Keith McCormick's Blog, page 2
March 7, 2021
OneTAKE Live with Ian Barkin
Welcome to OneTAKE Live! In this episode, host Ian Barkin speaks with Keith McCormick, the chief data science advisor at CloudFactory, about the complex world of big data, the value of democratizing analysis in an organization, how best to utilize analytics as a predictive tool, and much more!
Algmin Data Leadership
Building Analytics Teams with Keith McCormick.
Tech Society Podcast
A special 2 part episode this week. Alex and John from Ninja Software have the pleasure of chatting with Keith McCormick. Join us as we take a deep dive into the world of AI and Data Science, to clear the stigma surrounding the future of AI, predictive analytics and where to draw the line between AI and statistical analysis.
Special Election episode: Listen here
March 6, 2021
Roaring Elephant Podcast
With Machine Learning and AI being on everybody’s minds and lips these days, we invited Keith McCormick to joins us and discuss the do’s and don’ts of leveraging the undeniable power of ML in your organization. In this first part, we cover the technology part and we’ll have the human side of the story for you next week!
TDWI Event Interview
3 Elements for a Successful Analytics Practice with TDWI faculty member Keith McCormick.
UC Irvine DCE Magazine interview
UC Irvine DCE Magazine interview with instructor Keith McCormick.
Team Engagement Podcast
Short format podcast about teams. Keith McCormick shares his insights on analytics teams.
January 6, 2021
Defining Machine Learning
Am I the only one that finds that most definitions of Machine Learning are highly misleading? Google the exact phrase “without being explicitly programmed” and you’ll find thousands of variants that everyone seems to embrace. At the risk of being a contrarian, I find that this phrase conjures up images of computers learning like puppies or babies. Something which i believe is still years (or decades) from achieving.
Yes, I know that the phrase is attributed to Samuels (this is a topic worthy of a little internet searching too) and that it has a glimmer of truth to it, but I think we are all going to have to suffer though executive briefings where we are trying to explain why our supervised learning models have actually degraded over time, not magically improved.
I can across one particularly striking definition on Wikipedia: Machine learning (ML) is the study of computer algorithms that improve automatically through experience. It is attributed to Tom Mitchell’s Machine Learning (1997). I can’t claim to have read the whole book so I don’t want to offend by taking the definition out of context, but that sounds almost magical to me.
For the time being, supervised learning still dominates. We still need learning algorithms. See the 9th Law of Data Mining for more about how our models still degrade. Most techniques, other than Deep Learning, still need feature engineering to be effective. And since Deep Learning is black box many can’t use Deep Learning. We need to explain to our colleagues what machine learning is without appealing to magic. There are even t-shirts with phrase like “the future is unsupervised” which I find rather charming, but at client sites I’m still solving problems with supervised learning.
Not surprisingly Russell & Norvig do an excellent job with their definition.
An agent is learning if it improves its performance after making observations about the world. When the agent we call it machine learning: a computer observes some data, builds a model based on the data, and uses the model as both a hypothesis about the world and a piece of software that can solve problems.
Lately, I’ve been using the following definitions in my presentations and courses.
Machine Learning:
A broad term that generally refers to presenting carefully curated data to computer algorithms that find patterns and systematically generate models (formulas and rule sets).
Now, Supervised Machine Learning:
Given a dataset with a “Target variable” and “input variables” a modeling algorithm automatically generates a model (a formula or a ruleset) that establishes a relationship between the Target and some or all of the input variables.
So, while the algorithms are explicitly programmed, the models are not.
I could be accused of spending a lot of time reading definitions and crafting my own. I’d be guilty of that accusation. But I believe them to be important because carefully written they clarify what we are trying to do. Also, in a work setting, it can have the very practical impact of knowing “who should do what.”
How do we know what assignments to give the data science team if we haven’t decided what they do?
October 1, 2020
Data Mining Defined
As I write this in 2020, the phrase Data Mining is increasingly out of fashion. It has become associated with privacy concerns as shown by this quote from a Computer World article entitled Big data blues: The dangers of data mining.
As companies become experts at slicing and dicing data to reveal details as personal as mortgage defaults and heart attack risks, the threat of egregious privacy violations grows.
Clearly, privacy concerns are both valid and important but most of the projects that I’ve been involved in over the last 25 years have used data that is internal to the organization. For instance, it is common to look at mortgage defaults on a bank project, but it actually would be very unusual to try to acquire mortgage default information through a third party for a health insurance project. There are so many well-known and strange case studies in the news that it is common knowledge that attempts to acquire private data occur, but it is not as common as the famous case studies lead us to believe. Also, we have all been annoyed by pop-up ads that appear mysteriously after a Google search. It is so commonplace that we might reasonably believe that all machine learning is done in this way.
If there a negative connotation then why use the phrase Data Mining at all?
When the Cross-Industry Standard Process for Data Mining (CRISP-DM) was written in the late 90s the term was not yet associated with privacy concerns. The document is still influential today and is still considered to be the de-facto standard. Attempts to modify it have never truly taken hold, and they rarely depart greatly from the original. I discuss some CRISP-DM alternatives in a video from my Data Assessment course. Therefore, I still embrace the phrase Data Mining and the recommended process in the CRISP-DM document. Perhaps the safer phrase to use today, albeit a bit wordy, is Traditional Supervised Machine Learning.
Predictive Analytics was the phrase of choice for a while, but that is now less often used. Artificial Intelligence as a description is so all-encompassing in how it is used today that it is increasingly unclear what it is referring to, but it is safe to say that Data Mining is a subset of AI. One of my favorite attempts to define AI is by one of the co-creators of CRISP-DM, Colin Shearer. The same can be said for Data Science, which is also so broadly defined as to create problems.
I first started trying to nail down a good definition many years ago, but I know favor this one that I used in my LinkedIn Learning course: The Essential Elements of Predictive Analytics and Data Mining.
Data is the selection and analysis of data, accumulated during the normal course of doing business, in order to find (and confirm) previously unknown relationships that can produce positive and verifiable outcomes through the deployment of predictive models when applied to new data.
The course is in many ways a course about CRISP-DM, but with relatively few explicit references to the document as I describe the elements of Data Mining. In the first half, I describe the characteristics of a Data Mining project that you will always encounter and should therefore be prepared for. In the second half, I make explicit mention of each phase of CRISP-DM and also discuss the Nine Laws of Data Mining.
This video excerpt provides a high-level overview of the entire data mining process.
August 5, 2020
Kai-Fu Lee’s AI Superpowers and his reflections on AI in 2020
Kai-Fu Lee is uniquely qualified to describe the path the AI community has traveled since the 80s and to anticipate where it may be headed next. Although AI Superpowers: China, Silicon Valley, and the New World Order was written in 2018, it is just as timely in 2020, and in recent interviews with Time and the Aspen Ideas Festival he predicts that the predictions he made in the book may come even faster in the world that is being remade by the COVID-19 pandemic.
Lee was born in Taiwan, spent his teens in the USA, and then attended Columbia and Carnegie Mellon, earning his PhD in 1988. That his formative years were spent studying Computer Science and AI, with some of the best in the field, during that particular period has shaped his professional life. Neural Networks and AI were experiencing a brief but enthusiastic renaissance during those years, falling between the two so-called “AI Winters.” His rise was rapid. A young Lee can be seen in footage from 1992, standing next to Apple CEO John Sculley on the set of the Good Morning America, discussing a demonstrating Apple’s Caspar, which is in a direct lineage to today’s Siri.
Just prior, during the first AI Winter, neural networks were largely abandoned and expert systems reigned. Minskey’s and Pappert’s Perceptrons had initiated the period of skepticism in 1969, but backward propagation and the multi-layer perceptron brought renewed interest to neural networks in the early 80s. His insider perspective was a revelation to me, helping me better understand that period. Just a few years younger, I studied computer science as an incoming freshman about the time that Lee was finishing his Ph.D. Of course, his experiences studying at elite institutions with future Turing Award winners were different from mine in almost every way. Still, it finally made sense why no one was talking about artificial neural networks when I was an undergrad. I associated AI with expert systems and not neural networks during those years as they were not in vogue. However, I do remember being fascinated with Minskey’s The Society of Mind, which was written the year that I graduated high school.
AI Superpowers has been lauded by numerous tech celebrities, including Nicholas Thomson of Wired, Yann Lecunn, the recent Turing Award winner, and Chris Anderson of TED. It is hard to disagree. It is compelling and a great read. It is also an easy read at just over 200 pages. It gets my highest recommendation and is still the best contemporary commentary on the near future of AI politically, economically, and technically. For audiobook fans, the version on Audible is excellent and runs 9 hours and 28 minutes. If you require an appetizer to be convinced, consider his TED Talk, which touches upon some of the book’s themes, but with a particular emphasis on work-life balance and his experiences as a recovering workaholic and cancer survivor.
As I reflect on the book and his recent 2020 interviews, three themes stand out: Alpha Go and its impact on AI in China, the cultural differences between China and the US that are fueling China’s rise in AI, and the future of work.
In an interview with Wired’s Nicholas Thompson at the virtual Aspen Ideas Festival, Lee predicts that COVID-19 will accelerate AI trends for two reasons: using AI to reduce human contact during the pandemic and accelerated digitization from forces as disparate as increased food delivery and work from home. His Oxford talk, given shortly after the book came out, is perhaps the best summary of his thoughts on AI’s recent history. He explains that during the 80s, researching AI was for him a quest to understand the brain itself. A pursuit that he characterizes as failing, which is interesting because Demis Hassabis, CEO of DeepMind, is still on that quest:
“We think of DeepMind as kind of like an Apollo program effort for AI. Our mission is to fundamentally understand intelligence, and recreate it artificially.”
Demis Hassabis quoted from AlphaGo The Movie
Then, Lee describes moving on to trying to finding statistical patterns, but “you couldn’t get enough data”. 100 MB of data for his PhD thesis cost $100k in data storage for the university. Now, that would be a penny a day. Lee believes that the latest wave, Deep Learning, will succeed where the others have failed.
AlphaGo awakens Chinese AI
The first major theme of the book was China’s awakening by what Lee calls China’s “Sputnik moment”. The opening chapter of the book was a real eye-opener for me, and I think it would be for most Americans. While AlphaGo’s defeat of the world’s top players was newsworthy in the AI nerd circles in which I exchange LinkedIn posts and tweets, it was hardly the stuff of conversation with friends and relatives. In contrast, in Asia, 100s of millions of people watched the matches in real-time. The human players were known to the audience and are celebrities in Asia. DeepMind, the company that developed the technology has an absolutely remarkable YouTube channel with countless videos, including the Go matches and commentary, so if the events largely passed you by at the time, it is not too late to experience them.
There is recent news in regarding this topic, as well. Lee Sedol, considered the world’s top player, announced in late 2019 that he will be retiring from the game. Sedol played AlphaGo and won one match out of five in 2016. Alpha Go The Movie is a beautiful documentary, working to inform about the technology but also as a great film that any film festival fan would enjoy. Ke Jie, the player that Kai-Fu Lee focused on in the book, played AlphaGo in 2017. More recently, AlphaGo Zero beat the previous AlphaGo 100 matches to zero.
Breakthrough Innovation and Normal Science
A second theme of the book reflect’s Lee’s unique perspective as an executive in both the US and China and inspired the title of the book. He writes of the Silicon Valley’s penchant for supporting, nurturing, and gambling on the “breakthrough idea”. He describes a mix of economic (funding), cultural, and education styles working together to encourage this. In an extended interview on Lex Fridman’s AI podcast, they discuss this at some length, including brief mention of Fridman’s experience growing up in a Russian educational system. Lee’s description of China’s strength, the diligent work on persistently wrestling with an existing approach to optimize it by accumulating, manually labeling, and cleaning data on a scale unmatched by the West, reminded me of Thomas Kuhn’s description of Normal Science.
The success of a paradigm – whether Aristotle’s analysis of motion, Ptolemy’s computations of planetary position, Levoisier’s application of the balance, or Maxwell’s mathematization of the electronic field – is at the start largely a promise of success discoverable in selected and still incomplete examples. Normal Science consists in the actualization of the promise, and actualization achieved by extending the knowledge of those facts that the paradigm displays as particularly revealing.
Thomas Kuhn, The Structure of Scientific Revolutions
He’s quite specific in the interview with Fridman about this – the Chinese engineer does whatever is necessary to overcome problems and to get the algorithm to work. The US engineer is more likely to abandon the algorithm in favor of trying to invent a new approach. They both work, but Lee believes that the current state of the industry will economically favor the “low risk” Chinese approach because the application of existing technologies can solve numerous day to day problems. Lee also argues that China’s venture capitalist funding favors working with proven solutions. He believes that westerners conflate this with the copy cat phenomenon, and as a result, some tend to underestimate Chinese companies. This also reminds me of Kuhn who at times seems almost apologetic when describing Normal Science, elaborating that it is harder that it looks: “Few people who are not actually practitioners of a mature science realize how much mop-up work of this sort a paradigm leaves to be done or quite how fascinating such work can prove inn the execution.”
The resemblance to Kuhn’s theory makes Lee’s argument even more persuasive. If Chinese companies are the masters of this kind of “mop-up work,” they may lead the world in AI within 5 to 10 years.
The Future of Work
Lee anticipates high levels of unemployment from the rise of AI, soon enough to affect our careers and our lifetimes. Yet, he is an optimist. He rejects the generalized anxiety around AGI (Artificial General Intelligence) and the singularity. If that happens at all, he predicts that it is “many decades” away. He mentions another possible kind of dystopia. In the short story Folding Beijing futuristic upper-class life in techno-utopia while a lower caste of laborers works all night to keep the city clean and functioning. But Lee believes that there is no need for this future either as those same kinds of jobs done by the story’s “Third Space” can be done with automation. His optimism comes from the fact that “narrow AI” will shorten our work weeks, increase prosperity, and alter the very nature of work. In this way, he is somewhat reminiscent of The Future of the Professions, which will be released in a 2nd edition this year.
In the book and in his TED talk he relays the story of the colleague who developed an app for the elderly that was designed for ease of use and yet had a customer service team swamped with constant calls. It wasn’t poor design – it was that the elderly users of the app wanted more human contact. This represents a long-standing trend. I was early in my high school career when I was required to read the then bestseller Megatrends: Ten New Directions Transforming Our Lives. One of the ten trends was “High Tech, High Touch” which underscored that the more technical our lives become, the more we seek human contact in order to find balance.
In much the same way, Lee believes that jobs that involve compassion and caregiving will not only remain but thrive while easily automated jobs (and there will be many) will be lost. He thinks we can and should embrace this. He indexes jobs and places them on four quadrants. The jobs at the most immediate risk are both “asocial” and “low dextrous”. Robots work best in structured environments like factories so he thinks that we overestimate the likelihood of android like robots in the home. More likely are the robotic carts that one sees in Amazon warehouses. During a recent COVID19 quarantine all of his meals were delivered by robots from the restaurants to his apartment building. Apartment buildings are easier than individual homes so naturally, this is the first development and it has already arrived. He expects driverless trucks on highways before passenger cars on local streets for the same reason and expects China to lead in this.
AI Superpowers is just as timely as it was in 2018. Combined with Lee’s frequent interviews and social media outreach, it is still the best source for understanding AI’s recent history and short term future.
