Focus on AI — 1: What is AI?
Are AI technologies being used right now? And how? A robot will do my job in the future?
I am studying AI right now to continue my personal and professional growth path, and I am beginning to understand that I have never understood a lot of things when talking about AI: sometimes I realize that movies or books are to blame; on the other hand, the fault lies simply withmass information that points more to sensationalist headlines rather than making real information.
So I’ll try in a series of articles to explain better what is AI and it’s uses, if only to have a better understanding of the subject.
Let’s get into it!
1. How should be AI defined?
AI always peaked my interest, now more than ever as I’m working in the IT industry.
Currently, AI is a hot topic: media coverage and generic discussion is basically everywhere. However, there’s something that it’s important to notice: AI means different things depending on who you are talking to.
For some, AI is about artificial life-forms that will surpass humans; for others instead, any data processing technology can be called AI.
That said, let’s start simple: I’ll cover what AI is, how we can define it and which are closely related fields.
There are some example to illustrate which common AI aspects are: we will in future deepen our understanding on each subject.
a. Content recommendation
Pratically every information we come across is customized based on our web beheaviour. Think about Facebook ads or pages, Instagram feed and other social media content; online advertisement; Spotify or Netflix suggestions. And so on.
While the frontpage of a classic newspaper is the same for everyone in the globe, the frontpage of online version are different for each user.
The alghorithms that determine the content are based on AI.
As positive as it may seem (“I only see what interests me”), they could create situations not exactly positive: think of echo-chambers (an environment where a person only encounters information or opinions that reflect and reinforce their own), troll-factories (entities conducting disinformation propaganda activities) and fake news (the proliferation of false information on the web).
So it’s cool that Spotify introduces me to other artists of my favorite genre, but we must be aware of the implication of such technologies.
b. Self-driving cars
Self-driving cars are a hot topic as well. They require a combination of different types of AI: the search of the most convenient route from point A to point B, computer vision to identify any obstacles, decision making when moving in a mutable environment like a city.
Obviously, each one of these must work with a state-of-art precision to avoid grave accidents.
But we’re speaking the same for delivery robots (such as the ones used in Amazon warehouses) or flying drones.
c. Image processing
Face recognition is already a very used technology: think of automatic photo tagging, passport control, smartphones face unlock or payment management.
Similar technologies can be used to recognize obstacles around autonomous car or they can even be used to estimate wildlife populations.
But we already know that this technology could be used to generate or alter visual content: a lot of filters can adapt your personal photos to look like Lord of the Rings characters; videogames and movies can use motion capture to precisely replicate human movement; but it can also create deepfakes.
We’re going through the road that it will be almost impossible to distinguish a real footage from a fake one.
2. So, what is an AI? And what is not?
We already said that the public perception of AI is confused. People have started using the term AI when referring to things that used to have other names, even when talking about simple statistics or if-then-else rules.
But why is that?
a. No official definition
First of all, there’s no official definition: AI researchers are not capable of doing so because, with every passing hour, some topics are classified as non-AI and new topics emerge.
In 1970’s, for instance, automatic methods for search and planning were considered AI but nowadays these same methods are taught in school.
b. Books and movies depiction
Next, we can’t forget the various literary and cinematic works of science fiction that have influenced us over the years. When talking about robots and AI, we could think about Terminator or C-3PO, D&D’s warforgeds, Isaac Asimov Robot Series, “Becentennial Man” with Robin Williams or Quantic Dreams’ “Detroit: Become Human” videogame, and many more.
Mankind has always been fascinated by the concept of robots and the creation of intelligent life forms, and so the artistic work about it is enormous.
c. Easy tasks made difficult
Another source of misunderstanding is the difficulty to get which task is easy or not for an AI: one of the easiest tasks for us and the one we do most often is picking up any object.
It’s often an automatic gesture for us but let’s think about it more thoroughly: we first searched for the item we need with our eyes; than we “calculated” the trajectory for our hand/arm to reach it; we, in the end, moved our hand (and a lot of muscles in the meantime) to grab it and to keep it.
It means that, when speaking about robotic arms for instance, the researcher must scan the sorroundings for the item, doing calculations to reach the item, materially reach the object and applying just the right amount of force to keep it without brake it.
We have always to consider that these tasks are pretty much effortless for us because we’re used to complete them day after day (and with centuries of evolution behind our back).
d. Difficult tasks made easy
Ironically, for an AI is more difficult to pick and grab an apple rather than play chess.
The initial AI researches concentrated on mathematical task because they require a great effort for us, with years of practice and a lot of concentration.
It turns out that playing chess is very easy for computers, which can be programmed to follow simple rules and can compute a lot of alternatives at incredible speed, unthinkable for humans.
A famouse example is the Deep Blue vs Kasparov matches in 1997: the russian Garry Kasparov is a chess Grandmaster with a record of 255 months as a №1 rank. At that time he was World Champion, with already lots of records, but was beaten by the IBM’s Deep Blue computer.
3. At last, can we get a definition?
Before getting there, more useful could be a list of properties that are characteristic to AI:
Autonomy: the ability to perform tasks in complex environments without constant guidance by a user;
Adaptivity: the ability to improve performance by learning from experience.
When we talk about AI, we have to be cautios as many of the words that we use can be misleading: think of learning, understanding, intelligence.
If a system delivers precise navigation instructions, is it truly intelligent? Aren’t we suggesting that these systems could perform every task which an intelligent person could do?
Likewise, when MRIs artificial intelligence detects breast cancer, can we say that it understands the subject? Think of a computer vision system for self-driving cars: the car “understands” that it must be steady on the road, but are we sure that it will understand that is dangerous/wrong to drive over a person wearing a t-shirt with a road print on it?
Obviously, we would be wrong, because we’re using the so-called “suitcase words”, term conied by Marvin Minsky, a cognitive scientis and one of the pioneers in AI.
We have to realize that intelligence is not a single dimension measurable value: we could easily tell if in Bari the temperature is higher than in Lammi for a specific day.
But when we’re talking about AI, is a chess playing system more intelligent that a sentiment analysis algorithm? These questions make no sense in this environment: the capability of solve one problem does not imply the ability to solve another.
4. A pinch of AI
For everything we said above, it’s impossible to have a white-black classification of AI vs non-AI: while some methods are crearly AI and other definitely not, there are also methods that use “a pinch of AI”.
We also have to understand that AI is not a countable noun, as it is a scientific discipline (such as biology or computer science): so AI is a collection of concepts. It’s just wrong to say one AI, two AIs, and so on.
But, as we said at the beginning of the article, even this specific subject is misinterpreted by mass information: “Data form wearables helped teach an AI to spot signs of diabetes” or, worst, “Google’s artificial intelligence buil an AI that outperforms any made by humans”, are just wrong headlines.
To sound more expert, you can try use “an AI method”, which is also more professional.
Let’s go back at the definitions we said earlier, Autonomy and Adaptivity and let’s see some examples of what we could considere AI or non-AI, keeping in mind that sometimes the borders are very blurry.
5. First part conclusion
So, this is only the first part of this voyage. I’m learning more and more about AI, it’s functionality and ethics, hopefully to learn how to use it and programming it.
These series of articles are not be considered as a dogmatic truth, as I’m at the beginning of my path and maybe these a more useful to me to better grasp the subject.
That said, feel free to contact me, send me feedbacks or shout outs!
