June 26, 2019 at 09:23Parenting vs Machine Learning: compares a young mother
- Transfer
Elena Gerasimova, head of Data Science in Netology, translated an article by Lai Queffelec on how the processes of raising children and teaching AI are similar.
If you, like me, raised children and at the same time taught the algorithm, then most likely you compared these two processes. And even if you are not fond of artificial intelligence, but know a lot about children, welcome to the wonderful world of machine education ... oops, machine learning.
When writing this article, not a single child was injured. It’s just that I, like any parent, spend many hours observing how my child learns the world and am surprised to see his behavioral patterns. Just as a data scientist does, observing the results of train / test samples (training data selection for training the algorithm / the result of the algorithm on the new data - approx. ed. )
This is a quote from Jim Stern , author of Artificial Intelligence for Marketing: Practical Applications, from a lecture on machine learning - not about children (I love children!).
The essence of machine learning is to actually teach the machine to perform a specific task, just like parents dream of teaching children how to collect dirty laundry and put it in the washing machine while mom and dad are resting on the couch (admit it, tried?).
However, the main difference is that when a child is asked to wash clothes, he already knows what clothes look like; he knows how to walk, grab, pull and fold - he learned these actions thanks to other events in his young life.
So where is the key that ultimately opens up the possibility of being lazy on the couch while the laundry is magically washed? Context. We give children examples: we show how to perform each step and thank when they do everything right - because we love them.
To a large extent, machine learning is the same, except that the “virtual child” still has the experience of a newborn with the capabilities of a grown up baby. Therefore, you will have to start explaining from scratch: these five pieces, like sausages, stick out at the end of a long, sausage-like stick, sticks are fingers, hand and palm. Only then should you show how to use them to perform the necessary actions - grab and pull. The data set that you give the machine is all that is needed to get started, but also everything that exists in the world for it. What she does not yet possess is ...
Usually people distinguish between men and women with success. Liam, my son, also does a good job of this - and I did not give him a large set of tagged data at the input. I did not sit with him in the park and did not point out people saying “man, man, woman, man, woman” - because, honestly, that would be strange. Yes and no need. The luxury of common sense possessed by a child, and which it already uses when it first encounters a new concept, is inaccessible to the machine.
By common sense, I mean:
The ability to make the right decisions and make the right assumptions based on logical thinking and experience - Wiktionary
Of course, when a child decides to jump headlong into the ground from a height, we reasonably doubt that he has common sense. Nevertheless, it exists and allows children to learn from their entire experience. Moreover, no one is clearly broadcasting to them how to learn to distinguish between men and women.
Explaining the topic of AI to non-Datacientists, I like to use an analogy. The child needs only a little observation, a few examples and a couple of corrections in order to learn to say “Mr.” or “Mrs.”. And to train a car to do the same, you need to give it thousands of images. The lack of common sense is probably the number one reason why cars are not yet ready to take over the world.
Liam does strange things, for example, eating a hot dog, holding it by the ends and biting it in the middle. The standard reaction is to tell him: “Liam! They don’t do that! ” But then I hold back and think that the decision “out of the box” is not the best I can give him. Although when he tries to hold a spoon with his nostrils, he really has to set the boundaries of acceptable behavior at the table.
This is the great similarity between babies and cars - they are free from social norms and prejudices (or Bayes - from the English bias). And that is the difference between parents and the date of Scientists. Kids need to be given a set of values and social norms from which they will build their experience. “Good borders,” let's call them that. As a data scientist, you are most likely playing the opposite role. The machine must be free from your own norms and prejudices. Bias or addiction in algorithms is very dangerous.
Everyone loves gossip and hype. For example, Amazon's AI recruiter is a sexist ( Amazon's recruiting AI is sexist ), or a FaceApp “enhancement” filter is a racist ( FaceApp's “hot” filter is racist) This is a good way to explain to people unrelated to data science that the role of a scientist and the date of a scientist comes down to a large extent to preventing bias and creating the most ethical algorithm possible.
Image source xkcd
Correlation does not imply a causal relationship. And Nicolas Cage is not a monster that provokes drowning in the pool ( read about it at your leisure ). Nevertheless, I learned that this rule is not obvious to the child.
Not so long ago on vacation with my whole family, I informed the child that I was going to eat, and began to lay food on a plate. It was at that moment that he burst into tears, shouting at me (“Don’t eat, mom !!!”), clapping my hand and knocking the plug out of my hand.
When I managed to pick up the jaw from the floor, I tried to understand if my child was a monster who did not want his mother to eat and only two days later, laying him to bed, she realized where it all came from.
Our daily routine was as follows: I returned from work, fed the baby, bathed, put to bed, and then finally ate. As a result, each time, laying the baby in bed and reading a book to him, I ended the evening with the words: "Mom is going to eat." And after that I left the child alone for the next 10-12 hours of sleep. Thanks to this correlation, his mind created a causal relationship: "if my mother was going to eat, she would soon leave me alone." Oh ...
My mother’s task here is to change this pattern so that my son doesn’t learn the connection between food and separation. At Data Scientist, if a machine chooses the wrong symptom or cause, the main task is to acknowledge the error.
Back to Amazon’s unsuccessful use of AI as a recruiting tool. The 10-year sample of data they used to evaluate candidates chose men more readily because “most of the resumes were historically obtained from men, reflecting male dominance in the technology industry.”
And now Amazon’s AI seems to say: “Hey guys, most of the applicants are men, so you have to hire men, and if a woman sent a resume, I throw it away because it's an anomaly.”
No, AI. It just makes you a sexist. And it is here that kids have an advantage (and adults, let's be optimistic): it's never too late to learn not to be a sexist.
There is not a single parent who calls raising children exclusively pleasant and easy (and if someone still says so, he will brazenly lie). Each parent must constantly ask himself the question "what does the baby learn?" and adapt to its ever-evolving neural network.
To some extent, Data Scientists have the same responsibility.
When hiring or studying at Data Scientist, one cannot expect that all work will be associated only with programming. This is equivalent to the expectation that a happy adult can be raised from a child by training him, like a dog, with orders to “sit” and “turn over” all his childhood. From experience, this works until the child is 6 months old - and as soon as he has learned how to roll over, it is time to teach him human things.
I'll just leave a grinning emoticon here. If you are a parent, you already know everything.
If you, like me, raised children and at the same time taught the algorithm, then most likely you compared these two processes. And even if you are not fond of artificial intelligence, but know a lot about children, welcome to the wonderful world of machine education ... oops, machine learning.
When writing this article, not a single child was injured. It’s just that I, like any parent, spend many hours observing how my child learns the world and am surprised to see his behavioral patterns. Just as a data scientist does, observing the results of train / test samples (training data selection for training the algorithm / the result of the algorithm on the new data - approx. ed. )
“At first he's stupid like a cork”
This is a quote from Jim Stern , author of Artificial Intelligence for Marketing: Practical Applications, from a lecture on machine learning - not about children (I love children!).
The essence of machine learning is to actually teach the machine to perform a specific task, just like parents dream of teaching children how to collect dirty laundry and put it in the washing machine while mom and dad are resting on the couch (admit it, tried?).
However, the main difference is that when a child is asked to wash clothes, he already knows what clothes look like; he knows how to walk, grab, pull and fold - he learned these actions thanks to other events in his young life.
So where is the key that ultimately opens up the possibility of being lazy on the couch while the laundry is magically washed? Context. We give children examples: we show how to perform each step and thank when they do everything right - because we love them.
To a large extent, machine learning is the same, except that the “virtual child” still has the experience of a newborn with the capabilities of a grown up baby. Therefore, you will have to start explaining from scratch: these five pieces, like sausages, stick out at the end of a long, sausage-like stick, sticks are fingers, hand and palm. Only then should you show how to use them to perform the necessary actions - grab and pull. The data set that you give the machine is all that is needed to get started, but also everything that exists in the world for it. What she does not yet possess is ...
… common sense
Usually people distinguish between men and women with success. Liam, my son, also does a good job of this - and I did not give him a large set of tagged data at the input. I did not sit with him in the park and did not point out people saying “man, man, woman, man, woman” - because, honestly, that would be strange. Yes and no need. The luxury of common sense possessed by a child, and which it already uses when it first encounters a new concept, is inaccessible to the machine.
By common sense, I mean:
The ability to make the right decisions and make the right assumptions based on logical thinking and experience - Wiktionary
Of course, when a child decides to jump headlong into the ground from a height, we reasonably doubt that he has common sense. Nevertheless, it exists and allows children to learn from their entire experience. Moreover, no one is clearly broadcasting to them how to learn to distinguish between men and women.
Explaining the topic of AI to non-Datacientists, I like to use an analogy. The child needs only a little observation, a few examples and a couple of corrections in order to learn to say “Mr.” or “Mrs.”. And to train a car to do the same, you need to give it thousands of images. The lack of common sense is probably the number one reason why cars are not yet ready to take over the world.
Norms and Oddities
Liam does strange things, for example, eating a hot dog, holding it by the ends and biting it in the middle. The standard reaction is to tell him: “Liam! They don’t do that! ” But then I hold back and think that the decision “out of the box” is not the best I can give him. Although when he tries to hold a spoon with his nostrils, he really has to set the boundaries of acceptable behavior at the table.
This is the great similarity between babies and cars - they are free from social norms and prejudices (or Bayes - from the English bias). And that is the difference between parents and the date of Scientists. Kids need to be given a set of values and social norms from which they will build their experience. “Good borders,” let's call them that. As a data scientist, you are most likely playing the opposite role. The machine must be free from your own norms and prejudices. Bias or addiction in algorithms is very dangerous.
Everyone loves gossip and hype. For example, Amazon's AI recruiter is a sexist ( Amazon's recruiting AI is sexist ), or a FaceApp “enhancement” filter is a racist ( FaceApp's “hot” filter is racist) This is a good way to explain to people unrelated to data science that the role of a scientist and the date of a scientist comes down to a large extent to preventing bias and creating the most ethical algorithm possible.
Correlation and causation
Image source xkcd
Correlation does not imply a causal relationship. And Nicolas Cage is not a monster that provokes drowning in the pool ( read about it at your leisure ). Nevertheless, I learned that this rule is not obvious to the child.
Not so long ago on vacation with my whole family, I informed the child that I was going to eat, and began to lay food on a plate. It was at that moment that he burst into tears, shouting at me (“Don’t eat, mom !!!”), clapping my hand and knocking the plug out of my hand.
When I managed to pick up the jaw from the floor, I tried to understand if my child was a monster who did not want his mother to eat and only two days later, laying him to bed, she realized where it all came from.
Our daily routine was as follows: I returned from work, fed the baby, bathed, put to bed, and then finally ate. As a result, each time, laying the baby in bed and reading a book to him, I ended the evening with the words: "Mom is going to eat." And after that I left the child alone for the next 10-12 hours of sleep. Thanks to this correlation, his mind created a causal relationship: "if my mother was going to eat, she would soon leave me alone." Oh ...
My mother’s task here is to change this pattern so that my son doesn’t learn the connection between food and separation. At Data Scientist, if a machine chooses the wrong symptom or cause, the main task is to acknowledge the error.
Back to Amazon’s unsuccessful use of AI as a recruiting tool. The 10-year sample of data they used to evaluate candidates chose men more readily because “most of the resumes were historically obtained from men, reflecting male dominance in the technology industry.”
And now Amazon’s AI seems to say: “Hey guys, most of the applicants are men, so you have to hire men, and if a woman sent a resume, I throw it away because it's an anomaly.”
No, AI. It just makes you a sexist. And it is here that kids have an advantage (and adults, let's be optimistic): it's never too late to learn not to be a sexist.
Both parenthood and data science are about people.
There is not a single parent who calls raising children exclusively pleasant and easy (and if someone still says so, he will brazenly lie). Each parent must constantly ask himself the question "what does the baby learn?" and adapt to its ever-evolving neural network.
To some extent, Data Scientists have the same responsibility.
When hiring or studying at Data Scientist, one cannot expect that all work will be associated only with programming. This is equivalent to the expectation that a happy adult can be raised from a child by training him, like a dog, with orders to “sit” and “turn over” all his childhood. From experience, this works until the child is 6 months old - and as soon as he has learned how to roll over, it is time to teach him human things.
So what is easier - raising a child or raising a car?
I'll just leave a grinning emoticon here. If you are a parent, you already know everything.
From the editors
- Online course " Machine Learning "
- "Profession the Data Scientist »
- Online course « Big the Data »