Machine Platform Crowd

I started reading Machine Platform Crowd last night. I haven’t gotten far enough in the book to recommend it yet, but as I thought about the book this morning on the train, I feel like there is a medium post I can wring out of the introduction for teachers.

Playing Go

The authors of the book spent the first few pages talking about disruptions not uncommon in other books that deal with technology. What I find interesting was that computer programmers have had a notoriously difficult time programming a way to win at the game of Go. The writers attribute this to the sheer number of possible moves and, thus, the sheer number of “universes” that could be created after each possible move. Even a vast number of supercomputers would not be able to compute all the moves effectively.

Within this context, the authors talked about how so much of human knowledge is actually hard for us to explain. As an example they talked about a face. How can you tell each person is who they are? How do we daily differentiate the best place to sit on a subway? How do we choose which bananas to buy at the store? We certainly all have a way of remembering these things, but it is often based on patterns or some other type of knowledge rather then definitions or structured knowledge.

Thus, when the programmers finally did beat the best living Go player, it was when they started using pattern recognition and “deep learning” to try and find patterns and, for lack of a better term, intuition for how to win at Go.

Relation to teaching

What does this have to do with teaching?

  1. The variables to teaching are not unlike Go. We are not simply putting in efficient variables that will output good grades and outcomes for our students. In even just one class, there are 30 students with different life experiences and different ways of processing information. In spite of [Insert x Ed tech product here] promise that you can use their product to reach all students, the product will ultimately fall short. This is mostly because these programs are often built in a pseudo-industrial way where students are still all working towards the same goal (albeit at different paces). Such a model, despite Ed tech promises, is not so different from what a good teacher already does through scaffolding.
  2. Intuition is, in a sense, science. While it does not follow the scientific method or may not stand up to strict theoretical tests, our intuition, especially after many years of teaching, can sometimes be quite good. Rather than deny or write off this intuition, we ought to look for patterns that show us why these intuitive ways of thinking work. This may help lead toward more effective practice.
  3. While talking about these practices and intuitions will yield results, some of it may never be totally explainable. Just like the Go players who have difficulty explaining their answers, teachers may not always be able to explain why something worked. In fact, sometimes it is something unconscious or subconscious we are doing that makes it work in the first place. The best way I’ve found to surface these is to have lots of people watch your lessons and find ways to talk about it deeply.

Don’t Write off Machines

While I spoke disparagingly of Ed Tech, I must point out that the Go player in the book also thought that a machine could never do what he did. He felt that his understanding and knowledge constituted something only a sentinet human could do. It is not some grand calculation, but rather a combination of intuition, creativity, and skill. How could a computer do this? And yet, he lost four out of five games to the computer.

I think that when Ed Tech starts following the Deep Learning model, we are going to see some crazy shit from different types of computers. We must move away from this idea popularized by Hollywood that artificial intelligence and deep learning will somehow lead to rogue sentient beings that will take over the world. The more likely scenario will be that artifical intelligence will take advantage of network algorithms to first simulate and then emulate what great teachers do. Programmers will do this by observing what great teachers to: check for understanding, reteach, get to know students, etc.

If you say, but wait, a computer can’t get to know a student, you may be right in an emotional sense, but what’s to stop programmers from doing the same pattern matching that has been done with the game of Go? Already Siri, but to much greater extent Alexa and Google, can speak in conversational English. As pattern recognition gets better, we will continue to see these systems become more “human-lite.” We may not think they are human, but many people I know have already anthropomorphsized their Alexa and it’s because she is (relatively) good at the few tasks that she can do.

I don’t think we are more than five years off from a wifi-enabled speaker that can be a proto-teacher in the room. Even if they are not doing the teaching, they will at least be inputs for a teacher. Let us do a thought experiment through what I imagine we will see in five to ten years in every class.

A Thought Experiment

“Ok, class,” the teacher states. “We have just learned the parts of the cell. Please turn on your smart speakers for your assessment at your groups.”

We zoom in one group that turn on their speaker. There are four students here: Kiara, Adam, Kadijatou, and Kim.

“Hello,” the speaker states. “How are you all doing today?”

“We’re great,” Kiara says.

“That’s good to hear, Kiara. Since you spoke first, you get the first question. What is the powerhouse of the cell?”

“Um …” Kiara thinks for a second.

“The Mitochondria,” Adam whispers to Kiara.

“Hey, Adam! Why are you answering Kiara’s question?” The speaker stated.

“Sorry, I was just trying to help.”

“Ok, but now I have to give Kiara a different question. Kiara, why is the mitochondria called the powerhouse of the cell?”

“I’m not sure.”

“Do you remember the notes from yesterday?” The speaker states.

“It said something about little things on the mitochondria.” She responds.

“Does anyone remember what those little things are called?” the speaker asks.

“Ribosomes.” Kim states.

“Very good, Kim. Do you remember what the ribosomes Do?”

“I think they create something.”

“Does anyone at the group remember what they create?”

No one at the group responds.

“Let me show you this video on ribosomes to remind you.”

The speaker shows a video on ribosomes to the students as a remediation.

After five minutes, the teacher calls the class back together.

“The speaker has graded your conversation and 75% of you showed that you know all the parts of the cell today.”

Before you say this world is far off

Remember that 20 years ago, many teachers did their gradebooks by hand. They had to calculate percentages and weights themselves. Teachers were happy (mostly) to be rid of dealing with gradebook calculations and get to the heart of teaching when they could complete their gradebook on a computer. This has also been true for PowerPoints, spreadsheets for data, and a myriad of other possibilities that computers have opened up. And with the onset of networked grading, teachers can now easily share grades with other teachers, students, and parents at a moments notice.

What I prefer about this type of AI over something like Khan Academy is it would intuitively talk to a student the way a teacher or peer would talk when trying to reflexively question students to help scaffold supports. While this model still follows the “all students learn the same thing at the same time,” there is plenty of flexibility that could be built into such a system that might make it more like tutoring and self-directed learning possible.

And before you make the same mistake as the Go masters, we must consider that there is no ACTUAL reason why a machine that has learned language would not be able to question a student or other machines in order to teach them. And as a result, the argument that a computer simply can’t do it will not suffice any longer. We may find valid philosophical arguments that show why such a system might be flawed, but we must make those arguments seriously and consider all the future universes we may live in.