Boy, where does one start learning AI? I’m struggling a little bit with this one. I’m enough of a nerd that I really don’t learn something until I figure out how it works, which with AI seems like a pretty large task. Whether or not it truly is remains to be seen. But as I put more thought into it, I realize that I don’t really even have a terribly good grasp of what AI is. I’m versed enough in the tech world to know that when we talk about AI, we aren’t talking about “Skynet”, or “iRobot”, or any of those Hollywood scenarios where AI decides we humans aren’t worth the trouble (At least, not yet. More on that to come later). But, then what are we talking about? And how can it help my business, or those of my customers? What does it actually do?
In an effort to answer this question I started talking with a few folks in the industry who are further down this path, and with a little bit of guidance and direction began uncovering some basic “what is AI” courses available online and got started. (Links to the courses I took can be found at the bottom of this post) Here are some high level basics that I’ve learned so far. As I’ll attempt to always do, links to related or referenced articles or courses are provided at the bottom.
Table of Contents
What is AI?
The actual definition of AI that seemed the most straightforward to me was:
Artificial Intelligence (AI) is the theory and discipline of programming computer systems to learn from and spot patterns in data sets.
https://www.coursera.org/articles/how-does-ai-work
Wait…that doesn’t say anything about AI becoming my new robot best friend, driving my car, or eventually controlling the world. And yet, at this point in time, this is what AI is. From the article “How to Learn AI From Scratch in 2024” on Datacamp, there are three main types of AI:
Artificial Narrow Intelligence (ANI)
This is the most common form of AI we interact with today. ANI is designed to perform a single task, like voice recognition or recommendations on streaming services. Examples would include many voice activated tools like Siri, or Amazon Alexa. These tools have been around for a while (actually much longer than you might believe…think 1960’s) and are pretty commonly found throughout our day to day lives.
Artificial General Intelligence (AGI)
An AI with AGI possesses the ability to understand, learn, adapt, and implement knowledge across a wide range of tasks at a human level. Large language models like ChatGPT are pushing this level into new boundaries and have shown the ability to generalize and perform many “cognitive” tasks. With that said, according to the article this is still (as of late 2023) still a theoretical concept. While I may not completely agree with that sentiment, how they got there makes more sense the more you understand how these AI tools work.
Artificial Super Intelligence (ASI)
Yes! This is the one everyone is waiting for…when AI outsmarts us and eventually takes over the world. (Well, maybe we aren’t waiting for that part.) ASI is the final level of AI and refers to a scenario where AI surpasses human intelligence in nearly all valuable work. This sounds pretty cool, and a lot of people assume this is either where we currently are with AI or where we will be soon, but in reality this is many, many years away and whether or not it’s even possible still remains largely speculative.
It’s also worth pointing out that at this point, there isn’t even a consensus on the different levels of AI. Google’s DeepMind team is currently trying to change that with a worldwide push for standardization of the different levels of AI. DeepMind is pushing for a more granular six levels of AI:
So as you can see, while AI is certainly here and relevant already in today’s world, it isn’t quite the science fiction based, self-conscious, super-sentient being that many believe it is. It’s data. REALLY, REALLY large amounts of data and the processing power to quickly churn through that data, combined with mathematics. (More on this as we go forward in future posts.)
Ok, so I’m not scared of killer robots anymore, but I still don’t understand what AI is?
For any of you developers out there, a good, simple way of understanding AI as it is today is to compare it back to traditional programming. With traditional programming, we usually begin with a set of data and then program processing paths or routes for that data to travel, with the goal of arriving at an expected outcome. AI achieves this in a different way. Utilizing AI, we begin with a set of data (inputs) and another set of resulting data (outputs). What happens in the middle though, that’s the unknown. Contrary to traditional programming, AI takes that input and the associated output data and uses mathematics to figure out all of the possible pathways (called “neural networks”, there’s another key AI term for you) to get from the input data to the appropriate output results. This, in a simplified nutshell, is the act of training an AI model. An AI “agent” would then utilize that model to take future input data, run it through the neural networks that have been trained for that model, and provide the output.
Here’s the kicker though, and one of the primary benefits. Because AI is effectively teaching itself the pathways to process data, and because it’s relying on processors with computational power that we, as humans, simply can’t match, the AI models can handle much larger and much more granular sets of data than our brains could reasonably process.
Let’s go over an example
Here’s an example taken from the Google AI for Anyone course on Edx. (link below) An AI model was trained with large sets of data consisting of ocular scans, like the ones you get during your checkup at the eye doctor. Let’s say we take two of those scan images, one looking “normal” and one with some obvious abnormalities or health issues. Show those images, side by side, to the average person and they’ll fairly quickly be able to tell you which eye has a problem, but they won’t know what that problem is and will only have a high level and vague understanding of the issues within the image that led them there, say, dark spots maybe.
Show those same two images to an ophthalmologist, and they’ll quickly be able to tell you not only which scan has the issue, but most likely what the issue is as well and why they think that. But if you ask the ophthalmologist how old the patient is for either of the images, they will have no idea. This particular AI model, on the other hand, has analyzed the data and mapped neural networks for every single possible pixel on every image within the datasets and then mapped that data to output values, one of which is the patient’s age. The end result? The AI agent (another term we’ll discuss deeper in another post) can tell you within 3 years the age of the patient simply based on the scan of the interior eye. Now that’s pretty cool.
So let’s bring together the point of this first post. One thing I’ve learned early on in this process is that it is just as important, sometimes maybe more important, to understand what AI isn’t as much as understanding what it is. Now that we’ve determined that what we’re working with today isn’t scary and isn’t going to turn on us “for our own benefit”, we’ll spend some time in the next post breaking down the AI that exists today.
Resources
Here are some courses and articles that I thought were worth the time and/or cost. Some of these are super high level, some dive a little deeper, some are free, and some are not. Many of these sites will let you audit courses for free, but if you want to take the graded components and get a certificate of completion there may be a fee involved.
Please Note
This blog is all about pointing you to content I’ve found useful. While it doesn’t apply to everything, I do use affiliate links for some of these resources and may be paid a small amount if you decide to purchase any of these courses or programs from my links. That’s not the goal, and by all means please get as much free knowledge as you can! But if you are going to purchase any of these, it does help me out with general hosting fees and related costs for this site if you would be so kind as to purchase from these links. Thanks!