Software Development vs. Machine Learning

-
Artificial intelligence is a part of our daily lives. What we see on Twitter, Facebook, Google; the shows recommended to us on Netflix, Amazon Prime, and face detection are all using a form of AI called supervised machine learning (ML) i.e. you can teach a machine (computer) to learn something, it has no clue about, based on a lot of data. For example, you can show a machine thousands of images and train it to detect cats. After being sufficiently trained, you can input an image, which the machine has not seen before, into the machine and it would tell you if there is a cat in the image. The more data the ML models get exposed to, the better they get over time. 

Most of the practical applications we see today are using ML. And, everything ML is doing can be done by humans. So, why do we need ML? Three main reasons: 

1. Automation: You don't need to hire humans to figure out if there is a cat in the image or if an email is spam. Machines can do a good enough job. This can result in huge cost savings. 

2. Scale: Since you don't need to hire humans, train them, and retrain them, you can scale your product/business very fast. For example, you can go from classifying 10,000 emails per hour to classifying 1M emails per hour in seconds. Hiring people to handle 100x (1M/10k) workload can take years. Furthermore, there is just so much data now that it is hard for humans to fathom. For example, there are over one trillion pictures taken every year. 

3. Speed: Machines can do the jobs faster than humans can. 

When do we need ML? It is well suited for problems for which writing logic based rules is difficult. We have become used to logic based rules with software development over the last ~70-80 years. However, pure logic based rules are not well suited for all problems. For example, how do you write a software program to accurately detect a cat in an image. It would be really hard and not very accurate. Software can help find answers and ML can help you make rules that can predict answers. However, you have to know the answers and train the machine before you can use ML to predict answers. E.g. you have to know what a cat looks like before you can teach it to the machine and then the machine makes its own rule, based on various cat properties, on how to detect a cat. 

Interestingly, humans don't always understand the rules the machines come up with. 

The following image shows the fundamental difference between software development and ML: 


Software vs. ML



ML requires a different mindset, heuristics, and humility. As the world is getting digitized, we will see more and more applications of ML and other forms of AI. 

Popular posts from this blog

Obituary: Charles T. Munger

-
Image
My hero, Charlie Munger , passed away on November 28th, 2023. He was 99 years old.  I met Munger in 2007 at my first Berkshire Hathaway shareholder meeting . It was admiration at first few words. I liked his depth of understanding in multiple domains, his honesty, and his willingness to share his wisdom. Since then I have read all his available writings and seen him many times in Omaha or Los Angeles at various shareholder meetings. He profoundly influenced Warren Buffett and the entire investing and business world. The world has lost a great leader, thinker, and teacher.  Munger's response to my letter      Since meeting Munger, I continue to learn from him and share what I am learning.  Following are a few Munger teachings that have become my habits:  1. Inversion: When you want to solve a problem, start with inverting the problem. For example, if you have to figure out how to keep the airplanes safe, work on figuring out how they can crash.  2. Multi-disciplinary thinking:
Read more

Systems Thinking as taught by Ackoff

-
One of my heroes is Dr. Russell Ackoff . I have read a few books he has written and have learned Systems Thinking from him. I am surprised that the field of Systems Thinking is not well understood. Following is my attempt to share what I learned from one of Ackoff's recent lectures. Albert Einstein once said, "You can't solve a problem with the same mind that created it." According to Dr. Russell Ackoff most managers agree with Einstein's statement but not many know what it means. It is easy to agree with something whose meaning is vague. In the Renaissance era, when the science as we know it today was born, a scientific inquiry method called Analysis was developed. Analysis comes naturally to us. Just watch kids breaking new things and being curious about the parts. The understanding of something follows a three step process in analytical thinking: 1. Take it apart 2. Understand (function, role, behavior) what the parts do 3. Assemble the understanding of the pa
Read more

Dynamics Of The Technology Ecosystem

-
Image
Context  The last two decades have completely changed the technology (tech) sector. Cloud computing , smartphones, apps, and high wireless data speeds are taken for granted. These developments have created a dynamic tech ecosystem.  Figure 1: Technology Ecosystem abstraction with examples in each layer  Having worked with some very smart people to develop and market new products in all parts of the tech ecosystem - Networks and mobile phones ( SIEMENS ), Network Operators ( T-Mobile ), Semiconductors ( National Semiconductor ), Algorithms/Artificial Intelligence ( Audience ), Cloud-based Applications ( EVER , Moki ), Blockchain and Consumer Hardware ( Open Garden ) - gives me a unique perspective on how the technology ecosystem works and how it might evolve.  In this article, I share my view on what constitutes the technology ecosystem, what are the primary business drivers depending on a company's position in the ecosystem, and how the ecosystem might ch
Read more