Raising a child vs Training a machine

Raising a child vs Training a machineThe mom’s point of view on Machine LearningLai QueffelecBlockedUnblockFollowFollowingApr 25If, like me, you’ve tried to raise a toddler at the same time as training a machine, then let’s be honest, you’ve compared these two things before.

And if you’re not into artificial intelligence but have observed toddlers before in your life, welcome to the wonderful world of machine learning aka parenting a machine.

Before we get to the details, a quick disclaimer: no toddlers were hurt during the writing of this article.

Like any parent, I just spend many, many hours watching my baby discover the world and scratch my head at his behavioral patterns like any data scientist has done before in front of their train/test results.

“At first, it’s dumb like a rock”Think I’m lying?.This quote is from Jim Stern who gave a lecture I attended in Paris about machine learning (and not about kids; I’m not that offensive towards kids).

But yes, the main idea is that machine learning consists of training a machine to complete a specific task, just like you would want to train your kid to pick up dirty laundry and put it in the washing machine while you sit back on the couch (guilty!).

The main difference here, however, is that your child already knows what a piece of clothing looks like, and if you give them directions on how to do laundry then it’s safe to assume that they already know how to walk, grab, pull, and drop; those are just a set of actions they have already assimilated through other experiences in their young lives.

So what’s the key that will ultimately unlock your ability to be lazy on the couch while your laundry is magically done for you?.Context.

You give the child examples, show them how to do each step, and because you love them you congratulate them when they do it correctly.

Well, machine learning is pretty much the same, except the “child” has the capacities of a toddler but the experiences of a newborn.

In this case, you have to start by teaching them that those five baby sausage-looking things they have at the end of that one long sausage-looking thing are fingers, a hand and an arm, and then how you use them to do things like grabbing and pulling.

The dataset you then give to that machine is everything it needs to begin working, but also everything in the world for it.

What it is not able to have yet is…… Common senseUnless your name is Sangoku (or just Goku for your Dragonball nerds), you should have always been fairly good at the “man or woman” game.

Liam (my son, aka the toddler in this experiment) is pretty good at it.

I don’t feel like I have been giving him a large set of labelled data, however.

I haven’t, like, sat down with him in a park and pointed at people spouting “man, man, woman, man, woman” because, let’s be fair, that would be creepy.

But really, it’s also unnecessary.

The machine doesn’t have the magic of common sense that the child does on top of his first experiences.

But what do I mean by common sense?Ordinary sensible understanding; one’s basic intelligence which allows for plain understanding and without which good decisions or judgments cannot be made.

 -WiktionarySometimes, when your toddler decides to jump head-first to the ground from any height you legitimately question their common sense.

However, assuming it does exist, it’s what allows them to learn from all their experiences without being expressly told that they have to learn to differentiate between men and women.

This is how I decided to introduce the topic of AI in a presentation to my fellow consultants at my agency: a toddler learning when to say “Mr” or “Mrs”.

Where the toddler will only need a bit of observation, a handful of examples and a few corrections, you will need to give the machine thousands of image to start being good at this game.

Most of them don’t work in any kind of AI-related field at all so this explanation proved to be super effective.

To me, though, the lack of common sense is probably the #1 reason why machines aren’t ready to take on the world, but my son might be (he has a huge forehead and my brother calls him Megamind… watch out).

Norms and BiasLiam does strange things, like eating a hot dog by holding the ends and biting in the middle.

My response is to tell him “Liam!.That’s not how it’s done!” but then I hold back and think about how some of the best learning is done by trying things out of the box.

(Alright, to be fair, when he tries to hold his spoon with his nostrils, I do give him some boundaries.

)And that’s actually the great thing toddlers and machines have in common.

They are exempt from social norms and biases, but that’s also where being a parent and a data scientist differs.

To the toddler you have to give them a set of values and social norms from which to build themselves.

Good biases, if you will.

As a data scientist, you pretty much have the opposite role of keeping the machine free of your own norms and biases.

Biases are bad.

Very bad.

I’ll admit, I have this gossipy side of me that likes a good bias story, like Amazon’s recruiting AI is sexist or FaceApp’s “hot” filter is racist.

They are a good way to then explain to people who aren’t in Data Science that the role of a data scientist heavily features the prevention of bias and keeping the machine as ethical as they can be.

Correlation and causationxkcd on correlation and causationCorrelation doesn’t imply causation.

No, it certainly doesn’t, especially if you’re familiar with the anecdote, Nicolas cage is not a monster who causes death by drowning.

And, as funny as it is now taking a step back, I learned the hard way that a child doesn’t know this rule either.

Not too long ago I was casually hanging at my mother’s with the whole family when I informed my child that I’m going to eat and started to pick at my food.

That’s the exact moment when he burst into tears, yelling at me (“Don’t eat, mama!!”) while slapping my hand and sending the fork straight onto my plate.

I just sat there, jaw on the floor, trying to understand if my child was being a fat-shaming jerk who doesn’t want his mother to eat in order to fit into standard sexist beauty norms.

No Lai, calm down, he is only a 2 and a half year old and isn’t influenced by social norms.

Plus he’s been successfully brainwashed by his father at a young age to always say “Mom is beautiful.

”It was only two days later while putting him to bed that I realized where it all came from.

My daily routine consisted of coming back from work, feeding the baby, washing the baby, putting the baby to bed and then finally going to eat.

As a result, every time I put him to bed and read him a story I would finish by saying “Mom is going to eat now”, leaving him in this terrible predicament of being alone in the room with 10 to 12 hours of sleep ahead of him.

With this correlation, his mind made the “Mom eating causes Mom to abandon me” causality.

Err…In this case, my job as a mom consists of changing my pattern so he unlearns this causation.

As a Data Scientist, if your machine makes the wrong causation, well, sometimes that’s our job to admit we failed.

Let’s look back on Amazon’s failed use of AI as a recruitment tool: the 10-year dataset they used to measure whether a candidate would be a good fit for the job heavily favored men because “most [resumes] came from men, a reflection of male dominance across the tech industry.

”So Amazon’s AI came back saying: “hey guys, most of the applicants are men so you should be hiring men, and if the resume says things about women then I’m throwing it out because it’s a buzzkill.

”No, AI.

That just makes you a sexist jerk.

And that’s where toddlers have the advantage (and adults, let’s be optimistic): it’s never too late to learn not to be a jerk.

Parent and Data Scientist are both a human jobMy son pretending to read a bookYou won’t find any parent who will tell you that parenting is straightforward and easy (and if someone does, that’s called a lie).

You have to constantly question yourself, what you are teaching this toddler, and adjust to their constantly evolving neural network.

Meanwhile, Data Scientists to some extent hold the same responsibility.

If you are recruiting a Data Scientist or want to become one and expect it to be only about programming, it’s like expecting a parent to raise a child like a dog and hope he becomes a stable adult while hearing only “sit” and “roll over” during their whole childhood.

From experience, it works until they are about 6 months old.

Once they roll over, it’s time to teach them human things.

So, what’s the easiest?I’m just going to leave a smirk here.

If you’re a parent, you know.

One last thing.

Liam, if you ever stumble upon this article and get to this point, I love you and I’m proud of you because it means that you can read English.

Not too bad for a French boy!Again, thank you Jeff for the careful reading, the incredible editing, and for explaining me that you don’t use the word heuristic in the US.


. More details

Leave a Reply