Sometimes WHEN I talk about AI with my friends, I find that many people overestimate or underestimate AI. Some problems don’t need AI at all, and some problems can’t be solved even with AI.
So I decided to write a series that evaluated a question from all angles: “Should I use AI? Can AI solve my problem?”
Evaluation perspective: Characteristics?
The Angle that this article cuts into is: characteristic
Still remember my junior high school teacher told us a knowledge point of evolution: Russia is cold all the year round, Russian nose evolution is very long, so that the air into the body needs to walk a longer way in the nose, it will not be too cold. So russians have one obvious feature: long noses!
But not all people with big noses are Russians, africans have big noses too!
So, if we want to determine whether it’s Russian, we need more characteristics (evidence) :
- The nose is long
- tall
- Blue eyes
- White skin
- The eye socket is deep
- Body hair developed
When we find that a person has all of these characteristics, then the probability of that person being Russian is much higher.
- The man can speak Russian fluently
When we find the above features (evidence), it is almost certain that the person is Russian. Because this feature is too strong, or too convincing.
Interstitial – basic principle of artificial intelligence
Review the process above:
When we see a lot of Russians and people from other countries, we will sum up the characteristics of Russians based on experience: long nose, high character, blue eyes, white skin, deep eye socket, well-shaven hair, speaking Russian…
When we meet a foreigner we have never met before, we apply this set of “experiences” to the person to see if they match, and if they match many of the characteristics, we assume that the person is Russian.
The principle of artificial intelligence is basically the above process, as shown below:
Feature quadrants
But not every problem needs TO be solved by AI, and AI has the advantage of being able to deal with a huge number of features, not only surface features, but also hidden features. But in many cases, it’s not necessary to shoot mosquitoes with a cannon.
When we plot a coordinate between the number of features and determinism, it guides us to what problems are appropriate for AI and what are not:
Few features + weak determinism: suitable for manual solution
Fewer features + strong determinism: suitable for regular solution
Multiple features + strong determinism: suitable for regular solution
Multiple features + weak determinism: AI solution “can be considered”
PS: There are many other factors: cost, risk, measurability… I’m not going to consider anything here, because it’s too complicated.
Case description
The upper quadrant is too abstract. Let’s take a real case to illustrate.
The game industry has a thing called “plug-in”, simply said plug-in is a cheater, broke the fair game, let oneself have more advantages in the game.
Almost all well-known games have plug-ins, because plug-ins make money! So game manufacturers must be prepared to fight with plug-ins.
Game manufacturers hope to discover players using plugins for the first time.
I have a friend who works on a well-known fighting mobile game that also has a lot of plug-ins, and at first they used some fixed rules to find plug-ins. The results are good, but there are still some loopholes.
So they tried to use AI to catch plugins, and after a long time they found that the results were not much better than fixed rules.
The same is the fight against plug-ins, in some of the high complexity, high flexibility of the game (such as eat chicken, CS), the rules are not good, because it is difficult to sum up the fixed rules.
This is where the AI comes into its own. CS: There are some successful cases of GO and eating chicken:
Using Machine Learning to Catch the Dog
Ai has these applications in games. Let’s see how many do you know?
Application feature quadrants:
In mobile fighting games, the degree of freedom is not high. Players can only control movement, attack, skill and evasion, which are the four core operations. Strategy is not too strong, so roughly in line with the logic of “spend money + spend time ≈ strength”.
So there’s only one rule to follow: Is the player doing more than they can do?
By observing the player’s combat effectiveness, enemy difficulty, combat time and other ways can be more effective to judge whether the player is using plug-ins. For such a problem that even a human can judge effectively, AI is not necessary, but complicates the solution.
Let us see eat chicken to have what wonderful job add-on first: “eat chicken add-on daqo, let you see through the other party direct report!”
For shooters like CS: GO and Eat Chicken, the setting is very complex (big map, rooms, shielding…).
Player behavior is also complex (move, judge enemy positions, take cover, switch weapons, aim, shoot…).
In this case, it is difficult to determine whether a player is using a plugin with clear rules. You can’t say fast even with a fluoroscopic plug-in, some people are fast; Well, you can’t just say a lot of headshots make you use a headlock, because some people are good shots.
Finding these plug-ins requires analyzing large amounts of data and picking out “less obvious features”, which is where AI has special value.
conclusion
Today we explained which problems are appropriate for AI and which are not, from a “characteristic” perspective.
If we can sum it up in a word:
Problems that can effectively generalize some rules do not need AI, while problems that are difficult to generalize rules can be solved by USING AI.
If you want to evaluate, you can apply the following feature quadrants to see if your problem is appropriate to use AI technology:
In addition to the “feature” Angle, there are many angles that can help us decide: should WE use AI? This series will continue to be updated, follow my official account to see all the content:
Easyai. tech is the AI learning library for product managers