Introducing AI—Martin Levins

I worry about AI being seen as a black box that is considered to be magic.

“Any sufficiently advanced technology is indistinguishable from magic”—Arthur C Clarke

So, I would introduce the idea of AI by asking what was already known about the subject and teasing out inaccuracies and misunderstandings as well as existing knowledge.

As the Digital Technologies Curriculum has not bee taught in schools for an entire K-10 cohort, I’m putting forward an approach that should work from 5–10

I think I would see confusion between different types of AI and the conflation of expert systems and machine learning.

So, I’d continue by taking image recognition as an example and asking how it might work. Perhaps use an app like Seeing AI, or PlantNet as a hook—how does it work?

Based on replies, I’d use a dichotomous key (which many in secondary would have already seen) and have the students build a key for furniture.

But, I’d have the give a weighting to their decisions and introduce the concept of probability, so an initial decision point might be “has it got 4 legs?”  to which we might attach a probability that it is a chair or a table.

Running down a simple 2 or 3 level key would give a bunch of probabilities which we could multiply together to give an “overall” probability. There would be (I hope) a lot of discussion about ascribing probabilities, and I would lead this discussion to an understanding that the probabilities would become more accurate with the number of trials. I’d link this to human learning where we improve by analysing our performance on a task and modifying our behaviour so that we become better at the task.

I want the students to get an idea of the “fuzziness” of the key’s logic

I’d then have the class look at an antibullying AI and investigate whether “you smell” is really an unkind thing to say. We’d look at context, look at the way the AI behaves and then perhaps compare “you smell good” and “you smell”

The traidiotnal way of representing an artificial nueral network is to use criss cross lines indicating the relationships between nodes and I would hope that the dichotomous key example, with its single points of decision, would illuminate such diagrams

+ There are no comments

Add yours