AI for Everyone
A short summary of Andrew Ng’s course
Written by: Cameron Hernandez
Artificial Intelligence (AI) is ready for the world, but is the world ready for Artificial Intelligence? We find ourselves benefiting from such technology as self-driving vehicles, online purchase suggestions, and so on, but nary a member of the general populous appears to come to the realization that AI is behind all of it. Those who do appreciate the immense value of AI find themselves stalled at the starting line, simply unprepared to begin their own AI projects. Meanwhile, others are convinced that advancements in AI only bring us one step closer to bowing to “robot overlords” – but I digress.
“AI is changing the way we work and live and this nontechnical course will teach you how to navigate the rise of AI.”
Andrew Ng – AI for Everyone – Week 1 – Introduction
Many individuals find themselves at a fork in the road, where they can either continue in their current, yet dated, methods, or forge ahead into the future. It has become quite obvious – the desire, for most, to utilize AI in various projects and businesses is present, but the ability and straightforward information are lacking. This is where Andrew Ng begins his course AI for Everyone – with the understanding that informing the general populous, especially business men and women, is imperative to the progression of this invaluable resource in our society. Andrew’s course is organized such that a general knowledge of AI can be grasped in four weeks. Each week provides a particular set of material: What is AI, Building AI Projects, Building AI in Your Company and AI and Society, respectively. Additionally, Andrew ensures that all lesson topics are reinforced by a variety of examples, further solidifying the course’s material. One might even notice the subtle visual adjustments in the lesson videos as the lessons progress – keep an eye on the painting in the background as the weeks progress, as the close one gets to finishing the course the more impressive the image! I, for one, began this course with very little knowledge in the field of AI. After taking this course, I feel far more prepared to engage in general, yet productive, conversation with friends, family and colleagues regarding this rich field of study.
Andrew begins his course by asking the question, What is AI? This discussion begins by immediately addressing the elephant in the room – Artificial General Intelligence (AGI). Contrary to popular belief, there has been no significant, or even noteworthy progression towards AGI. However, there is a common misconception that any advancement in AI represents a leap towards a world where computers think and act the same, or even superior to humans – AGI. Andrew tackles this misconception by directing the conversation to the realization that all current AI is not general, but narrow – ANI. These AI algorithms are capable of handling a single task very well, even better than humans in certain cases (think playing chess or Jeopardy), but they are far from capable of replacing the human brain or being self-aware – so humanity can take a collective sigh of relief for now.
The primary machine learning method described in Andrew’s course is supervised learning, a type of AI that learns input to output (A to B) mappings. In order to lay a sturdy mental foundation for the lessons to come, Andrew provides various examples of supervised learning – for instance, considering an input as an email and an output as whether the email was spam or not. In essence, an input is placed in a proverbial “black box” which returns a single output. What this “box” accomplishes, as one might expect, varies by application.
Andrew manages to cover a generous amount of material in week 1, including the importance of data in machine learning and neural network training, what data is useful data, the difference between machine learning and data science, what machine learning can and cannot do, as well as what makes an AI company. I will try not to spoil everything, because it is highly recommended that the reader (that’s you!) take the course and experience it for his or herself! However, it is helpful to note that Andrew touches on a very important topic in week 1 – what makes an AI company. This is significant, since Andrew has a brilliant history of successfully leading the integration of AI into extraordinarily large corporations and up-and-coming businesses alike. In fact, he was the co-founder and leader of Google Brain, “a deep learning artificial intelligence research team at Google” responsible for multiple groundbreaking Google projects such as image enhancement and Google Translate. Let’s just say, if anyone is going to help you begin effectively utilizing AI in your business, regardless of the real estate, Andrew Ng is one of the most equipped and qualified to do so. That said, it only makes sense that integrating AI into a business be a recurring theme throughout the duration of this course. Using the five steps shown in the image above, Andrew continues to develop a general basis of knowledge in AI applications with an emphasis on business integration.
In week 2, as hinted in the image above, Andrew describes the process of Building AI Projects. It is learned that, as in any project development, there is a sequence of predictable steps that one can follow to go about seeing an AI project through to completion. However, a challenging question arises – how do you select an AI project? What constitutes a feasible project, and what is “biting off more than you can chew?” Despite having a practical workflow for AI projects, it is inherently required that an AI project be present. Therefore, a tried and proven framework for brainstorming plausible AI project ideas is also presented.
“It turns out that becoming good at AI is not a mysterious process. Instead, there is a systematic process through which many … can become good at AI.”
Andrew Ng – AI for Everyone – Week 1 – What makes an AI Company?
In the same token that a road map is necessary to find one’s way to an untraveled location, a workflow is necessary for successful completion of AI projects. Expanding on his week 1 distinction between machine learning and data science, Andrew offers workflows for both project types. For machine learning, the following key steps are proposed:
- Collect data
- Train the model, iterating many times until satisfactory
- Deploy model
By the above steps, it is inferred that the more useful data the better – with “useful data” being described previously in week 1’s lessons. It is also inferred that such training could take a rather long time, as many iterations are often necessary. For data science, a slightly different workflow is suggested:
- Collect data
- Analyze data, iterating many times to get good insights
- Suggest hypotheses/actions
The obvious differences between machine learning and data science project workflows are how data is used and what type of output is obtained.
Andrew takes a minor sidebar in the midst of week 2 to express the importance of solidarity in a business attempting to integrate AI into their processes. In particular, every job function must, in some way, learn how to use data to improve their line of work. A company which utilizes AI in, say, the hiring process, but not in sales or other aspects of the business, is no more an AI company than a clothing store which only sells one particular line of their t-shirts online is an internet company. It takes a combined, cooperative effort to turn an average company into an AI company.
Andrew continues to explain the process of brainstorming ideas for worthwhile AI projects in a business. He emphasizes the importance of having both AI experts, those who know what AI can do, and domain experts, those who know what’s valuable for a business, work together in this process. A brainstorming framework is presented as follows:
- Think about automating tasks rather than automating jobs
- What are the main drivers of business value?
- What are the main pain points in your business?
The first step shown above should encourage the reader to recall the most important lesson from week 1 – all advancements in AI up to this point have been advancements in narrow AI, not general AI. That said, it’s nonsensical to attempt to automate an entire job, which often includes a handful of individual tasks, as this is out of the scope of what AI can currently accomplish. The next two steps are dependent on application – solving which problem would increase profit margin the most, and solving which problem would save the most time and effort for goals to be reached? This topic of selecting the right AI project is so imperative that Andrew dedicated two separate video lessons to the subject.
“Don’t expect an idea to naturally come overnight. Sometimes it happens, but sometimes it also takes a few days or maybe a few weeks to come up with a worthy idea to pursue.”
Andrew Ng – AI for Everyone – Week 2 – How to Choose an AI Project (Part 1)
To conclude week 2, Andrew discusses how to work with an AI team – a valuable tool given that step 2 of the AI Transformation is building an in-house AI team. Andrew describes exactly how experienced, professional AI teams visualize data and what these teams require in order to be successful. Additionally, possible pitfalls such as machine learning limitations, insufficient data, mislabeled data, and ambiguous labeling of data is discussed. Such pitfalls will be expanded on in week 3’s lesson videos.
By now, perhaps you’ve noticed a pattern. Each successive week, the painting in the backdrop of the lesson videos is replaced to indicate how far you’ve progressed through the course. This subtle change serves to dismiss monotony and encourage course completion. Needless to say, I looked forward to see which image would appear as I moved on to the next week’s videos!
In week 3, Andrew examines Building AI in Your Company. This continues the ongoing discussion of AI integration into business applications. First, two case studies, smart speaker and self-driving car, are examined. These cases are meant to expose the reader to real-life applications of the AI project development process – exactly what is required to see an AI project through to its eventual realization. For instance, in the case of the smart speaker, developers would have to program a trigger word to “wake up” the device to listen for a command. Developers would then employ speech recognition to examine the audio the device receives from a user, which would proceed to pass through an intent recognition algorithm such that a specific command (i.e. joke, time, music, etc.) is triggered. Therefore, the steps to process a smart speaker command are as follows:
- Trigger word/wakeword detection (i.e. “hey device”)
- Speech recognition (i.e. “play ‘Dare You to Move’ by Switchfoot'”)
- Intent recognition (i.e. “music”)
- Execute command (i.e. play the requested song)
Note that all of these detection algorithms are supervised (A to B mapping). Audio (A) is mapped to some response (B), which can be a simple binary output (0/1) or a complete string of characters (“play some music”). The same could be said of the case of a self-driving car. All of the micro-processes involved with the complex system are broken into various independent machine learning algorithms which implement supervised learning, taking an input and producing a desired output.
Having laid the foundation for working with an AI team in the previous week’s lessons, Andrew now discusses the structuring of such teams. He uses the before-mentioned case studies to provide a practical example of how an AI team would be organized. In the case of the smart speaker, most of the team would consist of Software Engineers, while a Machine Learning Engineer would obviously be necessary to develop the supervised learning algorithm. In the case of the smart speaker, more than one Machine Learning Engineer would be necessary. “They might gather the data of pictures and cars and positions of cars, train a neural network or train a deep learning algorithm and work iteratively to make sure that the learning algorithm is giving accurate outputs.” Such work would be too much for a single Machine Learning Engineer, so the AI team would be larger than that of a smart speaker’s development. In addition to Machine Learning Engineers, an AI team might also benefit from a Data Scientist who would provide insights which allow for better, more educated decision-making by the management team. They would also be responsible for managing the vast amounts of data involved with deep learning projects. One thing that Andrew says which might encourage a lot of people reading this right now is, “you don’t need 100 people to do most AI projects.” Such a statement is precisely why it is important to first evaluate one’s particular needs, or a company’s particular needs, and perform pilot projects prior to assembling a team.
The remainder of week 3’s lessons are dedicated to examining each step in the AI Transformation Playbook in explicit detail, as well as further listing common pitfalls in AI and taking your first steps in AI. The AI Transformation Playbook is a strategic list Andrew developed to help companies truly become AI companies. It is tried and proven to jump-start companies into the complex, yet rewarding world of AI. So as to not spoil the fun, I will leave it up to you to go and learn these concepts. There is simply too much to summarize in a paragraph or two, and only watching the lessons for yourself will give proper justice to the material. Following his discussion of the AI Transformation Playbook, Andrew lists a few “dos and don’ts” of AI. These are particularly helpful to remember, so I have copied these below for review.
| Don’t | Do |
| Expect AI to solve everything | Be realistic about what AI can and cannot do given limitations of technology, data, and engineering resources |
| Hire 2-3 Machine Learning Engineers and count solely on them to come up with use cases | Pair engineering talent with business talent and work cross-functionally to find feasible and valuable projects |
| Expect an AI project to work the first time | Plan for AI development to be an iterative process, with multiple attempts needed to succeed |
| Expect traditional planning processes to apply without change | Work with AI team to establish timeline estimates, milestones, KPIs, etc. |
| Think you need superstar AI engineers before you can do anything | Keep building the team, but get going with the team you have |
Various options for “getting your feet wet” in AI are introduced next, from getting friends to learn about AI to discussing with your CEO/Board the possibilities of AI Transformation. Truth be told, even the fact that you are reading this right now is an excellent first step, and taking Andrew’s course in its entirety is another fantastic start. The sky is the limit with AI, and in many scenarios users themselves are the ones who can get in their own ways to progressing towards completing an AI project, whether personal or commercial.
This leads us to the final week in Andrew’s course, AI and Society – as indicated by the painting of course! The finish line is in sight and victory is just beyond our grasp. So much has already been learned, and behold, it is time to put it all together and conclude! Surely you’ve heard it said, “with great power comes great responsibility.” Such rings true in AI as well. As Andrew points out, “AI is a superpower.” With such knowledge, one can truly save the world – or damage it beyond visual repair. It is here that knowledge meets wisdom, and the reader learns to use their talents for the greater good.
“AI is a superpower that enables a small team to affect a huge number of people’s lives … it’s important that you learn about these trends, so that you can make sure the work you do leaves society better off.”
Andrew Ng – AI for Everyone – Week 4 – Introduction
While individuals with knowledge in AI can develop, say, spam filters for emails, the same can develop attacks, called “adversarial attacks,” to attack those spam filters. I said previously that the sky was the limit, but there is also a limit down in the bottomless pit! With advancements in narrow AI come a host of open-source resources and knowledge beyond the scope of this summary which are accessible by anyone with a computer. Andrew warns that such information being open to the public requires the “good guys” to continuously develop better ways of circumventing attacks from the “bad guys.” AI has a massive effect on society, so much so that how it is used carries importance of monumental proportions.
Andrew continues his concluding lessons by ensuring that the reader has a realistic view of AI, its capabilities and its effects on society. AI algorithms can be biased, just like humans; AI algorithms can be attacked; AI algorithms can do a lot, but not everything; AI algorithms can be misused; and the list goes on. Having a realistic view of AI is what separates those who manage to successfully implement AI solutions which contribute to society in a sensitive, yet groundbreaking manner from those who are too pessimistic or even optimistic to reach the finish line.
Andrew allocates his final lesson prior to concluding by expressing just how large of an impact AI has, and is expected to have, on society as the years progress. He points out that due to advancements in AI, it is predicted that the number of jobs created will outnumber the number of jobs displaced by the year 2030. Though it is certainly impossible to predict exactly what the outcome will be a decade from now, what is certain is that AI is making waves in our society and across the globe. The general populous and engineers alike would be wise to stay privy to how AI is progressing, and how their lives will be affected by such progression.
With a hearty “congratulations,” Andrew concludes his course having exposed over 200,000 individuals enrolled in his course, including me, to the rewarding, fast-paced, growing, intriguing field of Artificial Intelligence. As a side-note related to this course, I recommend viewing the “optional” videos at the end of each week’s lessons as they add depth to the material not obtained by simply watching the required videos. Additionally, you can take the quizzes to test your comprehension! I for one highly recommend anyone even remotely interested in learning about AI to take this course. You will certainly not be disappointed!
FAU CAP6635 Artificial Inteligence 