of attention to detail and sloppy thinking. If AI can simplify my life in this regard, I'm all for it. But the end result will probably be less of a need for humans to teach something like this, at least in the way it's traditionally been taught.

[Caracal]

But I'm thinking more along the lines of teaching and learning something like writing. I am on the verge of requiring that students run every single piece of writing they submit to me through a grammar/spelling checker or suffer the consequences. These are really really basic mistakes that students ought to have learned to avoid in elementary and high school and fundamentally indicate a lack of attention to detail and sloppy thinking. If AI can simplify my life in this regard, I'm all for it. But the end result will probably be less of a need for humans to teach something like this, at least in the way it's traditionally been taught.

When I've taught college writing classes, I didn't spend any time at all teaching grammar. I don't mean I ignored it. I corrected mistakes in papers and if there were enough of them it factored into grades. When students workshopped each other's work they would note grammatical problems and we'd sometimes talk briefly about them. I didn't teach comma rules though, and really, that isn't appropriate in a college level writing course. What you're supposed to be doing is teaching arguments and the importance of revision. Students might need reminders about the need to take more care in their work, but if the problem is that they actually don't know the basic rules of grammar, there is no way that a standard college level writing course can fix that. They need something more remedial that I wouldn't be qualified to teach.

Maybe grammar programs will get better, but right now they aren't very good. I leave my grammar check on in word and while sometimes it does spot mistakes, it has a really high false positive rate. It can be a useful tool to spot a mistake, but it wouldn't do you much good if you didn't know the mistake when you saw it.

[spork]

My employer does not have remedial writing courses. It should. But the presence of such a course would conflict with the admission office's marketing strategy.

[pigou]

We don't teach algebra and geometry so students can solve real problems any more than people run on a treadmill to prepare for running away from wild animals. There's value in *learning* how to construct a proof and how to set up a problem in mathematical terms, but not in applying it for the most part.

It's inevitable that AI will offer customized educational plans to students and we should all cherish this. A teacher standing in front of 30 students is an absolutely ridiculous model of education. Inevitably, 10 will be bored because the content is moving too slowly, 10 won't have mastered the necessary prerequisites and so will be lost, and the middle 10 will be moderately stimulated. That's a pretty horrendous hit rate, especially once you consider that it's not the same 10 people in every subject. So why bore 10 people in Math when they could be working on their English composition instead?

[polly_mer]

We don't teach algebra and geometry so students can solve real problems any more than people run on a treadmill to prepare for running away from wild animals. There's value in *learning* how to construct a proof and how to set up a problem in mathematical terms, but not in applying it for the most part.

I use the geometric proofs aspects far, far more than I use the basic algebra.  That's one way to knock down a problem to manageable by putting limits on what it possibly could be.

It's inevitable that AI will offer customized educational plans to students and we should all cherish this. A teacher standing in front of 30 students is an absolutely ridiculous model of education. Inevitably, 10 will be bored because the content is moving too slowly, 10 won't have mastered the necessary prerequisites and so will be lost, and the middle 10 will be moderately stimulated. That's a pretty horrendous hit rate, especially once you consider that it's not the same 10 people in every subject. So why bore 10 people in Math when they could be working on their English composition instead?

YES!  I have pretty bad memories of elementary and middle school because the class was almost never at the right level for me, but I had to be there all day everyday waiting for others to catch up or deployed to help others catch up--others who didn't want to catch up and resented me even more for being tasked with trying to make them stay on task and learn something one-on-one instead of quietly zoning out in the back.

[Caracal]

YES!  I have pretty bad memories of elementary and middle school because the class was almost never at the right level for me, but I had to be there all day everyday waiting for others to catch up or deployed to help others catch up--others who didn't want to catch up and resented me even more for being tasked with trying to make them stay on task and learn something one-on-one instead of quietly zoning out in the back.

Education, as John Dewey argued, isn't really about learning particular things. It is mostly about learning how to be a member of society.

[Antiphon1]

Education, as John Dewey argued, isn't really about learning particular things. It is mostly about learning how to be a member of society.

This. 

Plus, where is the discussion about when the cost of planned obsolescence in software and technology overtakes salaries?  We've need to address the marketing campaign that conveniently conflates programming with human interaction (teaching).  Programming is not inherently superior to human interaction just because we can program out human flaws.  It is weaker because those flaws are paths to new solutions.  Building systems on weak assumptions undermines the systems.  Which may be the basis of the planned obsolescence.  Technology is a manufactured tool.  Who benefits from the manufacture of these tools?  When we answer these questions, we'll have a better idea of how the technology can best be implemented. 

[spork]

Meanwhile, at Georgia Tech: http://www.omscs.gatech.edu/explore-oms-cs.

[Antiphon1]

Meanwhile, at Georgia Tech: http://www.omscs.gatech.edu/explore-oms-cs.

Low cost revenue stream.

[pigou]

Education, as John Dewey argued, isn't really about learning particular things. It is mostly about learning how to be a member of society.

This.

That may have been true in 1950 when you learned everything you needed at the assembly line. But that was also a time when you were almost always better off staying at home than going to a hospital, because of how unscientific medicine was.

Today's information environment is massively more complex and navigating the world (or being part of society) requires an ability to navigate and work with data. That's going to be even more true 20 years from now.

Plus, where is the discussion about when the cost of planned obsolescence in software and technology overtakes salaries?  We've need to address the marketing campaign that conveniently conflates programming with human interaction (teaching).  Programming is not inherently superior to human interaction just because we can program out human flaws.  It is weaker because those flaws are paths to new solutions.  Building systems on weak assumptions undermines the systems.  Which may be the basis of the planned obsolescence.  Technology is a manufactured tool.  Who benefits from the manufacture of these tools?  When we answer these questions, we'll have a better idea of how the technology can best be implemented.

I'm not sure I agree with this. Suppose that the best doctors and teachers are vastly better than the best algorithm. Does that mean we shouldn't use algorithms? Well, the vast majority of people won't have the best doctor and teacher: they'll have an average teacher. Half the population will have someone who is below the median in ability. So instantly you could make life better for 50% of the population if your algorithm does as well as average.

We see the same thing around self-driving cars: yeah, they make dumb mistakes on roads and they aren't as good as a human driver (yet). But they're vastly superior to a drunk driver and a texting driver. And accidents and fatalities happen when a human driver isn't at their peak performance. So the proper benchmark isn't zero error, or as good as a perfectly attentive driver. The proper benchmark is as reliable as a human driver on average. Anything better than that saves lives. In fact, even that's too conservative: self-driving cars could change the ownership model, which means we'd not need (at least) two parking spots for every car. Imagine what can be done with all the real estate that's currently dedicated to housing cars for when they're not needed (which is almost all day!). We could have more parks and transit would become more affordable.

I'm not sure I understand the argument around planned obsolescence. But one things that's great with algorithms is that you can do a software update over night and you've improved service for billions of people. Training, say, teachers is really hard and expensive and it takes many years... and you have to train each teacher from scratch. That's resources we could spend much more usefully in figuring out *what* works. The fact that teachers are exchanging (non-validated) materials on Instagram and Facebook is plenty of evidence that we're not doing it right at this time. That's a lot of wasted effort.

[spork]

Meanwhile, at Georgia Tech: http://www.omscs.gatech.edu/explore-oms-cs.

Low cost revenue stream.

And offering the equivalent degree at a more affordable price to people all over the world is bad how? (Not sure I understand your argument.)