- What Artificial Intelligence Really Is
- The Differences Between AI and ML
- How to Start Experimenting with AI and ML
- What Machine Learning Engineers Actually Do
Staying up to date with important news and technological discoveries can be really challenging--- especially when so many people create interesting books, blogs, and podcasts. I personally have a long "to read" list, and I am sure I won't have enough time in my entire life to finish it. If only someone could provide you with all that information in an easy-to-swallow pill in one read. That would be something, and that is what I aim to do in this AI overview.
This is not another technical blog post with confusing words readable only by those fortunate enough to call themselves "developers." As a business professional, I need to know how things work and what value they can bring to my customers. While I can't speak to the latter--- your specific customer needs--- I can cover the former. Like Morpheus in The Matrix said, "I can only show you the door."
Let's explore what Artificial Intelligence is and what it is not. Furthermore, I'll show you helpful links that will allow you to go down the rabbit hole and expand your knowledge.
Think of this post as an introductory guide.
What Artificial Intelligence Really Is
To get started, we have to realize where we are in terms of software development and how that is reflected in usability. As you probably noticed, emerging startups have SaaS (software as a service) solutions concerning almost every aspect of our lives (online wallets, food delivery, utilities, games, portfolio presentation, etc.). We can build such apps because we have reached a stage where we can digitize a lot of information. Cameras, laptops, smartphones, wearable devices, everything around us contain sensors that collect loads and loads of data.
So we have rows and columns of databases filled with data, but what happens next?
Following Edwards Deming's wise words:
"In God we trust; all others bring data."
We would like to use data to perform better--- be more efficient, more productive, work less, and have the same outcomes as previous generations. The problem is, we as human beings are limited in our cognitive abilities and time to analyze all the information that we can generate.
This is the moment where automated systems that can learn and analyze the changes in data come into perspective, which we call machine learning. Basically, we feed those systems data and wait until it gets back to us with results.
What do those systems provide?
These applications behave and act like humans. They read, analyze and try to find some relationships from which you can create a rule. However, as presented in the book "Thinking Fast And Slow" by Daniel Kahneman, unlike human minds, they do not have feelings or experience cognitive errors. The only thing that affects their performance is the quality of the data, which is an important topic.
The Differences Between AI and ML
As mentioned previously, Machine Learning (ML) is a subset of the Artificial Intelligence technological landscape. AI encapsulates hardware and software designed to make robots' capabilities as close to human intelligence as possible, simultaneously making them more efficient and precise; no emotions, just digits and boolean algebra.
We classify AI into three categories:
Artificial Narrow Intelligence (ANI)
Artificial intelligence specialized in a certain area. That's where we're at now in the world. Shape/object recognition, voice recognition, and text to speech engines.
Artificial General Intelligence (AGI)
Artificial intelligence at the level of human intelligence. AGI is not available yet. Planned for 2040.
Artificial Superintelligence (ASI)
An artificial intelligence that is smarter than humans with an infinite barrier for development. Planned for 2060. Skeptics say it will look like the Terminator movie.
Machine Learning is all about learning techniques. We divide it into three types:
This model is trained using labeled data; we have a dataset of inputs and outputs to teach our machine. The goal is to help our model find a mapping equation to correctly assign a new input to known outputs. Examples of usage include risk evaluation, sales forecasting, image detection, etc.
Limited to inputs, which we aim to classify by the outputs they provide and features they possess (clustering). We do not have to help our model, so it's unsupervised. As such, the machine can assign a new input to the right class with similar features or outputs. Application: recommendation apps, anomaly detection, customer segmentation, targeted marketing, etc.
Here we have no data. Wait...no data at all?
I am joking, of course. We have no information about input and output (no predefined data), but we have an environment. It is a certain set of boundary conditions or/and rules. The goal is to find a way to maximize profit (you thought about the stock market at first, didn't you?). The model uses the oldest learning technique, trial and error, and gets feedback about its past outcomes to apply a new strategy. It is used for self-driving cars, healthcare, vacuum cleaners, and games.
Some games have machine learning that uses this learning method to understand how you play to increase the difficulty level and win. Obviously, chess also has ML, but you have to choose your opponent's skill level, whereas in the example above, the difficulty increases automatically.
Examples of AI in Business Cases
That's it for the theory. Let's get to business!
Below I listed some daily routine activities that most of us don't realize are amplified by ML/AI.
- Recommendation Apps (YouTube, Facebook, Zalando, etc.)
- Face and smile recognition when you take a picture using your brand new smartphone
- Web search engines - Google and Yahoo: somehow they know what you are asking although you've just put two words, don't they?
- Lane assist in your brand new car
- Filtering and sorting goods on your favorite eCommerce site. You rarely see something irrelevant to your shopping desires, right?
There are plenty of usage scenarios currently, and there will be even more as time goes on because every day, we leave more and more data about us on the internet, social media, and mobile apps.
How to Start Experimenting with AI and ML
- Do your own AI experiments
- Great series from google about Machine Learning
- MIT Media Lab Open Learning courses
Here's a list of three sources from where you can find more information about these topics.
What Machine Learning Engineers Actually Do
In a nutshell, Machine Learning engineers take work results from data scientists to implement them, using proper infrastructure, models, algorithms, and machines. This position requires the right skills and experience, which depends on what you ultimately need to use machine learning for.
You are probably thinking, "How can I find someone like that for my project?" Easy. You can just email me, and I will help you find a great match for your next project.