Took notes while reading Kazuo’s principles.

• way of interacting with others: water only flows into the valley, stay humble.
• get to know yourself, what do you value, what comes easy or hard for you and
• Work-wise have a clear purpose will protect you from boredom.
• Have a purpose and work towards it by one mini-goal at a time.
• You have never failed on a project — how did you do it? “Before it succeeds, I don’t give up.”
• equation to success = attitude ( effort ) times talent times character (mindset)
• Heaven vs hell: there is a pot of delicious noodles, you have a pair of chopsticks that’s of 1 meter long. Hell = you try to use the chopstick and eat it yourself but messed up. Heaven is when you feed it others and others feed it back to you.
• Tend to trust others before others trust you. ( like others first )
• Talent: what are you naturally good/bad at
• If a task is beyond your ability, admit it and then learn about it.
• If you take the road where everyone is on ( less risk ), you might not get there on time b/c there is traffic.
• What differentiates two people who work equally hard is how long they keep it that way
• If you are obsessively focused on one domain and goes really deep, at one point you will reach a place where you meet all the people with enthusiasm and understand the first principles of this world. It’s fun!
• The direct experience you have from life and the indirect experience you have from books build your spiritual foundation of living.
• The essence of business: fully satisfy customers needs while maximizing the profit margin.
• DAILY: you must look at the goal and progress towards the goal. It’s like you need to look at the monitor while flying a plane.
• Constantly visualize your dream and goal so that one day when you run into the opportunity, you will be able to identify it and seize it.
• Stay honest and true to the principles.
• Diligence: works so hard until the universe helps

Would like to thank Baker Nanduru for being an inspiring mentor and a supportive friend. During the discussions, I learned about Peter Drucker’s work which paves the road for modern business management. Among the management classics, picked up “Managing Oneself” — here are the trimmed notes.

Drucker used the book to answer the fundamental question, how does one find out where does he/she belong? Drucker suggested four guiding questions to give him/her the clues.

1. How does one like to work? : just as people achieve results by doing what they are good at, they also achieve results by working in ways that they best perform.

• do I prefer reading vs listening?
• do I learn by reading/doing/writing etc?
• do I work well with people or am I a loner
• do I produce results as a decision-maker or as an advisor

2.1) What’s one’s strength?

to identify one’s strength

• 1) set the objective & deadline (e.g usually 6 months out )
• 2) review the result
• 3) derive learning between the result and the objective = gaps of knowledge/skills/strength

2.2) What're one’s values? = what kind of person do I want to see in the mirror in the morning?

3) What’s one’s proposed contribution?

• what does the situation require?
• given the strength, way of working and values, how to make the greatest contribution to what needs to be done?
• what results ( goal ) have to be achieved to make a difference? A result should be 1) challenging 2) feasible 3) measurable

Observation

• One may never know someone’s true character unless they are given the opportunity to violate moral or ethical codes.
• Had the Titanic not had that famous accident, as fatal as it was, we would have kept building larger and larger ocean liners and the next disaster would have been even more tragic. So the people who perished were sacrificed for the greater good; they unarguably saved more lives than were lost.
• The fragility of every startup is necessary for the economy to be antifragile, and that’s what makes, among other things, entrepreneurship work: the fragility of individual entrepreneurs and their necessarily high failure rate. For example, the local restaurant scene is antifragile due to the fragility of the individual restaurant.
• Every time you use a coffeemaker for your morning cappuccino, you are benefiting from the fragility of the coffeemaking entrepreneur who failed. He failed in order to help put the superior merchandise on your kitchen counter.
• Roman Statesman Cato the sensor viewed “comfort” as the road to waste, as it weakens the will. The record shows that, for society, the richer we become, the harder it gets to live within our means. Abundance is harder to handle than scarcity.

The theme of the observation is that things benefit from shocks; they thrive and grow when exposed to volatility, randomness, disorder, and stressors and love adventure, risk, and uncertainty. Yet, in spite of the ubiquity of the phenomenon, there is no word for the exact opposite of fragile. Antifragility is beyond resilience or robustness. The resilient resists shocks and stays the same; the antifragile gets better.

Inspirations

• Make and learn from painful mistakes: necessity is the mother of invention, which means the absence of challenge degrades the best of the best. The avoidance of small mistakes makes large ones more severe; the longer one goes without market trauma, the worse when the commotion occurs. (e.g he who has never sinned is less reliable than he who has only sinned once. And someone who has made plenty of errors — though never the same error more than once — is more reliable than someone who has never made any)
• Make sure painful mistakes don’t kill you by forming a network: individual unit is fragile but the network becomes antifragile on the expense of dividual fragility. For example, a team is more likely to survive than an individual. However, if the team stays alive too long without ever being challenged, it runs the risk of one day being wiped out completely, more than so than the team who faces interruptions every day. So, make and learn from lots of painful mistakes.

Freedom: build lean and reliable process to foster trust, keep people happy and protect their focus of doing what they are best at.

1)what’s the ideal outcome of this process?

2) if no one can make it to the meeting, what’s the alternative to achieve the outcome? (hint: leaner process)

3) can we replace the approval mechanism with clear and widely communicated decision-making system?

Responsibility: clarify and agree with teammates on the purpose and how to get there

1) what’s the goal and where are we now?

2) next concrete opportunity to capture?

3) what’s the concrete solution?

4) when’s the release?

5) what’s the status?

A quick word on compensation; it is based on skill scarcity and strength — that is how good are you at what you do multiplies with how much is it needed by the customers.

As Machine Learning is infused into everyday product and here are the key terms to know for PMs. Thanks to GoDaddy machine learning team for putting this together.

• Features: The properties of a dataset we use as input for making predictions. For example, when trying to predict whether or not an object is a cat, we might use features of color, is_scornful, and height_in_cm; a specific object might have corresponding values of black, True, and 28. These exemplify three common types of feature:categorical, boolean, and numeric.
• Labels: The thing the model is trying to match with its output. In the preceding example, our labels would be cat and not_cat, which we would probably represent as 0 and 1.
• Model: A mechanism for using features to predict labels. The main thing to note about this definition is that it doesn’t necessarily include the thing we’re trying to do with the model predictions. We may be using the aforementioned cat predictor to decide which objects around us to pet, but the thing we’re modeling is is_cat, rather than should_pet.
• Loss: The output of a cost function which takes as its inputs the labels (i.e. the ideal model output) and what the model actually produces. The loss is a pre-specified numerical measure of how poorly the model is doing, driving the adjustments that help the model learn. This is different from the discrete accuracy we may use when talking about a model’s success; for example, if the model is 90% sure the answer is A and 10% sure it is B, even if the corresponding decision it would make is “correct,” the difference between that prediction and {A: 100%, B: 0%} still counts as loss and can cause learning. There’s no standard scale associated with loss, but lower is better. More info in this SO answer.
• Training/validation loss: These two types of loss correspond to the training data and a portion of data we set aside, called validation data. Training loss is like the fuel that causes the model to change and learn; validation loss gives us information about how well things are going without changing the model. Ideally, we can observe both decreasing as our model learns.
• k-fold cross-validation: A technique for splitting up training/validation data. When k=10, we split the data into 10 parts, we reserve one of those for validation, and we train with the other 90% of the data. Then we train the whole model again from scratch with a different 10% selected as validation data, and we repeat this process multiple times. This gives us a more stable indication of model performance. More info in the first of five sections in this tutorial.
• Underfitting: What happens when our model is too simple to make good predictions. Both training and validation loss either fail to decrease or decrease by only a small amount. The standard image associated with underfitting is trying to fit a straight line to a curved plot.
• Overfitting: What happens when our model is too complex; the standard image associated with overfitting is a high-degree polynomial zig-zagging its way through a plot without capturing the overall trend. As a model overfits, training loss decreases, but testing loss does not follow suit: it may decrease by a substantially smaller amount, stay the same, or increase. The worst-case scenario is when the model “memorizes the training set.” At this point, it makes perfect predictions for all the training data, but these predictions are not generalizable to the testing data, and there is nothing left for the model to learn.
• Regularization: The process by which we curb overfitting. This involves either restricting the model’s complexity or penalizing it in some way. Continuing with the example of fitting a polynomial to a plot, we could require that the polynomial have no terms beyond degree 2, or we could allow higher-degree terms but incorporate their use into our cost function/loss to discourage the model from using them. Here’s a good intro to regularization.
• Hyperparameters: Essentially, the knobs you turn when trying to get a model to perform well. As the model trains, it’s already automatically changing the value of many parameters, so we use “hyperparameters” to refer to the higher-level variables we directly modify before training starts. Tuning hyperparameters can be challenging, and while some degree of automation is possible, it’s often prohibitively expensive to comb through this space.
• Neural network: A common type of ML model, and the one used for the workshop activity and nearly all of ShopperML. In a nutshell, the idea is to multiply a vector of your training data by one or more randomly initialized matrices, examine the output vector, and repeatedly adjust the matrix values to make the output closer to the labels. This is a huge topic, but if you have a few minutes, a former GoDaddy intern and soon-to-be full-time employee made this awesome interactive NN trainer. Or if you want to take a deep dive that goes far beyond the scope of this workshop, Andrew Ng’s Coursera course and Michael Nielsen’s free online textbook are excellent resources.
• Hidden layer: An intermediate transformation of input data existing inside your neural network. This is a big part of what makes NNs so powerful, and the inscrutability of these hidden layers is what sometimes causes complaints about the opaque nature of NNs. This 16-second youtube video shows you a visualization of the hidden layer for a NN as it trains to recognize handwritten digits. The 100 hidden layer neurons start out as random noise, and then as the model trains, they start to resemble number-like shapes. A handwritten digit input will activate these 100 neurons to varying degrees, and this will in turn determine what digit prediction the model makes.

Basic NN hyperparameters

The additional definitions below are the names of the hyperparameters you’ll be changing as you train your model, named as they appear in ShopperML. We definitely don’t expect folks who are new to ML to learn these prior to the workshop, but it would be great if you could take a quick look now, and then you’ll probably refer to the definitions and typical values again during the workshop.

• num_passes: How many epochs (complete times through the training data) your model trains for. This can be used as a regularization tool, in that you can try to stop training before the model begins to overfit, but with other forms of regularization included, you should be able to train for a long time, until both training and testing loss have plateaued. We typically use values between 5 and 250. Try a smaller number like 5 or 10 if you are checking to see if training is generally working, or a larger number like 50 or 100 if you care more about optimal performance or are working with a network with lots of complexity and regularization.
• batch_size: Tells you how many pieces of training data you pass through your model at a time. This has an impact on training speed, since a larger number lets you take more advantage of matrix multiplication efficiency, especially if you are using a GPU. However, this may also impact how well your model trains, since it corresponds to step size when trying to find your way to a minimum of cost. We typically use powers of 2 between 64 and 1024.
• learning_rate: When it comes time for your model to assess its loss and make adjustments, this hyperparameter will govern how aggressive these changes are. We typically use values between 0.00001 and 0.01. Too high of a learning rate often causes underfitting or an exploding loss function; too low, and learning progresses too slowly.
• num_layers: How many hidden layers your neural network has. For shopperML models, 1 or 2 are probably the only options you need to try. Deeper neural networks are very powerful in contexts with highly hierarchical features (e.g. edges composing shapes, composing pointy ears, tails, whiskers, composing cats), but in this context, there is no reason to believe that deeper networks will be especially successful.
• hidden_size: How big your hidden layer(s) are. This directly impacts how “smart” your model is: it can be reduced to regularize when your model is overfitting or increased when it is underfitting. In ShopperML, this can be an integer or a list of integers. Typical values for hidden layer size vary by context, but 256, 512, and 1024 are all standard choices.

More advanced NN hyperparameters

The hyperparameters in the previous section are sufficient to get great results for the workshop activity. But if you want to use ShopperML for the modeling contest, or you want to try tweaking some of the fancier knobs, or you’re just curious, read on.

• activation: After every layer of a neural network beyond the input, we need to apply a non-linear function to each individual floating-point number in that layer (for the curious, here’s why). The sigmoid function is one option, and as you can see from this graph, it squashes very large x down to just under 1, it squashes very small x << 0 up to just above -1, and it’s close to linear between -1 and 1. This is a typical choice for machine learning, especially in educational contexts, since it has some convenient mathematical properties, but it can also cause very slow learning. The other option in shopperML is ‘relu’. ReLU activation has its own problems, but as the graph in this article indicates, it often outperforms sigmoid.
• optimizer: The mechanism by which, along with learning rate, your model uses training loss to make changes. The common introductory NN option is ‘SGD’, stochastic gradient descent. Using ‘Adam’ will probably give you better results, since it makes model tuning depend less on the learning rate by making automatic adjustments as the model trains. Here’s more info on the Adam optimizer.
• nn_dropout: When set to True, the model uses dropout, a regularization tool that causes your network to randomly shut off a portion of its connections when training. This makes it harder for your model to overfit, and it forces the model to develop some internal redundancy. If your model is overfitting, setting this to True is a good idea.
• nn_dropout_rate What portion of connections are shut off during training. 0.5 is a good starting point. Note that some other NN libraries present this as the portion of connections that are retained.
• batch_norm: Models have a harder time learning when input features or internal layers have different numerical distributions, i.e. really big or really small numbers can mess up your model. Batch normalization is a way to deal with this: your layers are shifted and scaled using learned parameters to avoid problematic outliers. This can speed up learning, though it also makes your model more complex.
• res_net: Whether the NN is a ResNet. Briefly, this is a network with “skip connections,” in which earlier layers connect to later ones, and it allows for better deep learning. This shouldn’t have much impact on our shallower networks, and it can increase training time, so you probably want to leave it set to False.
• weight_decay: Sets the amount by which weights decay as the model trains. This is directly derived from L2 regularization, a process by which the squares of weights contribute to the trainer’s cost function. By encouraging multiple smaller weights capturing relationships rather than a single large one, you’re steering the model towards redunancy and away from overfitting. Good values to try are 0.01 and 0.001.

Thanks to Gail Giacobbe, am exposed to Dan Olsen’s work on Lean Product Playbook: How to Innovate with Minimum Viable Products and Rapid Customer Feedback. The book is available here. It presents a clear framework.

Nuggets:

The process to achieve product-market fit

The builds the intuition for the lean product process. Do have the question of whether we start with underserved needs or target customer since customer portraits are fundamentally humans with needs to different degree. For instance, everyone has the need of transporting from one place to another but with different demand on the time needed to access the vehicle will result in “uber customer” and “customer who owns a private car”.

Market ( problem space ) vs Product (solution space )

【why, not start with the solution space 】here is an example from the book “when NASA was preparing to send astronauts into space, they knew that ballpoint pens would not work because they rely on gravity in order for the ink to flow. One of NASA’s contractors, Fisher Pen Company, decided to pursue a research and development program to create a pen that would work in the zero gravity of space. After spending $1 million of his own money, the company’s president, Paul Fisher, invented the Space Pen in 1965: a wonderful piece of technology that works great in zero gravity.”

【why start with the problem space】

“Faced with the same challenge, the Russian space agency equipped their astronauts with pencils. You can actually buy a “Russian space pen” (which is just a cleverly packaged red pencil). This story shows the risk of jumping into the solution space prematurely and the advantage of starting in the problem space.”

Human needs stay the same.

When new technology is developed, users/buyers segments adopt the new tech in the following way.

The top needs don’t matter if the bottom needs haven’t been met yet.

Persona Attributes:

• Name
• Representative photograph
• Quote that conveys what they most care about
• Job title
• Demographics
• Needs/goals
• Relevant motivations and attitudes
• Related tasks and behaviors
• Frustrations/pain points with current solution Level of expertise/knowledge (in the relevant domain, e.g., level of computer savvy)
• Product usage context/environment (e.g., laptop in a loud, busy office or tablet on the couch at home)
• Technology adoption life cycle segment (for your product category) Any other salient attributes

e.g

How painful is the need ( problem ), how frequent it happens

need to identify the underserved needs: needs that aren’t yet filled.

GAP = importance — satisfaction

Opportunity Score = importance + MAX ( importance (pain intensity * how frequent the pain occurs) — satisfaction , 0 )

Opportunity Value Delivered = Importance * Satisfaction

Opportunity to Add Value = Imporatnce * ( 1- satisfaction )

Opportunity to Add Value to exsiting product = Importance * ( satisfaction-before minus satisfaction-after)

e.g

This is the continued reading notes from Yuval Noah Harari work, 21 Lessons for 21 Century. Recommended by Bill Gates as one his 5 books for 2018: https://www.cnbc.com/2018/12/07/bill-gates-this-book-will-help-you-stop-worrying.html

Inspiration

1, Human run the world b/c we think in stories and facts. Fundamentally we all want the same thing, meaning, health love, accomplishment, money and power. We will believe in the stories that can promise us these.

2, to manage the networks, government emerges

2.1 true measure in government is rooted in economic development since economic Is the power allocated back to the individuals ( that’s why it’s so important)

2.2 stability = liberty + equality

2, New technology e.g AI ( infotech + biotech) disrupts things and require updated stories.

• Goal: efficiency
• Elements of technology
• Connectivity?
• Updatability?
• Those who own data own the future.

3, Technology and humanity. What separates human and machine, currently, is that human is able to originate empathy and kindness.

【Terrorism】

Terrorism: a weak party that uses fear to achieve a political goal. It’s like a fly that angers a bull to destroy a shop. Usually, a terrorist group is so weak that they can’t wage war. In most cases, overreaction to terrorism poses a far greater threat to our security than terrorists themselves. Terrorists think like theater producers instead of generals ( who inflict material damages) A terrorist is like a gambler who is holding a particularly bad hand and tries to convince his rivals to reshuffle the cards. He cannot lose anything, and he could win everything.

The ability to use violence is the entry ticket to the political game. A state is built upon the premise to prevent its citizens from human violence. The most efficient answer to terrorism might be good intelligence and clandestine action against the financial networks that feed terrorism. Media often inflates the danger of terrorism. A strong motivation for it to be regulated by the government but run by the citizens.

【War】

Never Underestimate Human Stupidity. There is no winner in war, especially in an ever inter-connected world. One potential remedy for human stupidity is a dose of humility. National, religious, and cultural tensions are made worse by the grandiose feeling that my nation, my religion, and my culture are the most important in the world — and therefore my interests should come before the interests of anyone else, or of humankind as a whole. How can we make nations, religions, and cultures a bit more realistic and modest about their true place in the world?

【Humility】

You are not the center of the world.

Throughout history, humans have created hundreds of different religions and sects. A handful of them — Christianity, Islam, Hinduism, Confucianism, and Buddhism — influenced billions of people (not always for the best).

The vast majority of creeds, such as the Bon religion, the Yoruba religion, and the Jewish religion, had a far smaller impact. Personally, I like the idea of descending not from brutal world conquerors but from insignificant people who seldom poked their noses into other people’s business. Many religions praise the value of humility but then imagine themselves to be the most important thing in the universe. They mix calls for personal meekness with blatant collective arrogance. Humans of all creeds would do well to take humility more seriously. And among all forms of humility, perhaps the most important is to have humility before (your own), God. Whenever they talk of God, humans all too often profess abject self-effacement, but then use the name of God to lord it over their brethren.

【Secularism】

What then is the secular ideal? The most important secular commitment is to the truth, which is based on observation and evidence rather than on mere faith. Secularists strive not to confuse truth with belief. If you have a very strong belief in some story, that may tell us a lot of interesting things about your psychology, about your childhood, and about your brain structure — but it does not prove that the story is true. (Often, strong beliefs are needed precisely when the story isn’t true.)

Secularism’s measure stick is “feelings” — do you feel good about doing something e.g rape doesn’t feel good on the person who is being raped, therefore it hurts people, and hence should be forbidden.

This is the deep reason secular people cherish scientific truth: not in order to satisfy their curiosity, but in order to know how best to reduce the suffering in the world. Without the guidance of scientific studies, our compassion is often blind.

The twin commitments to truth and compassion result in the commitment to equality. Suffering is suffering, no matter who experiences it. Secular people recognize their responsibilities to their nations but also their responsibilities to humanities as a whole.

We can’t search the truth without the freedom to think. Hence, secular people cherish freedom.

Lastly, secular people value responsibilities. They believe it’s up to each one of the individual to be helpful to society.

【Ignorance】we know less than we think

Liberal story: Democracy is founded on the idea that voters know the best. Capitalism believes that customer is always right and liberal education teaches students to think for themselves. Yet all these are built on the ideas that humans are rational individuals. Most human decisions are based on emotional reactions and heuristic shortcuts rather than rational analysis. A human doesn’t think an individual, we think in groups. e.g it takes a tribe to raise a child, invent a tool and cure disease.

Steven Sloman and Philip Fernbach termed “ knowledge illusion” : we think we know a lot, even though individually we know very little b.c we treat knowledge in the minds of others as if it were our own. From an evolutionary perspective, trusting the knowledge of others has worked extremely well for Homo Sapiens. Most of our views are shaped by communal think rather individual thinking — the story is the glue that forms groupthink. Hereby, folks don’t like facts and they certainly want to feel ignorant. e.g convincing Tea Party supporters with statistical sheets is fruitless.

Power is about changing reality rather than seeing what it is. e.g when you have a hammer in your hand, everything looks like a nail. Power is a black hole that twists everything around them b/c the followers will see your giant hammer and hereby say thing with an agenda either consciously or unconsciously.

Here then, if we are ignorant about the world, how can we decide on what’s right and wrong.

【Justice】

Justice requires a set of values AND cause-effect relationships. One today is far more ignorant today than the hunter-gatherer. We don’t know where our lunch came from and what pension fund is doing with the money. Those who make no effort to know can remain in blissful ignorance and those who do make an effort will find it difficult to discover the truth. How to avoid stealing when the global economic system is ceaselessly stealing on my behalf without my knowledge. The problem is that it becomes extremely complicated to grasp what we are actually doing. e.g the charming English lady financed the Atlantic slave trade by buying shares and bonds in London exchange without ever setting foot in Africa. They sweetened the 4 o’clock tea with snow-white sugar produced in hellish plantations about which they knew nothing.

For each group and subgroup faces a different maze of glass ceilings, double standards, coded insults, and institutional discrimination.

Currently human isn’t able to grasp the complete the truth and hereby make moral/justice calls. Have we entered the post-truth era?

【Post-Truth】

Some fake news lasts forever. Humans have always lived in the age of post-truth and in fact, we might just be post-truth species, whose power depends on creating and beliving fictions. We are the only mammals that can cooperate with numerous stranges because only we can invent fictional stories, spread them around and convince millions of others to believe in them.

I am aware that many people might be upset by my equating religion with fake news, but that’s exactly the point. When a thousand people believe some made-up story for one month, that’s fake news. When a billion people believe it for a thousand years, that’s a religion, and we are admonished not to call it “fake news” in order not to hurt the feelings of the faithful (or incur their wrath).

Note, however, that I am not denying the effectiveness or potential benevolence of religion. Just the opposite. For better or worse, fiction is among the most effective tools in humanity’s tool kit. It inspires people to build hospitals, schools, and bridges in addition to armies and prisons.

Truth and power can travel together only so far. Sooner or later they go their separate paths. If you want power, at some point you will have to spread fictions. If you want to know the truth about the world, at some point you will have to renounce power. You will have to admit things — for example, about the sources of your own power — that will anger allies, dishearten followers, or undermine social harmony. E.g Christian priests, Confucian mandarins or communist ideologues placed unity above truth. That’s why they are so powerful. As a species, we spend far more efforts trying to control (gain power over ) the world rather than understand it. When usually try to understand it, the rational is to better control the world. We prefer power to truth. Human suffering is often caused by belief in fiction but the suffering itself is still real.

Tips:

• if you want reliable information, pay good money for it. If you get your news for free, you might well be the product.
• make the effort to read relevant scientific literature. ( peer-reviewed articles, books published by well-known academic publishers etc)

Scientists should be far more engaged with current public debates. It’s tremendously important to let the results/findings out. We certainly need good science, but from a political perspective, a good science-fiction movie is worth far more than an article in Science and Nature

【Science fiction】

We like the idea of shaping stone knives, but we don’t like the idea of being stone knives ourselves. So the matrix variation of the old mammoth story goes something like this: “Mind imagines a robot; hand creates a robot; robot kills terrorists but also tries to control the mind; mind kills robot.” Yet this story is false. The problem is not that the mind will not be able to kill the robot. The problem is that the mind that imagined the robot in the first place was already the product of much earlier manipulations. Therefore killing the robot will not free us.

e.g Disney loses faith in the idea of “free will” and authentic self. Disney has built its empire by retelling one myth over and over. In countless Disney movies, the heroes face difficulties and dangers, but they eventually triumph by finding their authentic self and following their free choices. Inside Out brutally dismantles this myth. It adopts the latest neurobiological view of humans and takes viewers on a journey into Riley’s brain only to discover that she has no authentic self and that she never makes any free choices. Riley is, in fact, a huge robot managed by a collection of conflicting biochemical mechanisms, which the movie personifies as cute cartoon characters: the yellow and cheerful Joy, the blue and morose Sadness, the red and short-tempered Anger, and so on. By manipulating a set of buttons and levers in Headquarters while watching Riley’s every move on a huge TV screen, these characters control all of Riley’s moods, decisions, and actions.

【Resilience】how do you live in an age of bewilderment, when the older stories have collapsed and no new stories have yet emerged to replace them?

• Education: “change is the only constant. With the ever-increasing speed of changes imposed on a human being, the tip is to know your self and know what you want from life. To run fast, don’t take much baggage with you. Leave all illusions behind. They are very heavy.”
• Meaning: “life is not a story. who am I ? what should I do in life? what’s the meaning in life? routinely says things like “Their sacrifice will redeem the purity of our eternal nation,” know that you are in deep trouble. To preserve your sanity, always try to translate such hogwash into real terms: a soldier crying out in agony, a woman beaten and brutalized, a child shaking in fear.”
• Meditation: just observe. Vipassana meditators are cautioned never to embark on a search for special experiences; instead, they are encouraged to concentrate on understanding the reality of their mind, whatever this reality might be.

Bill Gates shared his book list for 2018 and Yuval’s work is on top of the list.

Yuval Noah Harari is today’s “must read” big thinker about technology and the future of Homo sapiens.

His Ted Talk also explained Human beings live in dual reality ( objective and fictional reality) which give the competitive edge (ability to collaborate ) to win.

Inspirations

Why homo sapiens runs the world today instead of chimpanzees? If we put one human and one chimpanzee on an island. It’s more likely that the chimp outlives the human. Yet if it’s 1000 humans vs 1000 chimps. It’s more likely that human will run the island. Because human is the only creature that works well in large networks — we have emotions, that’s the key to unity.

Fundamentally we all want the same thing — the story that promises the things below can unify humans.

• meaning
• health
• love (from self and others)
• accomplishment
• money
• power

Humans think in stories rather than in facts. E.g the fascist story (created and killed by Berlin), the communist story (created and killed by Moscow) created and the liberal story ( created by London) and now shaked ( e.g rise of Trump and Brexit)

The liberal story acknowledges peace and prosperity for all, both are, sequentially the foundations of the 6 things listed above. e.g Bush and Obama administration.

Yet the liberal story stumbles since 2008

• money: financial crisis disillusioned the public
• power: resistance to immigration and trade is mounting & restrict the freedom of the media
• resulted in 2016: Britain exists and the rise of trump
• people start to believe that “liberalization and globalization are a huge racket empowering a tiny elite at the expense of the masses.”

• today’s political system is built during industrial era to manage oil engineers and machines, not software, internet AI nor blockchain let alone their explosive potential

From killing mosquitos to killing thoughts — the twin revolution = biotech + infotech

Human has been good at controlling the world outside of us e.g if a mosquito bothers, one kills it. Today, the human is empowered to control the world inside of us, such as designing brains, extending lives and killing thoughts at our discretion ( thanks to the biotech and infotech revolution)

The twin revolutions are started by engineers, entrepreneurs and scientists who are hardly aware of their decision implication to the global political system and certainly don’t represent anyone.

The liberal story is about ordinary people, but how can it remain relevant to a world of cyborgs and networked algorithms?

The rise of trump signified that American people are disillusioned by the “globalization” part of the liberal story due to its harm to the domestic economy. Yet still believe in human rights, democracy, free markets, and social responsibility. For instance, China is mirroring this scenario:

The true measure of a government is to see whether people’s lives improved over time — that’s the objective reality. This is rooted in economic growth. Or one word: “jobs”

So far the liberal story has worked well over the communist, fascist and any nationalist narrative. E.g when Vietnam was asked about the nuclear weapon that will affect the globe, they turn to the liberal story for answers.

Yet no narrative is ready to rationalize the technological disruption and ecological meltdown. Hereby we have left the task to create an updated story for the world. The development of biotech and infotech will require fresh narratives.

Regardless, worrying doesn’t help much.

Work

“when you grow up, you might not have a job” because human abilities can consist of two parts: physical (competed away by machines) + cognitive (competed away by AI)

AI the confluence of infotech and biotech increasing computing power enabled the biochemical mechanisms that underpin human emotions, desires, and choices. In the last few decades research in areas such as neuroscience and behavioral economics allowed scientists to hack humans. It turns out that the idea of “free will” results not from mysterious forces but rather billions of neurons calculating probabilities within a split second. “Human intuition” is in reality “pattern recognition”

Two particularly important nonhuman abilities that AI possesses are connectivity and updatability.

• Connectivity: human is individual. AI is in an integrated network. It’s not individual human vs individual computer. Rather it’s networked human vs integrated AI networks. e.g car driven by a human isn’t communicating with another individual car 100%. However, the self-driving car is able to do so 100% as they are elements of an integrated network. As a result, a poor child in a developing country might come to enjoy far better healthcare via her smartphone than the richest person today from the most advanced urban hospital. Ultimately, we are here to protect humans, not jobs.
• In the long run, no job will remain absolutely safe from automation, even artists. e.g Yet if art is defined by human emotions, what might happen once external algorithms are able to understand and manipulate human emotions better than Shakespeare, Frida Kahlo, or Beyoncé?
• Of all arts, music is the most likely replaced art since all the inputs and outputs are electronic waves. e.g “if your boyfriend eventually dumps you, the algorithm may walk you through the official five stages of grief, first helping you deny what happened by playing Bobby McFerrin’s “Don’t Worry, Be Happy,” then whipping up your anger with Alanis Morissette’s “You Oughta Know,” encouraging you to bargain with Jacques Brel’s “Ne Me Quitte Pas” and Paul Young’s “Come Back and Stay,” dropping you into the pit of depression with Adele’s “Someone Like You” and “Hello,” and finally helping you accept the situation with Gloria Gaynor’s “I Will Survive.”
• If art is really about inspiring human emotions, few musicians and artists would be able to compete with an AI. The job market of 2050 is well characterized by human-AI cooperation rather than competition. In fields ranging from policing to banking, human-plus-AI could outperform both humans and computers.
• Consequently, despite the appearance of many new human jobs, we might nevertheless witness the rise of a new useless class. We might actually get the worst of both worlds, suffering simultaneously from high unemployment and a shortage of skilled labor. Many people might share the fate not of nineteenth-century wagon drivers, who switched to driving taxis, but of nineteenth-century horses, who were increasingly pushed out of the job market altogether.

From exploitation to irrelevance:

• The government might slow down the automation process intentionally to alleviate the shock and allocate time to readjust. Something can be done doesn’t mean it must be done. E.g a drone pilot might need three years to reinvent him/herself as a virtual designer. Yet the government will need to support his/her family during this transition. For another instance, Scandinavia follows the motto” protect the workers, not the jobs”. If no such system is in place, then the people will retaliate which motivates the birth of new political systems (through wars).
• It is debatable whether to provide people with universal basic income (the capitalist paradise) or universal basic services (the communist paradise).

Liberty: big data is watching you.

Thatcher: there is no such thing as society. There is a living tapestry of men and women and the quality of our lives will depend on how much each of us is prepared to take responsibility for ourselves.

Elections are always about human feelings not about human rationality. If people are all rations-beings there is no need to give equal voting rights — there is ample evidence that some people are far more knowledgeable than others on certain subjects such as politics and economics.

“when it comes to feelings, Einstein and Dawkins are no better than anyone else. Democracy assumes that human feelings reflect a mysterious and profound “free will,” that this “free will” is the ultimate source of authority, and that while some people are more intelligent than others, all humans are equally free.”

Listen to the algorithm:

The liberal belief in the feelings and free choices of individuals is neither natural nor very ancient. For thousands of years, people believed that authority came from divine laws rather than from the human heart and that we should, therefore, sanctify the word of God rather than human liberty. Only in the last few centuries did the source of authority shift from celestial deities to flesh-and-blood humans. Soon authority might shift again — from humans to algorithms. Just as divine authority was legitimized by religious mythologies, and human authority was justified by the liberal story, so the coming technological revolution might establish the authority of Big Data algorithms while undermining the very idea of individual freedom. That our feelings are not some uniquely human spiritual quality, and they do not reflect any kind of “free will.” Rather, feelings are biochemical mechanisms that all mammals and birds use in order to quickly calculate probabilities of survival and reproduction. Feelings aren’t based on intuition, inspiration, or freedom — they are based on calculation.

Genes that help human evolve get passed on. Feelings aren’t the opposite of rationality, they are the evolutionary rationality. “Free will” will likely be exposed as a myth, and liberalism might lose its practical advantages. The authority is likely to shift from humans to computers ( the confluence of infotech and biotech) Computer will be able to calculate emotions and connect with humans. e.g computers have your information and they will recommend things to you that best suit your interests. ( Computers will know our personality type and be able to press our emotional buttons ) People won’t feel comfortable in adapting to the big data but we have no better alternatives.

As authority shifts from human to algorithms, we may no longer view the world as the playground of autonomous individuals struggling to make the right choices. Rather the universe is a flow of data and we are merely biological algorithms to make that happen. E.g every day we answer emails, tweets, and articles. We don’t see how our efforts fit into the great scheme of things or how the bits of data connect with the bits produced by billions of others humans b/c we are too busy answering these emails.

Digital Dictatorship:

We often associate emotions with compassion, love, and empathy. But during wartime, it’s associated with fear, hatred and cruelty, e.g 1968 U.S at My Lai massacre.

Should we be worried about AI to enslave humanity?

No, AI can develop intelligence, the ability to solve problems and consciousness, the ability to feel things. A plane is able to fly without the feather; AI is able to solve the problems without developing a feeling about them. For another example, an algorithm is able to solve a difficult math problem but not necessarily feel joyful about it.

To further examine the argument above — there are three possibilities we examine:

• Consciousness is somehow linked to organic biochemistry in such a way that it will never be possible to create consciousness in nonorganic systems.
• Consciousness isn’t linked to organic biochemistry but it’s linked to intelligence in a way that computer can develop them if they pass a certain threshold of intelligence.
• There are no essential links between consciousness and either organic biochemistry or high intelligence. Therefore computers might develop consciousness — but not necessarily. They could become super-intelligent while still having zero consciousness.

With our current state of knowledge, we can’t rule out any of the options above. Furthermore, humans currently have limited understandings of consciousness and will be difficult to develop as a result. The danger is that if we invest too much in developing the AI intelligence and too little on developing consciousness, as a result, it will only serve to empower the human stupidity. We are unlikely to face a robot rebellion in the coming days, but it’s likely we’ll deal with hordes of bots who know how to press our emotional buttons better than our mother does. For example, it would be tragic only to understand why the stock market falls instead of undressing the cause of suffering. It would be tragic to see the employee is so stressed out to quickly answer emails and that even eats into his lunch/dinner time that he/she isn’t able to taste the food, therefore ignoring his sensation. We hope to avoid downgraded human using upgraded computers to wreak havoc on themselves and the world. (Empathy and kindness makes us different from others)

Lastly, AI might extinguish liberty, political and economic equality, as a result, creating the most unequal ( money and power ) society, with a huge amount of wealth and power concentrated in a few elite, while most people suffer not from exploitation but irrelevance → social upheaval is a result of that. For human society, the road to extreme un-equality is something we need to avoid. Ithings don’t change, then we will have to do it the hard way — using war to change it. So either way, things gonna change. The members of the society will never be equal ( money & power) but we should avoid extreme inequality.

Equality: empathy and kindness makes us different from others

The rise of AI might eliminate the economic value and political power of most humans. At the same time, improvements in biotechnology might make it possible to translate economic inequality into biological inequality. The superrich will finally be able to buy life itself, instead of just status symbols. Humankind might split into a biological caste.

Globalization has resulted in huge inequality: today the richest 1% owns more than half of the world’s wealth.

The two processes together — bioengineering coupled with the rise of AI — might, therefore, result in the separation of humankind into a small class of superhumans and a massive underclass of useless Homo sapiens. To make an already ominous situation even worse, as the masses lose their economic importance and political power, the state might lose at least some of the incentive to invest in their health, education, and welfare. It’s very dangerous to be redundant. The future of the masses will then depend on the goodwill of a small elite. Maybe there is goodwill for a few decades. But in a time of crisis — like climate catastrophe — it would be very tempting and easy to toss the superfluous people overboard.

To prevent the concentration of wealth and political power, therefore, is to regulate the ownership of data. E.g attention merchants: Baidu, Google, Facebook — they capture our attention and sell them to advertisers to get us to buy more stuff. Furthermore, by accumulating an immense amount of data, which is worth more than the advertisement revenue. We aren’t their customers, we are their products. Furthermore, once the algorithm chooses and buys things for us, then the advertising industries will go bust. For example “Hi, Google. Based on everything you know about cars, and based on everything you know about me (including my needs, my habits, my views on global warming, and even my opinions about Middle Eastern politics), what is the best car for me?” If Google can give us a good answer to that, and if we learn by experience to trust Google’s wisdom instead of our own easily manipulated feelings, what could possibly be the use of car advertisements? Google now becomes the advertiser.

In the longer term, by bringing together enough data and enough computing power, the data giants could hack the deepest secrets of life, and then use this knowledge not just to make choices for us or manipulate us but also to re-engineer organic life and create inorganic life-forms. Selling advertisements may be necessary to sustain the giants in the short term, but tech companies often evaluate apps, products, and other companies according to the data they harvest rather than according to the money they generate.

A popular app may lack a business model and may even lose money in the short term, but as long as it sucks data, it could be worth billions.

Ordinary humans will find it very difficult to resist this process. At present, people are happy to give away their most valuable asset — their personal data — in exchange for free email services and funny cat videos. It’s a bit like African and Native American tribes who unwittingly sold entire countries to European imperialists in exchange for colorful beads and cheap trinkets.

Humans and machines are increasingly getting merged so completely humans will not be able to survive at all if they are disconnected from the network. If later, you refuse to connect, the instance company might refuse to ensure you and the company might refuse to employ you. Now the key question is: who owns the data? Is it me, government, or the company? Mandating governments to nationalize the data will probably curb the power of big corporations, but it might also result in creepy digital dictatorships. Politicians are like musicians, the instruments they rely on is the human emotional and biochemical system. They give a speech, there is a wave of fear in the country. They tweet, there is explosive hatred.

So we had better call upon our lawyers, politicians, philosophers, and even poets to turn their attention to this conundrum: how do you regulate the ownership of data? This may well be the most important political question of our era. If we cannot answer this question soon, our system might collapse.

Thanks:

Grateful for Andrew Ng’s crafting the lessons learned from leading Google and Baidu’s AI endeavor. The original writing “AI Transformation Playbook” is linked here

Inspiration:

Five steps for instilling AI DNA into a company:

1, Pilot Project Win

• within 6–12 months, marry an internal team (with deep domain expertise) and external team ( with strong AI capability) to deliver meaningful tractions
• define success quantitatively and qualitatively that delivers business, customer and strategic value.
• technically feasible

e.g Andrew Ng firstly applied AI to Google speech recognition to gain momentum, not Google advertising which affects the bottom line, to gain faith from for Google Brain team. Then applied AI to Google map as another internal customer. Finally, Andrew started the conversation with the advertising team, which affects the company bottom line.

2, Build in-house AI team — the rise of Chief AI Officer

• Build AI capability to support the whole company: execute an initial sequence of cross-functional projects. After completing these initial projects, set up repeated processes to continuously deliver a sequence of valuable AI projects.
• Develop company-wide platforms that are useful to multiple divisions/business units and are unlikely to be developed by an individual division. e.g: working with the CTO/CIO/CDO to develop unified data warehousing standards.

3, Provide broad AI training

CLO ( Chief Learning Officer) recognizes his/her job nature is to curate instead of creating content.

1, Executives and senior business leaders: (≥ 4 hours training)

Goal: Enable executives to understand what AI can do for your enterprise, begin developing AI strategy, make appropriate resource allocation decisions, and collaborate smoothly with an AI team that is supporting valuable AI projects.

Curriculum:

• Basic business understanding of AI including basic technology, data, and what AI can and cannot do.
• Understanding of AI’s impact on corporate strategy.
• Case studies on AI applications to adjacent industries or to your specific industry

2. Leaders of divisions carrying out AI projects: (≥12 hours training)

Goal: Division leaders should be able to set a direction for AI projects, allocate resources, monitor and track progress, and make corrections as needed to ensure successful project delivery.

Curriculum:

• Basic business understanding of AI including basic technology, data, and what AI can and cannot do.
• Basic technical understanding of AI, including major classes of algorithms and their requirements.
• Basic understanding of the workflow and processes of AI projects, roles and responsibilities in AI teams, and management of AI team.

3, AI Engineer Trainee ( ≥ 100 hours)

Goal: Newly trained AI engineers should be able to gather data, train AI models, and deliver specific AI projects.

Curriculum:

• Deep technical understanding of machine learning and deep learning; basic understanding of other AI tools.
• Understanding of available (open-source and other 3rd parties) tools for building AI and data systems.
• Ability to implement AI teams’ workflows and processes.
• Additionally: Ongoing education to keep up-to-date with evolving AI technology

4, Develop AI Strategy

• Build several difficult AI assets that are broadly aligned with a coherent strategy
• Leverage AI to create an advantage SPECIFIC to your industry sector
• Design strategies aligned with the “virtuous cycle of AI” positive-feedback loop

e.g: leading web search engines such as Google, Baidu, Bing and Yandex have a huge data asset showing them what links a user clicks on after different search queries.

Strategic data acquisition: Useful AI systems can be built with anywhere from 100 data points (“small data”) to 100,000,000 data points (“big data”).

Unified data warehouses: 50 different databases siloed under the control of 50 different VPs or divisions will be nearly impossible for an engineer or for AI software to get access to this data and “connect the dots.” Consider centralizing your data into one or at most a small number of data warehouses.

Recognize what data is valuable, and what is not: bringing an AI team in early during your process of data acquisition, and let them help you prioritize what types of data to acquire and save. e.g tragically CEOs over-invest in collecting low-value data, or even acquire a company for its data only to realize the target company’s many terabytes of data is not useful.

5, Develop internal and external communication

Talent/Recruitment:

• externally: employer branding helps attract the scarce AI talents. AI engineers want to work on exciting and meaningful projects. A modest effort to showcase your initial successes can go a long way.
• internally: AI has been over-hyped, there is fear, uncertainty, and doubt. Employees are also concerned about their jobs being automated. Clear internal communications both to explain AI and to address such employees’ concerns will reduce any internal reluctance to adopt AI.

Customer Relation: ensure proper product roadmap and messages are disseminated

Investor and Government Relation:

• finance: explaining a clear value creation thesis for AI in your company, describing your growing AI capabilities.
• legal: developing a credible, compelling AI story that explains the value and benefits your project can bring to industry or society is an important step in building trust and goodwill.

Andrew Ng on Building Today’s AI Company & Product was originally published in DataDrivenInvestor on Medium, where people are continuing the conversation by highlighting and responding to this story.

Deeply grateful to be introduced to Kim Scott’s work Radical Candor by Steven Aldrich, at 2018 Morehead-Cain Alumni Forum, I write to share the inspirations.

Radical Candor = build strong relationship + take on responsibility

“You need to do that in a way that does not call into question your confidence in their abilities but leaves not too much room for interpretation … and that’s a hard thing to do. I don’t mind being wrong. And I’ll admit that I’m wrong a lot. It doesn’t really matter to me too much. What matters to me is that we do the right thing.” — Steve Jobs

Build Strong Relationship = Offer Empathy + Goodwill + Increased Influence

Relationship first, business second. Besides offering empathy and goodwill, trusting the high-performing individual with a designated honor of teaching can increase his/her influence tremendously.

Introduce constructive feedback: “I am grateful for the this mentor, who had the goodwill to share with me the things I missed. I didn’t feel great about it at the moment but later really saw that as a gift. I am paying it forward. Would it be helpful if I share something with you?”

A thank-you note goes a long way. e.g Jim Ottawa is amazed to find a thank-you note still hanging onto the teammate’s wall many years after he wrote it.

Take On Responsibility = Offer Constructive Feedback

Constructive feedback = praise publicly+ criticize privately: use observations and facts ( situation → behavior → impact ) , no judgement, opinions or feelings. As a result, the manager surfaces the necessary information for people to make decision themselves rather than make the decision for them.

“At Apple, as at Google, a boss’s ability to achieve results had a lot more to do with listening and seeking to understand than it did with telling people what to do; more to do with debating than directing; more to do with pushing people to decide than with being the decider; more to do with persuading than with giving orders; more to do with learning than with knowing.” — Kim Scott

Use impromptu guidance in person: keep the “constructive feedback” talk to be 2- 3 mins and immediately after observed the problem instead of “ waiting long hours”, since the unspoken criticism will explode like a bomb.

Be conscious of culture. Everyone is watching you, but that doesn’t mean it’s all about you.

Team building = putting the right people in the right shoes, through hiring, promoting and firing

1) Understand teammate’s values, dreams and learning goals. Manager’s job is not to give purpose to the direct reports but to understand how each one derives meaning from his/her work.

• Values: “ tell me about your story” ,
• Dreams: “tell me about your dreams”
• Learning goals: “what do you need to learn in order to realize the dream?”

2) Growth management

• For the steep growth curve people, keep them challenged and figure out who to replace them once they leave
• For the average performers, find ways to move up him/her
• For the low performance individuals, fire with empathy, goodwill and not hurting his/her influence.
• *Focus on the work the person is doing not the status they achieve in the company for doing it .

Meetings: means to get results

• for one-on-one: direct report sets the agenda , manager listens and asks questions to clarify
• for staff meeting: manager sets the agenda, direct report listens and ask questions to clarify ( identify but do not make decisions)

• for big debate meetings: make it clear that it’s about debating not deciding
• for big decision meetings: push decisions into the facts, pull facts into the decision and keep the ego at bay.
• for all-hands = presentations to persuade people that the company is making good decisions + Q&As conducted so leaders can hear dissent and address it head-on.
• no-meeting time: block time off to do work
• Kanban board: priority requires visibility
• walk around: learn about small problems to prevent big ones