This is the first edition of EIM Review of Books and YouTube Videos, meant to provide you with concepts and information you can use during a job interview to help you prevail over your competition. The book “Prediction Machines” by Ajay Agrawal, Joshua Gans and Avi Goldfarb, can be purchased for $18.39 through Amazon, where 81 reviews average 4.5 out of 5.
Why is this book important to an executive in transition?
First, it is important for you to have a coherent concept of how Artificial Intelligence [AI] can impact the company that you are interviewing with and to be able to communicate that concept clearly and concisely. You may be asked how you think AI is likely to impact their company and how that might affect your specific role within the company.
Second, it is important that you have a conceptual map to understand how you can shape your expertise to be less likely to be replaced by AI, and more likely to have your value enhanced because of it. This book can give you these tools.
Synopsis of the Book
The authors state that an AI program is developed to deal with decision-making for a specific situation or complex set of situations. A lot of confusion surrounds what is meant by AI. In this book, when the authors speak of AI, they are not talking about an AI with the flexible, broad-spectrum intelligence of the human brain. Nor are they talking about super-intelligent AI that not only has the flexible, broad-spectrum intelligence of human intelligence, but that also enhances its own programming at incredible speeds.
For purposes of this book, an AI is merely a program that uses available information to make predictions about probable outcomes of specific actions that can be taken to deal with a situation. This is sometimes called “narrow AI.” An AI program may see patterns that humans do not see, and it may consider inputs of which humans are not aware. It can certainly consider much more input than a human brain and it may apply logic more dogmatically than humans. Most importantly, an AI learns from the feedback it gets from the consequences of actions taken and adjusts subsequent actions while human beings become attached to their decisions as correct, even when the feedback based on the consequences of those decisions indicates the opposite. AI has no such attachment.
An AI has the advantage of making the cost of prediction much less expensive, faster and more thorough.
The authors stress that any economists worth their salt will tell you that when something gets cheaper, we use more of it. AI makes the cost of applying arithmetic to decision-making much cheaper and faster across a large swath of industries and situations. Not only can an AI program learn faster and think more critically than humans, it can uncover solutions to problems that humans may never consider. That is why a computer is now the best “Go” player in the world . The AI that drove the decision-making in the game of “Go” came up with moves no human had ever thought of before, and so it is – or will be – with virtually every decision to be made under uncertainty. Not only do we use the new cheaper and faster capacity of AI to make decisions about things that are particularly arithmetical, such as decisions about insurance rates, but also about things that are not strictly arithmetical enterprises. Such as photography, music, product design, and communication. Eventually, any endeavor that relies on making decisions based on predicted outcomes will use AI as the predictive tool.
What is left for Humanity?
Humans will be valued for how they use the predictions of AI to make recommendations about what actions to take, but they will not be the engine that generates the predictions.
What is Prediction?
The authors state that prediction involves taking “information you have” to generate “information you don’t have.” This can be taking the last four quarters of sales to predict the next quarter of sales, or looking at pixels on an MRI image to determine if a person has cancer.
What are the implications of a plummet in the cost of prediction for society?
In all the areas where we have used human beings to make predictions, we will eventually use the faster, cheaper, more effective AI technology. We will also begin converting decisions we did not think of as involving prediction . . . into those that do involve prediction and we will use AI to solve them.
Example: how we use robots. Traditionally we have used robots under very controlled conditions, such as on a factory floor. Conditions under which robots act are controlled and certain, so there is no “if then” decision-making in these situations. However, as AIs have become more powerful, and sensors cheaper and more effective, we have begun using AI in uncontrolled conditions, such as having an autonomous vehicle drive down a highway with an almost infinite number of “if then” situations to consider. “If then” decision-making will be done by AI, not by humans.
With autonomous vehicles accidents will not be eliminated, but they will be greatly reduced. This “if then” decision-making will take place across many industries and in many applications. For example, if a customer buys a book on landscaping, AI will begin feeding them information on landscaping and marketing products and services that people who buy a book on landscaping have bought in the past. AI will use an individual clients’ decisions to buy or not to buy its recommendations as feedback to help AI improve its recommendations.
The best AI for playing “Go” did not even watch human players as they won or lost games. AI merely created millions of simulated games to review in its own programming and learned from watching which moves won or lost. Then it defeated the best human players in the world, plus all the computer software programs that had not used simulations as a major tool to learn tactics and strategies. This AI process is called “deep learning”.
AI is taking over language translation. If you have used “Google Translator” you may have noticed how much better it is at translation now than it was even one year ago . . . because every time it makes a mistake in translating one language into another it is learning, and the next time its translation is more accurate. Same for voice recognition and decisions in hiring, retention, compensation and promotions. Hordes of situations may benefit from this approach, and hordes of start-ups and substantial companies are beginning to apply “deep learning” to a specific area of decision-making. Google is working on 2,000 separate AIs, each being created to deal with a specific situation or complex of situations in order to make specific predictions.
What is Involved in Deciding?
The authors say that any situation that involves a decision can be based on these components:
- The situation
- The input
- The prediction
- The judgment
- The action
- The Feedback
- The outcome
Example: the authors first take a common situation and break it down into its components.
- A public speaker bangs her knee on the podium as she leaves the stage. Next morning, when she wakes up, her knee hurts, so she goes to the doctor. [The situation]
- The doctor asks questions and takes an x-ray of the knee. [The input]
- Based on that input the doctor decides that there is a 90% chance that the knee is bruised and a 10% chance that there is a hairline fracture. [The prediction]
- The doctor decides an MRI is not warranted by the 10% chance of a hairline fracture and recommends that the patient go home, stay off the leg as much as possible over the next week and if, after a week it still hurts, come back. [The judgment]
- The patient agrees and follows the recommendation. [The action]
- One week later the knee still hurts. [Outcome and feedback]
- The patient comes back for an MRI on the knee. [Input]
- A hairline fracture is discovered with 100% probability. [Prediction]
- The doctor advises the patient to elevate the leg and to rest while the bone heals itself, and also recommends the patient ice the affected area for 24 to 48 hours, reduce their activity and, for pain, use NSAIDs (nonsteroidal anti-inflammatory drug), such as ibuprofen, naproxen, or aspirin. [Judgment]
- The patient agrees and does so. [Action]
- One week later the pain goes away. [Outcome and feedback]
ON A PERSONAL LEVEL: I knew a person who was having a problem with blurry vision. The person went to one of the highest rated ophthalmological centers in America and, after gathering as much information as the doctors could, they concluded that there was probably no danger, and they told the patient to go home and see if the blurriness abated after a few weeks. However, unbeknownst to the patient, the doctors fed all the input they had considered into an AI program. The AI predicted that the patient had a high probability of a rare optical disease that would lead to total blindness if not dealt with immediately. The physicians called the patient back, operated immediately; the blurriness abated, and the patient did not go blind.
Why is it Important that you know this before you go into any executive level interview?
It is important that any senior-level corporate decision-maker understand the components of a decision and why AI has advantages and disadvantages compared to a human being when it comes to making predictions based on input and feedback. Remember, an AI program is a prediction machine, not an experienced expert with a strong general background and the flexibility of a human being in understanding a situation and making recommendations for actions.
- An AI can gather input faster and more thoroughly. For example, an AI can read all the research on a specific topic and every opinion of any expert in the time it takes a human to read one 10-page research paper. It can read medical journals completely and keep up on medical research much more thoroughly and faster than any doctor.
- An AI can ask questions more dogmatically and change the questions it asks and the values of the inputs it considers more effectively based on the feedback it gets from the actions that it previously recommended.
- An AI cannot use its knowledge in a generalized fashion like a human being. A small change in the way you are providing it input and feedback can confuse it. If you are not specific in what you are asking an AI to accomplish, and careful in the input you are feeding it, an AI may not give you the results you seek.
- To date, an AI cannot garner the kind of personal relationship or use the emotional sensitivity and empathy a human being might deliver in making predictions. But that may change in the future.
- The authors state that an AI is not as adept at making judgments and recommending actions as a trained and experienced human, although it is getting better at making recommendations all the time.
The authors cite their research into the recommendations made by Amazon based upon the decisions its clients have made in the past and found that people tend to buy one out of 20 of those recommendations. The magic in AI resides in the way it improves its predictions and recommendations based on feedback. So, Amazon will continue to get more accurate over time.
As an executive, if you want to stay relevant over the next 10, 20 or 30 years, you need to accept the superiority of AI at gathering and analyzing input to make predictions in narrowly defined situations and focus on refining your judgment when recommending actions. The value going forward will be in owning the initial input and feedback that informs an AI. There will also be value in the ability to make effective action recommendations based on predictions of the AI. You do not have the ability of an AI, but you have greater general knowledge and a better capacity to flex when the situation changes.
SUMMATION: Do not get confused when talking about AI to a potential employer. They are likely not talking about AI that has human-level general intelligence or super-intelligent self-programming AI. They are probably talking about the kind of narrow AI that this book describes.
As always, when asked a question, keep your answer short but, if you incorporate the way this book breaks down the decision-making process and the way it talks about an AI program’s advantages, you are likely to impress them. One approach is to ask, “What are the two or three most important decisions a person in the position I am interviewing for is likely to make on a consistent basis?” Then, select one of those decisions and describe how you would use AI as you understand it from this book synopsis in your answer. Then you are sure to impress them. “Go” is an abstract strategy board game for two players, in which the aim is to surround more territory than the opponent does. The game was invented in China. – Wikipedia