Artificial intelligence, machine learning, and deep learning all seem like fairly similar concepts. More on the point, they seem the same to an untrained eye. A lot of people see them as the same concept because, in part, they kind of are. Just as nuclear or theoretical physics are parts of physics as a whole, machine learning and deep learning are the branches of a colossal tree that is artificial intelligence. Think of them as Russian nesting dolls, each being a part of something bigger.
Artificial intelligence, machine learning, and deep learning.
The devil, as always, is in the details.
As John McCarthy, the proclaimed godfather of AI technologies put it, “Artificial intelligence is the science and engineering of making intelligent machines.” As you can see, the definition itself is rather broad. If we were to figure it out from a technical point of view, we’d be stuck for years in a philosophical debacle trying to identify what is and isn’t intelligence.
I can adapt to change. Am I intelligent, Albert?
Additional definitions of AI state that it is a branch of computer sciences revolving around a machine’s ability to simulate and emulate human-like behavioral patterns. If put simply, an intelligent machine should imitate the behavior of a human being. More to the point, AI is to perform a number of tasks that require human intelligence, including speech, visual perception, and decision-making.
If we were to look at an example of an AI-based app like, say, an e-mail verifier, we’d see that it filters through a bunch of messages and pinpoints the ones that are likely to contain SPAM or viruses. That being said, the software doesn’t actually read nor understand the text it processes. The machine focuses on finding patterns and similarities flawed inbox entries share in common. Such a program is capable of analyzing data and making a call based on a series of hand-programmed if-then scenarios. This approach to AI development is called GOFAI (Good, Old-Fashioned AI).
Obviously, there are dozens upon dozens of applications for such a machine. One can teach it to be an accountant, personal assistant, or a search bot, to name a few options. Being the most-used application of artificial intelligence on the market, these types of applications are the current representatives of the technology. They are powerful, quite useful, and incredibly potent. But these kinds of machines are all programmed by humans, leaving a lot of wiggle room for the critics of the technology. The machine did not beat a player at chess or Go. The programmers behind it did. If only there was a way to avoid this argument?
Machine learning can teach sight to robots.
Machine learning is one of the more interesting subsets of AI. Its fundamental difference from all other applications is the fact that it allows the machine to learn from experience.
Machine learning (ML) is dynamic. It allows applications to modify themselves based on the data they are obtaining in real time. In simpler words, an ML-powered AI solution does not require involvement or intervention from human beings.
This brings up the argument of a computer defeating a person in a game like the aforementioned chess or Go. One cannot program a set of skills required to outperform his own abilities. To succeed, the computer is to adapt on its own.
How do machines learn?
Obviously, there’s no school or college that graduates beings of artificial intelligence. Computers have to do everything on their own.
A machine’s learning principles stand on two staples: minimization of error and increasing the likelihood of predictions.
Both are achieved through having a clear goal—the objective function. The function is, essentially, what the program is trying to do.
Step 1: Give the machine a goal.
Step 2: Provide the correct answer.
Step 3: Feed it data.
The machine then will make guesses as to the inputs’ nature. Most of the results will be incorrect. That’s where the learning kicks in. The AI will analyze the “wrong answers” in comparison to the “correct one,” finding where it most likely went wrong.
Voila—you’ve just designed an AI program. You would still have to network it to the inputs and give control of outputs to be a neural network.
The seemingly simplistic nature of machine learning (it’s obviously much more complex than it looks) combined with the exceptional calculative power of modern processors and a nearly limitless access to the world’s largest pool of data—the internet—make machine learning-based applications into the biggest and sweetest piece of the AI pie up to date:
- Netflix alone saved more than $1 Billion on its ML algorithms that personalizes content to subscribers.
- The only reason why Amazon’s one-day shipping is available is machine learning.
- The industry is currently growing from $1.4 billion in 2016 to $59.8 billion by 2025.
For once, artificial intelligence is the next rational step in GPS navigation services. Not only can it simultaneously analyze geolocation data from a multitude of vehicles in real time, but it can also make traffic predictions. This becomes exceptionally potent once you realize that the lion’s share of automobiles is not equipped with a GPS device, making AI’s predictions based on known data patterns much more useful. If put simply, the AI doesn’t need to analyze every car in the stream to predict and help avoid a traffic jam.
There’s more: AI can prevent the jam from happening in the first place. According to this research, one driverless car increases the average speed of 14 other vehicles by 50%.
10% of AI-based cards double the number of vehicles passing through on a ramp.
Predictive tech could also predict the cost of rideshare applications, such as Uber, based on current road situations. In fact, Uber uses machine learning technologies to base its rates during surge hours on rider demand.
Video surveillance is yet another prime example of ML algorithms in action. The same principles of treating data as a set of patterns—some of which are leading to the desired outcome, while others do not—opens the doors to predictive surveillance where suspicious behavior can and will trigger a red flag.
To put things into perspective, a person standing on a bridge for a significant amount of time is not “normal behavior”. A machine seeing a person like that through the lenses of a camera can message 911 and potentially save the life of a suicide victim.
More on the matter, every “false alarm” or “true prediction” will help in training machines to make it more productive or accurate over time.
Social media giants like Facebook are using predictive technologies to personalize your newsfeed, show you the ads you’ll be interested in, and hide content that may be offensive to you as a person. The fact that the network has 2.7 billion of willing guinea pigs has helped a lot. Facebook’s algorithms have gotten so smart that it can predict your behavior better than your significant other based on as little as 300 likes. 70 likes will make the media giant better at judging your character than a friend or a roommate.
As you can see, machine learning technologies have become a breakthrough for artificial intelligence. They’ve made the software more adaptive and capable.