Log In | Subscribe | | |

My boss keeps saying he wants an algorithm. Is it soup or a sandwich?

Classification: 

Dear Uncle Bert and Aunty Gert

I'm a young intern with a political party. I'm a woman and my boss is a man. Honestly, I'm a bit confused by some of the language used and I'm not sure if some of it is innuendo.

Recently, my boss has taken to demanding that I give him an algorithm. I don't know if he means he wants me to pick up his lunch, find an esoteric document or to get under the desk and try not to bang my head.

Can you help me?

Out to Lunch

That all seems like a tortuous process. Can you streamline it?

Of course you can: you can short circuit all that rigmarole by asking a colleague a broad question: what kind of bread does the boss like? If the person you ask knows the answer, then you have data to put into your algorithm. So the next stage of "AI" is to check a database and see if there is an answer to the question there. In this example, your colleague is like the database. But tomorrow, when he says the same thing - you will already know the answer so you don't need to check the database: you've learned and stored the data in your own database, a.k.a. memory. This is what AI people like to refer to as a "learning machine."

They also like to think that the machine is able to build its own algorithms - and so it can, up to a point, just as you can build your own because after a week of going out to buy the boss's lunch, you'll not need to go through all the questions: you'll just know that sometimes he likes ham and cheese toasted, sometimes he likes tuna with olives and capers and sometimes he likes roast beef and mustard. Importantly, you know that when you arrive in the shop, the decision making is different: now it's you looking at the available options, deciding which one looks nicest today and ordering that.

So, you've taken the initial instruction, gathered data, processed it and acted upon it but all within parameters set by your boss and by the data.

That is all there is to it. Yes, there can be multiple algorithms, yes they can each contain many options but it all comes down to the simple question: "if it is this, then the following happens; if it isn't this, then something else happens." Every choice is binary, it's a yes or no answer. The end result looks like something clever has happened but it hasn't.

Gert says :

AI people hate simplicity: they are driven by making machines do something they can say is "cool" and the more complicated the software and the more entertaining the result, the more "cool" it is. Or they want to prove that their expensive and incomprehensible "AI" is needed to do what are actually very simple data-matching tasks. And they all get very, very upset when someone says that while building the sequences of instructions is technical and, if done properly, clever, the program itself, and the computer than runs it, aren't intelligent: the apparent intelligence is nothing more than a transfer of intelligence from the techies who built it. And here's why: you, yes, I'm talking to you can design it: you've been doing it all your life.

Here's why.

You were taught if-then-else loops at school, probably when you were about ten years old and if you were anything like me, they were daunting. But what you didn't realise was that you'd been using a life-saving algorithm in your daily life from the moment your parents took you out for a walk. Yet, in fact, algorithms are the very first things we learn, even before we learn to speak.

"Am I hungry? Yes, I cry; no, I go to sleep."

"Is my nappy dirty? Yes, I cry; no, I go to sleep."

Does my tummy hurt? Yes, I cry; does it help when mummy or daddy picks me up and pats my back? Yes, I burp, no, I cry."

Later, when your parents took you out for a walk and you came to the kerb, they taught you a little saying to help keep you safe "Look right, look left, look right again and if all's clear, cross the road." *

Write to Bert and Gert: