Module 5: Artificial intelligence in business and society

Professor Patrick Winston gives a detailed example of another type of AI, discussing some of his recent research on building AI systems that learn in much the same way that humans often do: by understanding stories.

• Rule based systems, 
• searching many possibilities
• linear regression, probabilistic reasoning
• next
• 5 to 10 years… What’s possible with data and deep nets
• algorithms are free… data is the asset
• AI Forth Wave
• Massive, free computing
• Excited people
• Emerging round table
• Accumulated progress
• Better questions 
• What makes us humans different than other specieis, past and present?
• Human have been around 200k years
• about 70k began to advance
• We can take two disparate concepts, and make a new concept without impacting the original two concepts.
• “Merge operation” give us an inner-language…. only we can tell stories
• AI
• Artificial Perception
• Story understanding
• Recipe following: 
• reasoning
• Strong Story Hypotheses:
• 
• 
• 
• self awareness is a suitcase term
• `System will be able to explain what they do and why?
• machines and programs will tell their own stories.
• 
• smarter application
• applications that can explain themselves
• applications that understand us
• better understanding of ourselves and each other, and that will take us to a new level.

In Module 4, you learned about robotics, which was defined as the automation of physical tasks. However, some people are also using the term “robot” to refer to systems that automate certain kinds of purely information-processing tasks. For instance, the financial world refers to so-called “robo-advisers” that help clients manage their investments, sometimes with a human adviser in the loop and sometimes not. 

MIT Professor Andrew Lo discusses some of his research on AI and investment management. He uses the example of index funds to illustrate how algorithms can be applied to the financial world. He introduces the idea of “precision indexes,” which (akin to “personalized medicine” that is specific to a given individual) are automated portfolios that take an individual’s personalized criteria into account to make decisions. Professor Lo notes that organizations are currently missing the opportunity to model actual human behavior, rather than modeling assumptions about what human behavior should be in the investment world. He closes by elaborating on the notion of “bounded rationality.”

• index funds… assets held in proportion to their market capitalization
• People tried to do equal balances, but it was difficult to manage
• Three new criteria for index funds
• Transparent
• Investable
• Systematic
• If you look at a spectrum from index to hedge, what if you could take a full spectrum approach…”Precision Indexes” E.g. the Shawn Harris 500, based on my particulars… The hardware and software exists today to do this, what we don’t have yet is the algorithms… See “Personal indexes” paper.
• We are missing the ability to model actual behavior, not just modeled behaviors… artificial stupidity…artificial humanity… learned common sense…
• Bounded Rationality
• We don’t know what the optimal solution is, we develop rules of thumbs that are good enough.
• 4 Themes:
• Evolution models of behavior
• surveyor of investor risk preference
• heuristics and algorithms to automate systems
• learn from big data to learn from actual behaviors

The Future of Work

AI and robots are set to play a big role in the future workforce as collaboration between people and computers increases. Although the media seems to relish reporting that robots will replace human workers, it is more likely that people’s jobs will change and evolve, so that people work alongside AI and focus their energies on the tasks they do best.

O-ring principle as a collection of tasks that need to be done together to successfully accomplish a main task. If some of the tasks involved can be automated, the economic value of the human inputs for the other tasks that can’t be done by machines will increase. O-ring Harvard Economist Michael Cramer… As you improve the reliability of other components, other items become more important…. Humans becoming the  O-rings.

Never Enough Principle “Insatiability” As we gain more wealth, as tech expands, we think of more things to do.  Invention is the mother of necessity.

Issue will be how wealth is used. Think Saudi Arabia vs Norway. With respects to the use

Are we making progress in areas that are not making significant productivity gains.

“We all know more than we can tell.”  we have gotten past this now.. The degree of uncertainty is truly unknown.

Humans are good on small data, based on our models of the world. We can make inferences based on disparate data.

The key challenges for executives, will be:

(1) shifting the training of employees from a focus on prediction-related skills to judgment-related ones;

(2) assessing the rate and direction of the adoption of AI technologies in order to properly time the shifting of workforce training (not too early, yet not too late); and

(3) developing management processes that build the most effective teams of judgment-focused humans and prediction-focused AI agents.

Professor Daniela Rus discusses the impact that robots will have on the workforce. She uses an example of applying an any-time optimal algorithm to match the taxi supply and demands in New York City, which reduces the number of taxis needed. She examines whether this could result in taxi drivers losing jobs. Professor Rus also explores the potential benefits of autonomous vehicles on mobility and quality of life. She discusses AI’s impact on fields such as healthcare, law, and education before talking about the current limitations of putting AI to work.

• NYC 14,000 taxis, MIT algorithm says only 3,000 taxis are required with a capacity of 4 passengers, satisfying 98% of demand within a 2.8min waiting time and a trip delay of 3.5min.
• Level 4 autonomy is here. autonomy in some environments… Level 5 has a way to go…
• Think of the walking stick being replaced.
• Machines are better medical/legal/teachers predictors, still need human judgment and emotional connections 
• Machine learning has potential applications in so many fields, and medicine is a great example. Machines today can read more radiology scans in one day than a radiologist will see in a lifetime. So, a new AI-based approach was tasked with classifying radiology scans of lymph nodes as cancer or not cancer. The machine had 7.5% error, as compared to the 3.5% error of the human. But working together with a human, the machine and the human together achieved 0.5% error, which is a significant improvement over the state of the art.
• 
• Gaps/limitations in AI in breadth and depth perception, reasoning, creativity, thinking….
• no universal tools
• Crunching data does not translate in to knowledge.
• complex calculations doe not produce autonomy.
• 99.99% is exponentially harder than 90% correct.
• Perception and action
• tasks  with physical contact….
• 

Professor Rus talks about jobs in terms of the tasks they entail. She sees a future partnership between people and machines in which each performs the elements of the job to which they are best suited: machines doing what’s easiest for machines and people focusing on the strategic tasks. She discusses two points of concern: productivity and job quality or wages. Lastly, she emphasizes lifelong learning and reinforces the idea of collaboration between people and computers.

• 
• There will  be a focus on tasks, not jobs.
• 
• 
• must be a lifelong learner…

Professor Malone asks Professor Frank Levy about the implications of AI on employment and the future of work.  

• Politics needs to be a part of the story
• More people will be knocked out of mid-skill jobs
• Physical issues at the bottom, non-repetitive work at the top.

• Writing for MIT Sloan Management Review, H. James Wilson, Paul R. Daugherty, and Nicola Morini-Bianzino describe the jobs that AI will create and divide them into three new categories: trainers, explainers, and sustainers. Humans in these roles will work alongside machines, ensuring that machines are working in an effective and ethical manner.
• Read about which aspects of various jobs could be automated and which are less able to be automated, and gain further insight into the graph that Professor Rus used in Video 2 to illustrate automation across different activities in different sectors.
• Will a robot take your job? In January 2017, McKinsey Global Institute published a report estimating that by 2055 (give or take 20 years), around 50% of today’s work tasks could be automated. In this interactive graphic, you can input a job title or industry to find its automation potential.
• Ravin Jesuthasan and John Boudreau, in the Harvard Business Review, provide a four-step approach for thinking about how automation will affect job design. 
• Read about the productivity benefits of automation along with its impact on various industries and implications for policymakers. (Access the report by clicking on the download link.)
• Have a look at five management strategies for getting the most from AI.

Over the longer term, a task-based view of work will be needed to make the best use of AI, to understand which tasks can be automated and which ones are better suited for people to do. New jobs will be created that are still unimagined. Institutions and society – the education system in particular – have a role to play to unlock the full potential of both people and machines in the future.

General ethical concerns surrounding AI

MIT Professor Iyad Rahwan highlights general ethical concerns about AI and explores why organizations should care about AI ethics. He describes how to balance the benefits and risks of AI, and he explains how people have been addressing these problems by using “a human in the loop.” He describes the regulation of human behavior and then examines the challenges involved with regulating the behavior of machines. He concludes by discussing preconditions to promote public trust in machines. 

• Have discussed… technical aspects of AI, business value, strategic value, future of work
• Ethics: moral principles that govern behavior
• AI benefits:
• Better recommendations
• Safer Cars
• Better medical diagnosis
• Plus+
• AI Risk
• filter bubbles
• fake news
• unfair matching
• Need to put a “human in the loop”
• AI = prediction human = judgment?
• Society in the loop
• human in the loop, with a social contract.
• Regulatory forces, more than just the law…
• Law
• Norms
• Market
• Architecture
• We have safety standards, liability laws, and consumer expectations
• Regulating AI is different
• not passive, 
• have autonomy, 
• have intentionality 
• can adapt and learn
• cant certify at design time, as they will adapt and learn as they interact with the real world.
• will need to enforce the law, as they act
• Agency vs Experience
• machines don’t care about out norms
• How do you assign intentiality
• Need to understand emerging norms. What do we expect from AI, need to help adoption, while not over regulating.

Professor Rahwan provides a case study of the ethics of autonomous vehicles. He poses a scenario: what if an autonomous vehicle’s brakes become inoperable, and the car is heading toward a group of pedestrians who would be killed if the car hit them. The car has a choice to swerve and hit only one pedestrian rather than the group. Should the car swerve? Or, what if the car could swerve and avoid the pedestrians, but would thereby harm the occupant(s) in the car? At issue is that a machine would be making a moral decision. Professor Rahwan discusses approaches to this dilemma, including how different countries have started tackling such questions.

• Accidents by car: 1.2m -> 120k human to autonomous…. 90% accident are attributed to human error.
• what socially acceptable behavior
• social dilemma:
• I would not want to be sacrificed
• but everyone else should
• Our choices have externalities, we can’t be selfish
• People are less likely to purchase a car that will sacrifice them
• adaption and capacity
• theory of mind, mind perception
• bars on the front of the car. US ok, Europe no
• our new issue is that its a software decision
• open issue
• Germany created an autonomous car ethics commission
• Legal scholars
• Ethics experts
• Engineers
• Consumer protection groups
• Religious leaders
• recommendations
• avoidance of critical dilemma situation.
• should not be any discrimination…
• you can take total number of casualties in to account, not mandated
• person who does act on purpose, should not jeopardize people int he car.
• Check lout the Moral machine
• AI is a new kind of challenge, that we need to take seriously to address Law, Norms, Market, and Architectures.

2.3 Crowdsourced workers

Platforms such as Amazon’s Mechanical Turk (MTurk) and CrowdFlower, as well as vendor-managed systems like Clickworker, let companies hire contract workers to complete tasks, down to the level of microtasks that may only take a few seconds to complete. Read about the ethical issues that arise regarding the many crowdworkers whose low-paid, behind-the-scenes labor underlies many AI systems.

2.4 Biased algorithms

AI systems can excel at identifying patterns that let companies target specific customers more precisely. As you’ve seen throughout the program, this ability helps companies serve the unique needs of niche customers. But, sometimes this targeting can go awry. For example, Facebook’s algorithms enabled advertisers to reach self-described racists. Facebook’s COO, Sheryl Sandberg, publicly apologized for this “totally inappropriate” outcome and Facebook pledged to add 3,000 people to its 4,500-member team of employees to review and remove content that violates its community guidelines.

In another example, Microsoft set out to build a chatbot that could tweet like a teenager.  Microsoft announced “Tay” on March 23, 2016, describing it as “an experiment in conversational understanding” and released it on Twitter. The idea was that the more Tay engaged in conversation with people on Twitter, the smarter it would become. Unfortunately, Tay learned all too well. As people sent racist, misogynist, anti-Semitic Tweets its way, Tay started responding in a similar tone, not simply repeating back statements, but creating new ones of its own in the same unfortunate vein. At first, Microsoft deleted the offensive statements, but, within 24 hours, shut down Tay to “make some adjustments.” 

1. Four researchers in the field of AI share their views, concerns, and possible solutions for reducing and avoiding societal risks associated with AI. 
2. Read about the differing views of Elon Musk and Mark Zuckerberg on the safety of AI.
3. According to Sandra Wachter, Researcher in Data Ethics at the University of Oxford, although building systems that can detect bias is complicated, it is in principle possible and “is a responsibility that we as society should not shy away from.”
4. In 1942, science fiction writer Isaac Asimov coined his Three Laws of Robotics in his short story, “Runaround.” The three laws are outlined in Figure 1.

Humans seamlessly integrate perception, cognition and action.

AI raises serious ethical concerns, as smart machines will make decisions that may have life-and-death implications. Many AI systems also rely on low-paid workers who labor behind the scenes. Finally, AI has the potential to exacerbate and amplify the negative qualities of humans. As a result, executives considering adoption of AI systems need to reflect thoughtfully on the ethical aspects of their choices.

Understanding robotics

• Professor Daniela Rus introduces the world of robots. She touches on computation in motion and different robot forms, and highlights some potential future applications of robots. She also explores advances in autonomous driving.
• most robots are built for specific tasks.
• Automated devices that perform physical tasks in the real world.
• Ways that humans and robots can be more effective together. trusted partners. 
• 
• Computation in motion
• Sensors to actuators
• Come in many different shapes/forms
• 
• advances in HW and algorithms driving automated cars.
• Cheap Sensors and Actuators
• Body and brain
• Body: sensing
• Body: Computation
• Body: Actuation
• Body: communication
• Brain: Algorithms
• 
• Professor Rus delves more deeply into the topic of autonomous vehicles. She describes the two different phases of autonomous systems and the architecture of the systems now being developed. This includes how autonomous vehicles rely on planning and control, localization, and feature vectors to get from a starting location to a destination. A feature vector is a list of numbers representing an object’s important characteristics, such as size, shape, color, and brightness. She also explains parallel autonomy: when a system corrects a driver’s errors by making small adjustments, such as turning the wheel to ensure safety.
• Two phases for autonomous vehicles
• Phase 1 – Maps the environment, map is a feature space.
• Phase 2 – Plans a path and executes from start to destination, perception
• localization
• detect obstacles
• 
• 
• 
• perception is performed by looking for features, to identify uniquely most locations in the world
• curvatures of curbs
• textures of buildings surrounding roads
• 
• From normals, you can extract out features to form curves, etc..
• 
• have a virtual bumper, to give enough time to create an alternative movement plan
• import traffic rules, but know when to break the rules per Vienna convention
• 
• IMU (Inertial measurement unit)
• Series Autonomy: either human or computer is in control. Think Google
• Parallel Autonomy: human in control, but “guardian agent” is there. A Shared Controller works between human and Series Autonomy system.
• 
• Professor Rus examines the applications of autonomous vehicles by using two examples: an autonomous vehicle trial in Singapore and the potential impact of driverless taxis in New York City. She also discusses technological and policy challenges associated with the adoption of autonomous vehicles.
• Cars and public transportation system are connected.
• Algorithm Optimal path for single vehicle, pooling. system implemented using taxi data in NYC
• 
• autonomy and ride sharing can transform in to a utility, reduce congestion, decrease pollution, improve quality of life.
• what’s hard
• congestion
• human gestures
• bad weather, poor visibility
• high speeds
• Computer vision is getting better, but perception is still lacking
• Policy is lagging.
• 
• How:
• dedicate lanes
• parallel autonomy
• automated logisitics/highways
• Professor Malone interviews Professor Rus to discuss four topics: 
• Why the robots in use today don’t walk around like robots in sci-fi movies?
• Important to think about capabilities of body, relative to task.
• legged robots are hard in the real world. Where to place the next leg. Boston Scientific.
• Robot would need extraordinary perceptions and agility of motion.
• What is easy and what is hard in robotics?
• easy
• mobility is easier than manipulation.
• will we see a flying car first, or Level 5  vehicle?
• hard
• manipulation, but soft robotics are advancing… will have greater compliance like human manipulation.
• Must be able top figure things out on their own.
• better at interacting with people
• making new robot bodies faster.
• How people and robots are likely to interact?
• Working side by side.
• Shared control of certain tasks. they should be able to override each other.
• leveraging each others best assets, being better together.
• The kinds of robots, besides self-driving vehicles, that are mostly likely to be used in business in coming years?
• that a machine and human working together will be better than individually.

Business applications of robotics

• Professor Malone discusses the three other main ways robotics are being used today: 
• in factories…
• to perform miscellaneous physical services. 
• deliver, manipulate, locate physical objects
• Sensing, Computation, and Actuation

• Factory robotics:
• Acorn Sales Company, a small manufacturing company, talk about how they have deployed a robot called Sawyer, built by Rethink Robotics, to improve productivity.
• Gained acceptance from employees to do the dull and dirty tasks of lifting and positioning material onto conveyer belts and sawing and drilling wood.
• Acorn uses Sawyer as part of its cost leadership strategy to be able to keep prices in line with those of competitors. Goals:
• Less reliant on other suppliers
• increase quality
• keep prices in line with competitors
• business continuity
• Baxter can learn from EEG sensor connected humans who provide reinforcement through thought.
• Warehouse robotics:
• Kiva’s robots have revolutionized operations at Amazon’s warehouses. To achieve efficiency gains, Amazon re-engineered its warehouse processes based on Kiva’s capabilities, rather than trying to fit the machines into existing processes.
• The use of robots in warehouses enables productivity improvements, thus supporting a company’s cost leadership strategy, particularly in times of labor shortages. Robots can lift and move heavier loads, bringing items to easy reach of human workers who then select the products to fulfill each customer order. 
• From a competitive advantage standpoint, it’s interesting to note that Amazon acquired Kiva Systems, renamed it Amazon Robotics, and then chose not to renew the contracts Kiva had with other companies, thereby confining the technology to Amazon and keeping it out of the hands of competitors.  
• Performing miscellaneous physical services:
• Savioke’s robots, Relay and Botlr, are armless robots used in hotels to transport amenities autonomously to guests’ rooms. In the next two videos, Professor Malone interviews Savioke’s CTO and Chief Robot Whisperer, Dr. Tessa Lau, to find out more about Savioke’s robots, their appeal to customers, and their implications for business strategy.
• Dr. Lau provides an overview of the role of Savioke’s Relay robot in hotels. She talks about how Relay moves around hotels to deliver items to hotel guests, and she discusses the benefits of using a robot like Relay in the hospitality industry. She also mentions how Savioke’s robots are used in other industries. Professor Malone and Dr. Lau discuss how Savioke has created robots that navigate through human spaces and the company’s focus on human-robot interaction design.
• Notes:
• deliver robots for indoor space. Relay is for hospitality. controls elevator through wifi…
• Highest turnover in the US… Robots provide consistency, at a lower price point deliver less than 5 min predictably
• FedEx uses 7 robots, repair facility, relay brings parts to them as they need them. (like Kiva)
• focused on human spaces. Robots that are safe for people. Makes Relay very aware of surroundings. 
• Human-Robot interaction design.
• Politeness, 
• Asks for help, 
• Human has empathy for robot.
• Anthropomorphizing robots
• By making him cute and approachable increase adoption, actually protects him in the real world…
• Professor Malone asks Dr. Lau to expand on the AI technology that drives the Relay robot, and they discuss the qualities that Relay has in contrast to a human employee. Dr. Lau highlights some of the limitations and challenges that Relay still faces, and she concludes by providing advice to people who are interested in using robotics.
• Notes:
• primary technology is robot navigation/path finding. How do I get from one point to another safely, within constraints and obstacle….to construct a plan….use AI to actively replan.
• biggest competitor is human labor.. but brings reliability, consistency, automation, and predictably.
• limitations/Barriers:
• How to make relay safe in some environments, (e.g. escalators, stairs, 
• limited to wheelchairs can go. ADA creates standards that robots can use.
• Still needs to be programmed to do the task. The more you can cast your problem in the eyes of repetition. Maybe processes need to change to become structed for repetition, you don’t have to think.
• Key points and strategy
• Robots that operate in the same environments as humans must have enough understanding of their surroundings and of the humans in them, so they can navigate those environments safely. This requires machine vision (like autonomous vehicles have), as well as “path finding” (the ability to get from Point A to Point B even though there may be many ways to get there). The robot must also be able to reprogram if the environment changes, such as if a person steps into its path or the robot ends up on the wrong floor.  
• One reason to give a robot a human “face” and “personality” is to encourage empathy for the robot. Just as babies have big eyes to trigger an emotional connection, a robot that can create an emotional connection will be more accepted and be perceived as trustworthy by people. Savioke designs its Relay robots with attention to making them “polite” in interactions with humans.
• A delivery robot can perform certain tasks less expensively than human labor can, supporting a cost-leadership strategy in industries such as hospitality that have high turnover requiring training of new workers. However, there are many tasks robots cannot do, and thus they cannot replace human workers entirely. A delivery robot can help to ensure consistency of service, thereby also supporting a differentiation strategy. For example, a hotel brand could advertise that a guest, staying in any of its properties, could expect room service delivery within 5 minutes, because that is the typical time it takes a delivery robot like Relay to fulfill a housekeeping request.
• Robots are currently deployed in factories and warehouses, and sometimes behind fences or in restricted areas to keep them separate from humans, for safety reasons. But as robots become increasingly able to distinguish humans from other objects, they are becoming safe enough to operate in the same physical spaces as humans, such as in hotels and restaurants. Organizations that automate manual tasks so that robots can perform them can potentially reduce errors and lower costs while improving efficiency and quality, often all at once. In many settings, the most successful outcomes have been achieved through robots and people working together, each doing the tasks that are easiest for them to do.

Module 2: Machine learning in business

Unit 1: Key features of machine learning

Machine learning as a discipline seeks to design, understand, and use computer programs that learn from experience (i.e., data), without being explicitly programmed for specific modeling, prediction, or control tasks. In his interview with Professor Malone, he suggested three requirements for using machine learning:

1. The problem can be formulated as a machine learning problem.
2. There is much relevant data available that could be used by machine learning algorithms.
3. The system has enough regularity in it that there are patterns to be learned.

Understanding machine learning

Professor Jaakkola provides a brief definition of machine learning, and discusses how it can be used to solve prediction problems. He also gives examples of the types of prediction problems that can be addressed, and touches on supervised learning and convolutional neural networks. He then highlights the recent progress made in the machine learning field, discussing deep learning architectures, with DeepMind AlphaGo used as an example.

• programmers use tools to abstract out the complexity of writing machine code
• what about creating general learning program that learn from experience, data
• 
• Tyopes of Prediction problems
• future events
• predict properties not know yet.

• formulate ml problem, by learning from examples
• Labeled data is a group of samples that have been tagged with one or more labels. Labeling typically takes a set of unlabeled data and augments each piece of that unlabeled data with meaningful tags that are informative. For example, labels might be indicate whether a photo contains a horse or a cow, which words were uttered in an audio recording, what type of action is being performed in a video, what the topic of a news article is, what the overall sentiment of a tweet is, whether the dot in an x-ray is a tumor, etc. Labels can be obtained by asking humans to make judgments about a given piece of unlabeled data (e.g., “Does this photo contain a horse or a cow?”), and are significantly more expensive to obtain than the raw unlabeled data. After obtaining a labeled dataset, machine learning models can be applied to the data so that new unlabeled data can be presented to the model and a likely label can be guessed or predicted for that piece of unlabeled data.

Supervised learning –  Supervised: given explicitly input examples and target label, that I wish to predict: … is the machine learning task of inferring a function from labeled training data.[1] The training data consist of a set of training examples. In supervised learning, each example is a pair consisting of an input object (typically a vector) and a desired output value (also called the supervisory signal). A supervised learning algorithm analyzes the training data and produces an inferred function, which can be used for mapping new examples. An optimal scenario will allow for the algorithm to correctly determine the class labels for unseen instances. This requires the learning algorithm to generalize from the training data to unseen situations in a “reasonable” way (see inductive bias).

• Typically had to program in detail what you wanted a machine to do, that is  then translated in to lower level code. But now instead, you can write a general program that included instruction to learn from experience (i.e. data).. then teach them what to do by giving them more data and direction….

Formulate by learning from examples.

• Training set: Label examples of movies I do and don’t like +1 / -1
• Translate in to feature vector, based on questions we would ask…
• Genre, famous lead, etc?  Then calculate in to a binary feature vector
• Test set: Then I go through the same process for movies that I have not seen.
• Training set give us clues, to predict the label in the test set. Then the algorithm classifies new movies presented in to like and dislike.

In computer vision

• Learn from idneitifyed featured in multiple passes.
• Layeers of transfomariont, then you get a deep model … Convulution Neural networks

• Movie recommendation
• Take descriptions and translate in to a feature vector 
• Ask questions about the movies, then calculate those answers in to a vector… What is the genre, famous lead, etc… then calculate in to a binary vector….
• Then calculate the same vector for movies you have note seen…
• In the training set you have labeled vectors; in the teat set you just have the vectors that you need to provide the label for…
• Move into a geometric form and put the vectors as a point in space, based on labels.
• problem, how to classify test set, training set gives me clues to classification of data generally.
• You can divide data in to halves of -/+ then bring back test set and classify them in to the two halves, based on the learning form the training set.
• Convolution neural network,
• Classify images in to content categories

•  progressive learning of what an Item is by looking at a fraction (small features) of the pixels then gradually taking in more and more of the image.
• Recent progress
• 
• 1- Error has gone down by 50% to 100% per year… 2012 -> 2014…
• 2 – The red represents deep learning approaches, and you can see how they have taken over computer vision
• We also see advancements in machine translations, captures semantics, etc… You are giving an example of the correct behavior, then you are trying to automate the process of finding the solutions
• Learning to act: Playing Games
• GO – Looked at game board and learned to match it to what actions you could take, instead of thinking about all of the possible actions upfront. it did this by watching human experts play. However, you could probably do better by watching a computer play. Deepmind alpha go 

Professor Jaakkola provides four main reasons for the recent advances made in machine learning, as follows: 

1. The accumulation of huge amounts of data 

2. Advances in computational power

3. The growing complexity of models
• Large models are easier to train

4. The new possibilities created by deep learning architectures
• Flexible neural “lego pieces”
• Common Vector representaiton
• 
• Allows you to take an image and generate a word, or sentence, or take a word or sentence and generate an image… Because this model has been put in to vectors that are cross referenced. Allows the transfer from one domain to another in a very simple way…
• Recurrent neural networks, takes known information and new information, to derive a new vector representation. and see how it functions for the classification task you will want to use it for.
• 
• Google translate works in this way…
• Takes a lot of data to get examples of behavior. 
• Amount of data: What data i have prior to trying to solve a problem… What additional data do I need.
• 
• Types of machine learning:
• 
• Unsupervised learning is a type of machine learning algorithm used to draw inferences from datasets consisting of input data without labeled responses. The most common unsupervised learning method is cluster analysis, which is used for exploratory data analysis to find hidden patterns or grouping in data.
• Semi-supervised learning is a class of supervised learning tasks and techniques that also make use of unlabeled data for training – typically a small amount of labeled data with a large amount of unlabeled data. Semi-supervised learning falls between unsupervised learning (without any labeled training data) and supervised learning (with completely labeled training data). Many machine-learning researchers have found that unlabeled data, when used in conjunction with a small amount of labeled data, can produce considerable improvement in learning accuracy.
• Active learning is a special case of semi-supervised machine learning in which a learning algorithm is able to interactively query the user (or some other information source) to obtain the desired outputs at new data points.[1] [2] In statistics literature it is sometimes also called optimal experimental design. [3] There are situations in which unlabeled data is abundant but manually labeling is expensive. In such a scenario, learning algorithms can actively query the user/teacher for labels. This type of iterative supervised learning is called active learning.
• Transfer learning or inductive transfer is a research problem in machine learning that focuses on storing knowledge gained while solving one problem and applying it to a different but related problem.[1] For example, knowledge gained while learning to recognize cars could apply when trying to recognize trucks. This area of research bears some relation to the long history of psychological literature on transfer of learning, although formal ties between the two fields are limited.
• Interpretable modeling
• deep learning is pretty opaque, so we need to make them interpretable… highlight the rationale to the solution/predictions.  This will help us guide them.

Professor Jaakkola explains, at a high level, the difference between 

Shallow learning and 

• say in image prediction (recognition) it would be taking all pixels and bouncing off of a simple classification model to see if you can determine what it is… based on all available pixels. Pixels -> Cat or human
• “C” or “D”

Deep learning architecture 

• layers of processing, features, combination of features  breakdown of data in to discrete parts to understand them, then bringing together to provide the prediction. Pixels -> arm -> hands -> fingers -> eyes -> edges  -> wiskers -> etc etc etc etc ->humam or cat
• “Dog” or “Car”

and talks about 

neural networks and 

convolutional neural networks. 

• detectors for each area of an image… Spatial analysis

The two professors discuss what is currently both easy and difficult to do with machine learning algorithms. Professor Jaakkola poses three questions to help you decide whether machine learning is possible as an application to solve a problem. 

• Can you formulate in to a machine learning problem? 
• You can use supervised learning to predict many things. Provide training data and expected results, and this can be used to find answers in target data sets.
• Do I already have tools available to address that formulation?
• Classifying images
• Interpretating natural language
• Business decision (stock picking)
• etc.
• Is it likely to work well. as a solution to the problem?
• With formulation and tools, still unable to develop high confidence results…
• Example: Planetary motion can be chaotic…

They talk about the differences between 

supervised, 

• labeled data with target outcomes.

unsupervised, and 

• Trying to understand based on observation/how things work.. what are the regular structures… Then when you get feedback, it increasing the learning.

reinforcement learning, 

• used to learn to action, based on changes to the sate of the world, you need to act differently…

as well as the amount of training data or examples that machines need to make predictions.

• Supervised: Do you have historical data, actions and outcomes? featurize data.. attributes being tracked.
• You can use unsupervised if you dont have a lot of data, just to start.
• If you have feature representation exists, then very little data one example will allow fpor a prediction, but that is based on a lot of data to train the system
• Starting from scratch – Tabula rasa refers to the epistemological idea that individuals are born without built-in mental content and that therefore all knowledge comes from experience or perception.
• if it’s easy (high correlation) could be few 100 examples…
• if it’s hard (low correlation) could be much more… 
• exactly how much data is based on the complexity of the formulation problem, and can be assessed by a data scientist.E

machine learning systems currently can’t properly explain how they came up with an answer and whether this will change in the future. 

• Active area of research. Interpretability. How and why a decision was made.
• No commercially available solutions at this time. Possibly next year or two. Need will drive this… Think medical fields.

Lastly, he explains how formulation is key to understanding machine learning.

• Formulation is the key. That is… In this case, this is what I want…and here are illustration of what I want. to then run through ML.
• What are the inputs, and what are the outputs you are looking for… 
• Recognize machine learning problems all over the place…

Three requirements for using machine learning:

1. The problem can be formulated as a machine learning problem. Formulation is the key. That is… In this case, this is what I want…and here are illustration of what I want. to then run through ML.
2. There is much relevant data available that could be used by machine learning algorithms.
3. The system has enough regularity in it that there are patterns to be learned.

Unit 2: Business applications of machine learning [± 2 hour 30 minutes]

An executive’s guide to machine learning:\

In an article from McKinsey & Company, machine learning is explored from an executive’s perspective. The article discusses how organizations are using machine learning for insights, the importance of strategy in getting started with machine learning, and the role of senior executives in leading such initiatives.

Sensing

• Perceiving large amounts of data from sensors in the world, and learning what to recognize what there.
• Recognizing movement, sounds, temps, light, vibrations, faces, retina, finger-prints, scenes for autonomous cars…
• Key points and strategy
• Applications like Shoegazer could be used to reduce marketing costs (customers point the app to a pair of shoes and click to buy a pair for themselves) or to support the development of new features for shoes (customers use the app to show features they like). This kind of machine learning can be used not only for images, but also for sounds (for example, Shazam is a mobile phone app that can identify songs)

Image analysis

• Key points and strategy
• People play a role in improving the accuracy of the scene analysis by identifying the objects that the machines have flagged as unknown, and then feed those images into the training set for the machine, so that future versions become smarter. Mobileye’s customers use the company’s AI technology to differentiate their products.

Predicting

• What will happen in the future.financial fraud, disease, mechanical failure, crop yields.
• can customize these models by firm by customer base to get very customized based on the situation.
• Key points and strategy
• As Professor Lo explained, machine learning algorithms examine different parameters from those which traditional credit-scoring models examine. The insights gained from the machine learning techniques provide banks with a new lens that enables banks to predict consumer delinquency with a higher, finer-grained accuracy. Using this type of machine learning application, banks can pursue a cost reduction strategy that reduces their losses from non-payment.
• As this example shows, AXA uses machine learning to predict which customers are most likely to cause accidents that would cost AXA US$10,000 or more. Such predictions support a focus-based strategy, because they enable AXA to write policies only for lower-risk customers.
• PayPal uses machine learning to support a low-cost strategy of being efficient in customer service while avoiding fraud and customer inconvenience. Specifically, PayPal uses machine learning to determine whether a visitor to the site is a trustworthy customer. If the user seems suspicious, the system will ask for additional verification. The combination of neural networks, vast quantities of data, and deep learning have greatly improved fraud detection, but the machines cannot do it alone; people need to decide which data is relevant for the machine to use.

Personalizing product offerings

• Key points and strategy
• Pursuing a strategy of differentiation, Netflix combines machine learning with human curation to tailor its offerings precisely to the tastes of each individual customer.
• Like Netflix, Stitch Fix uses machine learning to pursue its strategy of differentiation, by personalizing its product offerings for each customer. Furthermore, human stylists and the company’s algorithms work hand in hand. The algorithms augment the human stylists’ productivity by doing tedious tasks such as matching client measurements to different brands and products. The stylists meanwhile read the personal notes that customers have sent and analyze their Pinterest boards to determine each customer’s nuances of style and taste. Stitch Fix’s approach illustrates three lessons about how to combine human expertise with AI systems. 
1. It’s important to keep humans in the business-process loop; machines can’t do it alone. 
2. Companies can use machines to supercharge the productivity and  effectiveness of workers in unprecedented ways. And 
3. Various machine-learning techniques should be combined to effectively identify insights and foster innovation.

Improving product performance using better predictions

• Prof. Randall Davis… Digital Cognition Technologies (DTCclock)
• Analyze the product and the process, to screen people for cognitive problems.
• Benefits
• Reliability – test always done the same way from person to person.
• Detailed and informative measure – accuracy of screener
• Early indications of cognitive problems 
• You can do the same in other aspect of business. Know it before it breaks  
• Good sensors, 
• good human interpreters of sensors, and 
• good machine learning tools
• Not just about the technology, it’s an interplay between human and computers
• Humans: Development of initial system, and then dealing with exceptiopns.
• Analyze pen-strokes for classification, then take in geometric and temporal data for assessment, looking for infirm and normal.
•  sketch understanding
• recognize most indicative features of data gathered.
• Buyer considerations for AI applications
• Does it work, show me tests, and performance, show me failures.
• Do you have the right kind of expertise in-house to develop, improve, and maintain?
• Keep the no magic principal…
• Key points and strategy
• As Professors Malone and Davis discussed, the machine learning aspect of the DCTclock can be generalized to many different applications of a screening test to provide early indication of a potential problem. In the case of the DCTclock, the machine looks for patterns that indicate cognitive impairment. The machine was trained using an expert’s knowledge (a neuroscientist) about the features that are early indicators of potential problems. The machine learned to find the patterns that indicate the early onset of impairment. Human and computer work together in that the machine can suggest which patients a clinician may want to watch or follow up with, and the clinician then provides the diagnosis and treatment therapy. The same process applies to the early screening of any kind of problem, such as a factory machine or a jet engine. Computers can alert people to an engine that may need repair, for example. The people then use their expertise to pinpoint the cause of the upcoming problem and take preventative action. Systems like these can support a differentiation strategy by allowing for more accurate predictions about future problems.
• Professor Alex Pentland from MIT talks about honest signals and how they can be used to interpret human intention and emotion. He talks about the “second language” that people use, consisting of body language and signaling behavior, as a result of some of our basic neural processes. Professor Pentland uses the example of Cogito’s software, which has been used to analyze conversations between call center agents and clients in real time. Computers and people work together, as the machine learning system interprets whether the conversation is going well or not, and then provides feedback to the agent to adjust their behavior for a more engaging and effective conversation. 
• Listening to the non-linguistic aspect of a conversation to make all center reps more effective.
• Think about the signals we ca not see/hear, as we focus just on words.
• Cogito…slow down, stop talking, redirect…
• increase customer engagement
• more effective communication
• less conflict
• lower call center turnover
• Key points and strategy
• Professor Pentland’s research has found that “unspoken” language – the nonlinguistic signaling of interest, attention, dominance, and so on – accounts for 40–50% of the outcome of a conversation. People tend to focus consciously on words, so it is more difficult for them to tap into these signals, but computers can help them out. Professor Pentland used supervised machine learning in creating Cogito to “listen” to these unspoken features and predict how the conversation between a customer service representative and a customer is going. Cogito’s software supports a differentiation strategy by facilitating higher-quality customer service interactions by letting representatives know if the customer is paying attention or is getting angry, for example.

Module Artifacts:

MIT AI M2U1 Video 1 Transcript.pdfMIT AI M2U1 Video 2 Transcript.pdfMIT AI M2U1 Video 3 Transcript.pdfMIT AI M2U1 Video 4 Transcript.pdfMIT AI M2U2 Casebook Video 1 Transcript.pdfMIT AI M2U2 Casebook Video 2 Transcript.pdfMIT AI M2U2 Casebook Video 3 Transcript.pdfMIT AI M2U2 Casebook Video 4 Transcript.pdfMIT AI_M2 U2 Casebook.pdf

Unit 1: Outcomes of this program

• It not about smarter machines, it’s about smarter organizations…
• Collective Intelligence: Groups of individuals working together, in ways that seem intelligent…
• In tech think of Google, or Wikipedia, Technology enabled collective intelligence
• Question: How can people and computers be connected, so that collectively they act more intelligently than any person or group has done before.
• Connecting people together in new ways so they can act more intelligent as a group
• Connecting people to computers that have more Artificial intelligence
• After the course, I will know…
• more about AI in business
• opportunities for AI in busienss
• ideas and concrete plan as to how MY org can use AI
• Will understand generally about AI in the world.
• If you are afraid of AI, the hope is that you will not be at the end.

Unit 2: An overview of AI: definition, history, and current state

• Defining AI
• The term “artificial intelligence” is not easy to define. The word “artificial” is more straightforward, meaning something that doesn’t occur naturally. By contrast, “intelligence” has been defined in many ways. One good definition, by the psychologist Howard Gardner, focuses on problem-solving: “Intelligence is the ability to solve problems, or to create products, that are valued within one or more cultural settings” (Gardner 1983). Informally, people sometimes use the term “artificial intelligence” to mean only those activities that are hard for computers to do (like understanding English) as opposed to simpler activities computers routinely do today (like accounting). 
• An important distinction in the field of AI is between “narrow AI” and “general AI”. Narrow AI is defined as “a machine-based system designed to address a specific problem (such as playing Go or chess)” (Kiron 2017). By contrast, general AI refers to machines with the ability to solve many different types of problems on their own, like humans can. To date, all applications of AI are examples of narrow AI. Although general AI is currently a hot research topic, it is still likely decades away from true realization.
• For the purposes of this program, 
• Professor Malone’s intuitive definition of AI is that it is: “AI is…machines acting in ways that seem intelligent”. 
• Professor Patrick Winston’s more formal definition is: AI is about the architectures that deploy methods enabled by constraints exposed by representations that support models of thinking, perception, and action. And of course, it’s not just about doing, it’s also about learning to do.
• Professor Patrick Winston on the history of AI
• Professor Malone’s intuitive definition of AI is that it is “machines acting in ways that seem intelligent”.
• Computation isn’t just changing somethings, it’s changing everything
• “Founders” in the field of artificial intelligence
• Alan Turing, 1950 paper.. “Turing test”.. 5 min computer of person… trying to deal with objections of humans, vs just a test of intelligence.  – We can do this…
• Marvin Minsky and the suitcase of words for AI… “It’s a suitcase term.”  Minsky’s paper “Steps towards artificial intelligence 1961” – What to do to get there…
• AI’s 1st Wave… 1960’s
• James Slaegal, MIT work in symbolic expressions. All about Problem Reduction, or breaking problems in to simpler, and simpler problems…
• If you get the representation right, you’re also most done.
• Professor Patrick Winston “AI is about models of thiking, preception, and action…”
• model … behaves like the real things…
• representations .. support models… set of conventions for describing situations.
• representations expose constraints
• methods to deals with constraints
• Architectures for the methods.
• AI’s 2nd wave – mid 1970
• Ed Shirtliffe, Stanford , MYCIN – Diagnose infectious blood disease, Rules based expert systems 
• More History…
• Summer of 1956 researchers got together at Dartmouth to start laying foundation in researching and developing artificial intelligent human being.
1. Design goals of artificial intelligence
1. Reasoning
2. Knowledge Representation  – John McCarty invented Lisp, language to helpd define knowledge
3. Planning (including navigation)
4. Natural language processing
5. Perception. How doe we feel, hear, smell, things in the world
6. Generalized Intelligence. (including emotional intelligence, creativity, moral reasoning, intuition, etc.)
• Many boom/bust cycles – AI winters
• 1960’s  – Translation – 60 Russian sentences in to English. 701 translator – punch card demo – no semantics…context idioms…meaning was lost.. Funding killed for a decade
• 
• 1970’s – Micro-worlds understanding language in smaller contexts. Eliza – talk therapy…
• Pick up a blue block
• put blue bock on top of red block…
• 
• 1980’s – expert systems – 
• 
• The breakthrough…
• Deep learning. McCullough and Pitts, 1940’s modeling computer on the way the brain works… 
• Effectively let’s write code that mirrors the neuron to neuron system in the brain , using weights
• 2012 – Andrew Ing used 10m YouTube videos Google brought data and scale… It found cats, they recognized 16% of other objects in the stills. They did not have to tell it what to look for. it found it. THis is a Data Up approach
• why7 now…
• 
• How does it work
• Define the number of layers, this is the deep… then give each layers some neurons… links are build between the neurons in each layers strong/week… the system will then determine what’s in the data… setting the weights on each connection,
• 
• 
• Artificial intelligence -> Machine Learning – >deep learning
• AI can make humans better…
• The future of intelligence
• 15 to 25 years to reach general AI… has been the same for the last twenty years.
• We do not understand enough about ourselves…
• Collective intelligence will most likely lead the way. Human/human computer/human, computer/computer
• The theory of multiple intelligences was developed in 1983 by Dr. Howard Gardner, professor of education at Harvard University. It suggests that the traditional notion of intelligence, based on I.Q. testing, is far too limited. Instead, Dr. Gardner proposes eight different intelligences to account for a broader range of human potential in children and adults. These intelligences are:
• Linguistic intelligence (“word smart”)
• Logical-mathematical intelligence (“number/reasoning smart”)
• Spatial intelligence (“picture smart”)
• Bodily-Kinesthetic intelligence (“body smart”)
• Musical intelligence (“music smart”)
• Interpersonal intelligence (“people smart”)
• Intrapersonal intelligence (“self smart”)
• Naturalist intelligence (“nature smart”)

Unit 3: Combining people and computers [± 1 hour 20 minutes]

• All real uses of AI involve people and computers, people are:
• Creative software
• selecting applications
• fixing problerms
• actions only a human can do
• We are not just trying to optimize computer systems, they are human-computer systems.
• Questions:
• What tasks should computer do, vs people?
• machines can remember hug amounts of info
• people interact flexible with other people
• We should not think about how computers will replace people, but how we can do things better together, and new things together.
• Think Google man made content, indexed and cataloged by Google’s algorithms… Lost some reference librarians, but gained many jobs in search and advertising. and Wikipedia bots checking content as it’s actively updated… CSAIL cyber security systems machine can find event, people can reason the event… find 3x more events..
• Four roles computers can play:
• 
• 
• 
• People and computer working together: combination was more accurate and more robust.
• 
• Crowdforge – machine and people working together, with the machine having oversight and coordination
• Cogito… listening to customer calls to assess tone and mood, to inform rep.
• How can this system improve over time?
• Ever evolving systems, learning from experince top get better and better over time.
• Cyber human learning loop
• Humans get better
• programmers improve machines
• Machines learn from experience, with various forms of machine learning
• Strategic organization need to:  think carefully about how we divide the tasks betwen humans and computers, and constatnly learn from experience.

Unit 4: How to gain strategic advantage [± 2 hours 4 minutes]

• Professor Malone begins by describing Michael Porter’s framework for understanding how organizations can gain an advantage over their competitors. Porter identified three generic strategies that companies can use:
• Cost leadership (being the low-cost producer)
• improving operations, loss prevention, using robots, 
• Differentiation (being unique on dimensions that customers value, such as quality)
• incorporating new features that could not be identified before. Look at Google assistant Watson and patient data.
• Focus (tailoring products to a narrow segment of customers)
• Uniques need of customers and individuals. Think recommendation from Netflix and Amazon.
• In general, your strategy for using AI should be consistent with your organization‘s overall strategic approach. However, in some cases, AI may make an entirely new strategy feasible for your organization.

Module Artifacts:

15 key moments in the story of artificial intelligence.pdf2017-report.pdfMIT AI M1 U1 Video Transcript.pdfMIT AI M1 U2 Casebook Video 1 Transcript.pdfMIT AI M1 U2 Casebook Video 2 Transcript.pdfMIT AI M1 U2 Casebook Video 3 Transcript.pdfMIT AI M1 U2 Casebook.pdfMIT AI M1 U3 Video 1 Transcript.pdfMIT AI M1 U3 Video 2 Transcript.pdfMIT AI M1 U4 Video Transcript.pdfMIT AI _ M1U1 Tom Malone.mp4Preparing for the Future of Artificial Intelligence.pdfWhat Managers Need to Know About Artificial Intelligence.pdfWhat can AI do right now.pdfai100report10032016fnl_singles.pdf