AI - A primer for educators

By Cathy DeanJanuary 4, 2019Share  

Introduction

You can't watch or read your news (likely online) without seeing some mention of how AI is changing the world. This article is meant as a primer for educators, mostly to explain the basics given the amount of disinformation in today's media (and the marketing of some tools claiming to use AI for really cool stuff)

Basic definition and types

Artificial Intelligence is the capability of a machine to imitate the intelligent human behavior

The term itself is a misnomer. It is neither artificial nor is it intelligence. It's not artificial because it is man-made, and it is not intelligent because it is usually designed for one specific task.

Basic AI

AI can be as simple as a filtering task ('If the envelope is small put it in my mailbox, otherwise put it under the desk') or playing a game (checkers, chess or the famous Go used in AI news). We all played and lost a game of chess against the computer. Other than some specialized heuristics, most chess engines build a tree of possibilities to find the best outcome based on its score. This is the traditional definition of AI. Now notice how there is little intelligence here. It's a pre-backed tool that reacts to pre-defined set of inputs (you won't be playing chess with a checkers game engine).

Machine Learning

More complex AI is called Machine Learning, and examples include facial recognition and spam filtering. You use a lot of data to build a specialized model for the purpose you are trying to achieve. These models are typically Neural Networks and they can get pretty complex with all sorts of fancy modeling. But the essence is the same. You define a model and 'train' it with a large subset of your data, and then you 'test' it with another subset of your data to make sure the model works well.

Deep Learning

In all cases today, AI engines are good at one specific task only. The next level of AI, called Deep Learning, is still a dream. We are not there yet. It is the ability of computers to start having a better ability to learn from their experiences. In other words, being high on the cognitive skills.

The takeaway is that we can now analyze much more complex problems, with much more data and do so with a standardized set of tools

And this is what it boils down to, we can now do some really cool things with all the data that we are collecting in our growing databases.

The lazy person's tool

Machine Learning, which is what most people refer to when they say 'AI', has evolved thanks to two major developments:

  • Processing power has gotten cheaper
  • Data storage has gotten cheaper

Today, a good programmer can build a good AI model in 10 mins

And herein lies the danger. Instead of actually understanding a problem, we can now throw a lot of cheap processing power at it. You get some data, use one of the open-source AI tools, play with it a bit, and you have a 'black box' that you can use to (try to) make good predictions.

Even with the best black box, if we do not have a good understanding of the inner workings of a model, we can't really use it

And now comes the fun part. Do we trust this model to yield accurate predictions? Would you make important decisions for your school or district based on what an AI model tells you? This is the perfect segway to our next section.

AI in education

There are two main branches that are worthy of mention when it comes to using AI in classrooms and beyond.

Predictive performance

This is the sector of AI we are heavily investing in as a company, and frankly, the most useful and promising.

Knowing students' prior performance, can we easily and accurately predict future performance?

Note the use of the word easily here. The actual end-user of this data (teacher, principal, leader or even parent) is not an AI expert. It has to be easy enough that they would use it and get meaningful results from it.

The second challenge is accuracy which depends on a whole slew of conditions: are we measuring and predicting in the same conditions; for example: Are we using the performance of A middle-schoolers to predict that of special needs high-school students?

Here is a harder one:

Do we have enough data (both in quantity and scope) to build a model useful enough for decision-making by our educational leaders?

Stay tuned, we believe we have the answer to that one (hint: upcoming blog post)

Personalized learning

If you have been around long enough, you have seen these tools come and go. Even big publishers had at some point invested in them. Many of the content that was based on Personalized Learning (PL) is now considered sub-par by many educators (some were even sold at discount to foreign publishers - as you are likely aware).

The challenge? Well, back to basics. The developers of these tools were expecting students to be motivated enough to use them (almost) unsupervised to literally learn by themselves. As experienced educators know, teachers have a much bigger role than imparting knowledge, especially for younger students. PL will stick around for a while, but is not likely (in our opinion) to flourish beyond being great helpers, because in its essence, PL takes away from the social aspect of learning.

Conclusion

We hope you enjoyed this article and feel you are in a better position to react to the hype surrounding AI. New AI startups will continue to launch with great fanfare, and most will disappear silently. But overall, we will have better tools to craft our trade. We are working on it.

References

Cathy Dean
Cathy Elizabeth Dean is an educator with over 10 years’ experience as a teacher and freelance writer. Her academic background includes an MBA and three degrees in education. Other significant roles include several years as a family law paralegal and volunteer work as a CASA advocate for foster children in crisis. Her passions include traveling with her children, journaling, and genealogy

Other interesting articles

Get the latest in thought leadership straight in your inbox

We only email every couple of weeks at the most