Hey there, business owners! Are you ready to take your business to the next level and stay ahead of the competition? Then it’s time to embrace the power of Artificial Intelligence (AI) strategy! AI is no longer a futuristic concept but a reality that is transforming businesses across all industries. And in this article, I’m going to show you how to leverage AI strategy to help your business thrive.

So what exactly is AI strategy?

It’s the use of AI technologies to enhance business operations and achieve strategic objectives. By integrating AI technologies into your business operations, you can create a competitive advantage that helps you achieve your business goals faster and more efficiently. And it’s not just about implementing AI solutions — it’s about understanding how to use them to drive growth and innovation.

Advantages of AI strategy

AI strategy can be used in various areas of your business, such as marketing, customer service, operations, and finance. By automating repetitive tasks, reducing costs, and improving customer experience, AI can help you to streamline your business operations and stay ahead of the curve.

One of the biggest advantages of AI strategy is automation. With AI, you can automate repetitive tasks such as data entry, customer support, and scheduling. This frees up your time and resources, allowing you to focus on more strategic tasks that require human creativity and decision-making.

AI can also provide valuable insights into customer behavior and preferences, helping you to improve customer experience, personalize offerings, and increase customer loyalty. By identifying operational inefficiencies and streamlining processes, AI can help you to save costs, increase productivity, and allocate resources more effectively. And by enabling you to innovate and differentiate yourself from your competitors, AI can provide you with a powerful competitive advantage.

How do you create an AI strategy for your business?

Here are the steps:

1. Identify your business objectives. Determine what you want to achieve with AI technology and how it can help you to achieve your strategic goals.
2. Assess the data you have available. Determine how your data can be used to drive growth and innovation. Identify any gaps in your data and determine how to fill them.
3. Identify AI solutions that can help you to achieve your strategic objectives. Research the market and identify the most suitable solutions for your business.
4. Develop an implementation plan. Outline how you will integrate AI solutions into your business operations, determine the resources you will need, timelines, and potential challenges.
5. Monitor and evaluate the effectiveness of your AI strategy regularly. Use metrics to measure progress and identify areas for improvement.

Successful AI strategies in action

1. Amazon uses AI to personalize customer recommendations, optimize pricing strategies, and improve supply chain management.
2. Netflix uses AI to recommend personalized content to its customers based on their viewing history and preferences.
3. Uber uses AI to optimize pricing strategies, improve route optimization, and enhance customer experience.
4. IBM uses AI to improve business operations, develop new products, and enhance customer experience.

Implementing AI strategy in your business

When it comes to implementing AI strategy in your business, there are a few steps you need to follow:

1. Educate your team about AI technology and its potential benefits for your business.
2. Identify AI solutions that can help you to achieve your strategic objectives.
3. Develop an implementation plan that outlines how you will integrate AI solutions into your business operations, determine the resources you will need, timelines, and potential challenges.
4. Train your team on how to use AI technology and its potential applications in your business operations.
5. Monitor and evaluate the effectiveness of your AI strategy regularly.

Conclusion

Of course, implementing AI strategy can come with its own set of challenges. Data quality and availability can be an issue, and you may need to invest in new technology and training to get the most out of AI. But with the right strategy, and vision, you can build a competitive advantage for your business.

You can follow me on Twitter / LinkedIn to get updates on stories like this. Feel free to visit my blog, where I write about topics like MBA admissions, life in Canada, and my experiences with cutting edge technology.

From Survival to Thriving: How to Leverage AI Strategy for Your Business’s Success was originally published in Predict on Medium, where people are continuing the conversation by highlighting and responding to this story.

The Menace of ChatGPT

Not even a couple of months has passed since the ChatGPT has launched, yet the teachers all over the world are now terrified of ChatGPT. Can you blame them, though? After all, it’s simple to see how the tool could be abused. Students might submit essays and assignments just by entering prompts at ChatGPT. In conjunction with other AI assisted writing and plagiarism checkers, students might produce a highly coherent and plagiarism-free writing sample in a matter of minutes. These samples would be indistinguishable, if not superior, to those submitted by hardworking and honest students. Essentially, take-home assignments cannot be utilized to assess and reward subject knowledge. This is quite troubling for educators.

Educators Fight Back

These AI written-text generator tools have the potential to bring about a generational shift in learning. That is indeed a challenge to the status quo. And the school administrators are starting to take this challenge head on. Responses include curriculum overhaul to increase emphasis on oral exams, live in-class internet-free exams, and group projects. The New York City Public School system went as far as blocking ChatGPTl from their Wi-Fi networks. While the reactions are predictable, they are not all encompassing. For example, NYC public schools can’t stop students from using their smartphones to access the tool. Can they do better than this? Can they even embrace ChatGPT and disrupt the teaching model itself? Turns out they can! Like flip the teaching on its head.

Flipped Classroom Model

“Flipped Learning is a pedagogical approach in which direct instruction moves from the group learning space to the individual learning space, and the resulting group space is transformed into a dynamic, interactive learning environment where the educator guides students as they apply concepts and engage creatively in the subject matter” (The Flipped Learning Network, 2014).

That may sound complicated, but it simply means that students complete “schoolwork at home and homework at school.” Still vague? Let’s give it another shot. In the flipped classroom model, students complete the lower level of cognitive work, such as reading assignments, financial modeling, and so on, before class. When they come into the class, they engage in higher levels of cognitive work, such as addressing a real or fictional problem in collaboration with teachers and peers. Now, that does sound familiar. Ever heard of Harvard Business School’s “case-based” method? Students are essentially placed in the shoes of case protagonists and are required to offer a well-articulated and reasoned recommendation. However, outside of business schools, case-based methods are not that popular.

Embracing The Inevitable AI era

The reality is that students will use artificial intelligence in their working life, in some form or another. Just like we use calculators for the simplest things including calculating tips. Wouldn’t it be amazing if we could start AI tools in the classroom right away? For example, we can let students use ChatGPT and learn all of the theory concepts at home, take positions based on the information, obtain supporting arguments for their positions, and so on. I bet that will lead to a much better classroom discussion than what we have ever seen.

What do you think? Let me know your thoughts in the comments!

You can follow me on Medium as well as on Twitter / LinkedIn to get updates on stories like this. Feel free to visit my blog, where I write about topics like MBA admissions, life in Canada, and my experiences with cutting-edge technology.

ChatGPT and the Flipped Classroom Model: Why Case Method is well suited for the era of AI tools was originally published in Predict on Medium, where people are continuing the conversation by highlighting and responding to this story.

Over the last decade or so, Artificial Intelligence (AI) has evolved into a catch-all term for any accomplishments of computer algorithms that formerly required human reasoning and thought. Everything from AlphaGO defeating the then Go-champion, Lee Sedol, to autonomous vehicles testing on public roads fell under the tent of the AI debate. The recent achievement in producing art through stable diffusion, which many label as “theft of creativity,” has taken the globe by storm.

Given such widespread success, it’s no surprise that enterprises rushed to adopt AI in the hopes of gaining a competitive advantage. However, recent research from Fivetran has signaled that a significant majority have failed in that regard. The key result of the survey was that while 87 percent of enterprises believe AI is critical to their company’s survival, 86 percent do not trust AI to make business decisions without human input. Why is that?

Before we delve into the reasons, let’s set the expectations right. We’re not dismissing AI because it isn’t what we expected it to be based on science fiction; we’re talking about AI that already exists, performs well in demos, and is presented at conferences and summits, but isn’t being deployed for commercial purposes. So what could be going wrong here?

Let’s start with the cost of running AI systems. Over the past years, there have been phenomenal advancements in computing hardware, algorithms, and infrastructure services (such as Google TPU and Amazon AWS Lambda). As a result, setup and per-unit costs have been fundamentally lowered, thus decreasing the barriers to entry. However, the growing size of benchmark setting neural network models is keeping AI expensive. Training and deploying larger models such as object detection in autonomous vehicles or Bert-like networks in NLP applications currently requires utilizing several Nvidia GPUs or $$$ in AWS instances. It is also costly to the environment because it uses a lot of power and energy. Turns out most of the large neural networks are designed for general purposes and thus are an overkill to the specific applications at the enterprise level. Instead of throwing humongous networks, practitioners should employ domain knowledge to prune the network. Apart from using transfer learning to avoid retraining a large network, enterprises should also consider knowledge distillation techniques to come up with a slimmer network for their specific use case. Lastly, AI accelerators should be used for inference purposes as such devices are designed to run in real time while consuming less power.

Another major reason for AI systems missing expectations is the low availability of data — both in terms of quantity and quality. In fact, the survey highlights that about 71% of organizations are struggling to find the data they need to run AI programs, workloads, and models. Furthermore, only a quarter of those that manage to obtain data are able to transform that data into actionable insights. Even the successful ones opt not to commercialize AI because they don’t trust it. The reason for this is that the decision making skills of supervised learning algorithms, the commonly deployed forms of AI, are only as good as the humans who label the underlying data. That is, it inherits all of the labeler’s biases and preconceived notions. While the human is removed from the loop, i.e., time is saved, it does not instill much trust in decision making. In the last couple of years, there has been increased conversations on biases in historical data and ways to rectify it. But that may not be sufficient until neural networks become more transparent. Enterprises should actively look into the sources of biases that may have creeped in and use data masking techniques to eliminate the effect of those in the final outcomes.

Despite advances in AI, the currently deployed neural networks are still very much black-boxes. The explainability worsens the deeper the neural network gets. Since decisions that cannot be explained cannot be trusted, their commercial deployment is limited. As AI gets more complex, it’s important to have clear governance around things like explainability, fairness, and bias. This is a challenge for many companies because it’s hard to establish these guidelines when there is no agreed upon standard. This gray area is where no large company’s executives would like to operate. It’s hard to get buy-in from decision makers when they don’t really understand what AI does or how it works. And we ML practitioners are not making it easy.

The communication gap between ML engineers and the executive team makes matters worse. Since the KPIs are different for the two different teams, most AI projects don’t go beyond the pilot phase. The ML engineers are not effectively converting metrics such as accuracy, mean average precision, and latency into metrics such as revenue growth and cost savings, the KPIs executives care about. This has been a big source of my personal frustration through my years as a machine learning researcher. Building a layer of product managers between the developers and the executives to translate metrics into $$$ can improve the outcomes. Simultaneously, the executives should participate in AI learning programs for business leaders. So that your engineers do not roll their eyes when you talk about the AI initiatives at your company, The impact of this issue will further subside as enterprises all around the world are striving to be at the forefront of digital transformation, resulting in the appointment of technology executives to the coveted CEO position.

The potential commercial benefits of AI should compel corporate executives to expedite the transition from isolated pilot initiatives to a comprehensive AI strategy. An AI-centric business strategy requires continual investment in cutting-edge AI research while ensuring that the benefits are shared throughout the organization. This can be accomplished by making created and acquired datasets available across the organization, investing in AI infrastructure and talent, and encouraging executives to become more AI-literate.

You can follow me on Twitter / LinkedIn to get updates on stories like this. Feel free to visit my blog, where I write about topics like MBA admissions, life in Canada, and my experiences with cutting edge technology.

Why enterprise AI has been a bust, and how it can recover was originally published in Predict on Medium, where people are continuing the conversation by highlighting and responding to this story.

Applicants spend several hours writing their essays, working with their recommenders, and fine tuning their resume. This is on top of the prep needed to get a solid score on the GMAT or GRE. So how does one even know if their application will stand out? When can they stop obsessing about revising their essay? Here is one approach that worked for me and others I mentored.

We begin by putting ourselves in the shoes of admissions committee members. They have read your essay, your responses to short-answer questions, your resume, and the reference letters you submitted. Now if one has to distill all of that information into a sentence made of 3–4 adjectives, that forms a description—a description of you as a candidate. The adjectives start with your ethnicity and nationality, followed by gender and gender identity, and then your profession and side projects/hobbies. For example, when I applied, I could characterize myself as an Indian male engineer who works in auto/self-driving cars, delivers invitational speeches, and loves to teach and volunteer at local high schools. Another example could be of a Hispanic female consultant who was a D1 tennis player and is now a coach. Notice that the first half is for demographics because the business schools want to build a diverse class, and the second half is to define you professionally and personally. What are you good at? What drives you? What motivates you?

So once you write your application package, ask someone who doesn’t know you to describe you using this framework. The first step would be to verify if the reader’s description matches what you intended. Thereafter, think hard and long about whether that is unique. In an ideal scenario, your descriptive sentence is so distinctive that you are the only one in the application pool. There could be several Hispanic females, but also being a consultant narrows it down, and then being a former D1 tennis player who is now a coach narrows it down even further.

We have seen many applicants with lower than average GPA, GMAT, etc. make it to the top business schools because their descriptor sentence is very unique. But what if it is not? That can happen with over-represented applicants such as Indian males, consultants, and bankers, etc. That’s where hard facts come into the picture, like your GMAT score, your undergrad GPA, the prestige of your undergraduate institution, the selectivity of your employer, the awards you have won, and so on. The goal should be to be the best person associated with that descriptor.

This framework is even more powerful when you are a few years away from your MBA application. It helps you to make career and personal choices that enhance your “uniqueness” and thus improve your chances of getting into a top business school.

If you like this story and want to see more like it, please give me a follow. Also consider visiting my personal website where I write about topics like these :)

DISCLAIMER: This piece is my personal opinion and I am not asked or paid for by HBSOnline.

Less than two years back, I was constantly hearing about HBSOnline CORe (called HBX CORe then) on LinkedIn. Some of my connections took it, and they had really nice things to say. I checked out the syllabus and found it interesting. The course is comprised of three subjects: Financial Accounting, Economics for Managers and Business Analytics. For detailed syllabuses, check the following links: FA, EfM and BA. As the syllabuses indicate, there are 7 modules in Financial Accounting, 5 in Economics for Managers, and 5 in Business Analytics — totaling to 17 modules. What intrigued me was this idea of “case-based” method of teaching — where important concepts are explained via case studies on real businesses. I felt that the tuition fee was a bit steep. Also, it was difficult for me to get my employer pay for it, as I work in a technical field and this certificate isn’t exactly technical. I decided to apply anyway and figure along the way. The application was fairly simple and one can finish that in about 10 minutes. The admissions committee is mostly interested in knowing a bit about you and your motivations of taking CORe. After a couple of weeks, I received an admit letter. I was very excited, but I still didn’t think it was worth the money. I had the option to defer it few times, so I did exactly that for 15 months! Crazy, I know!

Anyway, in September 2018, I was trying to move out from the technical team and get into the strategy team. It needed convincing a lot of people at very high levels in my company, and a certification such as CORe would have helped tremendously. I also realized that a tuition fee raise was planned from 2019. If I couldn’t join October 2018 cohort, I might have to pay 300$ additional for subsequent cohorts. Also, October 2018 cohort was an extended cohort, i.e. the duration of the cohort was 17 weeks, which is the most relaxed cohort. There are CORe cohorts of 8 weeks too, the other end of spectrum, which I can imagine being very difficult to manage. Even the extended cohort takes up about 10–15 hours per week. 8-weeks CORe is probably twice as challenging as Extended CORe. With extended CORe’s duration being 17 weeks, it fits well with “A Module Every Week” schedule.

In short, a bunch of reasons made October 2018 cohort very attractive to me, so I paid the fee and enrolled. Around mid-October the course began. I noticed that the total cohort size was about 900, with participants split into two sections. There was a Facebook group too to allow participants to casually discuss the course. From the interactions there, I realized there was so much diversity in the cohort. Not just in terms of race, gender or nationality, but also in their education, work experience, and motivations to pursue CORe.

Over the next 17 weeks (+2 weeks for winter break during end of December), I completed all of the 17 modules. Completing a module involves going through the course material as it is presented (you can’t skip ahead), answering in-module questions, going through “cold calls”, providing your reflections and rating reflections from peers and then passing the module quizzes. Module quizzes have 20 multiple answer style questions, with moderate difficulty level. I haven’t personally experienced this, but scoring below 50% doesn’t meet the passing requirements for that particular module. For whatever reason if one cannot finish the module on time, the course automatically assigns a score of 0. The participant still has to finish the module to appear for final exam and get the course certificate.

Talking about the course material itself, the difficulty level perceived is very subjective. Financial accounting was most difficult for me, while Business Analytics seemed the simplest. Having an engineering background helped, I guess. Material presentation is highly systematic and the later modules reinforce the understanding of earlier modules. I felt that Economics for Managers was a little verbose at times, but that’s just the nature of the subject. For me cold calls were the most interesting feature of the course. During a cold call, a question would appear on the screen, and the participant is given 1-2 minutes to answer the question. These questions are situation-based and one has to put his/her executive hat on to answer the question. Thus there are no right or wrong answers.

On January 8th, 2019, during the middle of our cohort, HBX rebranded itself as HBSOnline to raise awareness of its online courses. I always associated HBX brand with Harvard Business School, but apparently, the rebranding of HBX to HBSOnline improved the perception, even if it wasn’t lacking. As per this article, rebranding has led to a huge spike in enrollment.

At the end of it all, there will be a closed-book final exam.This final exam will be proctored and will be taken at a Pearson approved testing centers. (A side note about testing center: I was constantly distracted by loud conversations between the test administrators. And some of my batch mates shared anecdotes about their computers freezing midway during the exam!). The whole thing may sound scary but its not too bad. HBSOnline provides you course summary kit, and a set of practice exams which are good representatives of the final exam. Now, this is my final exam experience — and the format may change in the future. For all three subjects, I had to answer 45 multiple choice questions in 60 minutes. The difficulty level is slightly higher, but everything is covered in the syllabus — no surprises.

Few weeks after the exam window ends, CORe announces the results. As per the CORe guidelines — Final grade is comprised of 50% final exam score, 33.33% quiz scores average and the remaining 16.67% comes from participation. For computing quiz score average CORe drops the lowest score in each of the three subjects and then average out the remaining ones. The mechanism to calculate participation grade is (intentionally?) kept vague, but some aspects are made clear by the program coordinator. Like sharing articles (still recommended) doesn’t count, but reviewing other student’s reflections and helping out peers in “Peer Help” section is highly encouraged.

A lot of participants, including I, initially hoped that by completing CORe, they will be given preferential treatment for MBA admissions at Harvard Business School. If you were under a similar impression then sorry to disappoint you, but that’s not true at all. As per my understanding, HBS does appreciate a participant taking initiative and completing CORe. But this is neither necessary nor sufficient to get admit to the school. Even if you have the “High Honors” grade! As you have heard more than thousand times, that MBA admissions committee looks at your application holistically. I have this feeling that the committee notices if you mention HBSOnline CORe but they are probably more interested in how your professional and personal life has changed after CORe. So, make sure you do cover that in your MBA application.

After completing the HBSOnline CORe, one can do a lot of things. I for one, used my experiences at CORe and the brand of Harvard Business School to move laterally within my company. That was one of my objectives in taking the course. You can also choose to do other courses from HBS Online. The complete list of courses offered by HBS CORe can be found at this link. There actually is a past participant discount. You can find all of the information on that here.

So, what’s the final word? I strongly endorse HBSOnline CORe and I felt it’s worth the money spent. If you manage to get your employer pay for it, that would be icing on the cake. But before you go enroll, have a plan to put it to better use than a LinkedIn badge.

That will be all from my side. If you like these story, please show your appreciation in form of claps. You can also follow me on Twitter / LinkedIn to get updates on stories like this.

Adagrad:

In Adagrad, the variable c, called cache, keeps track of the per-parameter sum of the squared gradients, which in turn is used to normalize the parameter update element-wise. Essentially what happens is that the weights receiving high gradients will have their effective learning rate reduced. On the other hand, weights that receive small or infrequent updates will have their effective learning rate increased.

Here ϵ is the smoothing term (usually takes value of 1e-6) needed to avoid division by zero. One issue with Adagrad is that in case of Deep Learning, the monotonic rate usually proves to be aggressive and leads to early learning stoppage.

RMSprop:

This highly effective method is cited through lecture 2, slide 6 of Geoff’s Hinton coursera class on Neural Networks for Machine Learning. The RMSprop update counters the aggressive and monotonically decreasing learning rate behavior of Adagrad by utilizing a moving average of squared gradients (unlike squared gradients in Adagrad).

Here γ, the decay rate, is another hyperparameter with typical value being 0.9–0.99. We notice, that the weights (x) update is similar to that of the Adagrad, however the cache c is slightly different. Thus, RMSprop still provide the learning rate of each weight based on the magnitudes of its gradients. However, by using moving average of squared gradients, it ensures that the updates do not get monotonically smaller.

Adam:

Adam update can be considered as RMSprop with momentum. In place of raw and noisy gradient vector dx, a “smooth” version of gradient m is used.

Adam often works slightly better than RMSprop and considered the default update method for practitioners. However, carefully used SGD with momentum can surpass the performance of Adam and may lead to a better local minima (see this).

The clues are:

Solutions of this edition will be posted in the next edition of this crossword series. And the solution of the second edition:

EDIT:

Since, there will not be any more editions of the crossword series, the solution of this edition is HERE.

Follow me on Twitter / LinkedIn to get updates on stories like this.

The clues are:

Solutions of this edition will be posted in the next edition of this crossword series. And the solution of the previous edition:

Follow me on Twitter / LinkedIn to get updates on stories like this.

The clues are:

Solutions will be posted in the next edition of this crossword series. Follow me on Twitter / LinkedIn to get updates on stories like this.

Boosting uses very simple classifiers as base classifiers, called “weak learners”. These weak learners are essentially decision trees with only 1 splitting rule. In Adabost, the most popular type of boosting, we start with a “weak learner” and “focus” on the samples it got wrong. In the next iteration, we train another “weak learner” that attempts to get these samples right. We achieve this by putting a larger weight on these training samples. Again, this 2nd classifier will likely get some other samples wrong, so we’d re-adjust the weights … and the cycle continues till we meet a certain performance criterion.

In a nutshell, we can summarize “Adaboost” as “adaptive” or “incremental” learning from mistakes. Eventually, we will come up with a model that has a lower bias than an individual decision tree. So, what’s the key point here — Boosting fixes bias and thus prevents under fitting.

The random forest algorithm, on the other hand is actually a bagging algorithm. In ordinary bagging algorithm we draw random bootstrap samples from our training set across the full set of features. However, in random forest algorithm, in addition to the bootstrap samples, we also draw random subsets of features for training the individual trees. Due to the random feature selection, the trees are more independent of each other compared to the ordinary bagging. This often results in better predictive performance due to better variance-bias trade-offs compared to bagging. And bagging in general, avoids over fitting too. It might be worth mentioning that random forest algorithm usually work faster than bagging as it works on smaller subset of features.