The article includes the following parts:
AI/ML: The Confusing Context
Often interchangeable terminology used within marketing and technological gigs is Artificial Intelligence (AI) and Machine Learning (ML). There are relevance and relation between AI and ML; however, AI is a tree, and ML can be considered as a branch of AI. The significant usage of MI/AL within the marketing sector has created a lot of buzz and confusion at the same time, wherein in some cases even the software services company are not sure if they are offering MI or AL within their product.
Artificial Intelligence (AI)
A piece of code running on the machine that thinks and decides as per human intelligence based on receiving inputs falls under the definition of AI. Laymen often refer to Terminator or Data from Star Trek as AI, which is a specific form known as Artificial General Intelligence (AIG) that matches or exceeds human intelligence against any given matrices. Thinking and Acting are the primary two approaches that define AI. AI usually falls under two categories, and each is having sub-categories based on capabilities and functionality.
Type-1 AI is capabilities-based.
- Narrow or Weak AI is the current generation of AI that we usually interact with on a day-to-day basis. Apple’s Siri, Chess, E-Commerce purchasing suggestion, and similar applications fall under Narrow AI.
- General AI is where the machine performs intellectual tasks with efficiency as humans. Researches are underway to achieve General AI.
- Super AI will be the machines that could surpass human intelligence and is still a theoretical concept.
Type-2 AI is functionality-based.
- Reactive Machines are the most basic type of AI which gathers current statistics and provides the best possible output accordingly.
- Limited Memory stores experience for a short period and provide relevant outcomes, therefore. A good example would be self-driving cars.
- Theory of Mind is not developed yet; however, in arguments, these understand people and emotions and decide its actions as required.
- Self-Awareness hypothetically will be self-conscious and self-aware machines that will be smarter than the human mind.
As per Stuart Russell and Peter Norvig in “Artificial Intelligence: A Modern Approach”, there are four different approaches that define AI.
- Thinking humanly
- Thinking rationally
- Acting humanly
- Acting rationally
AI plays with raw data and depicts the output as per the processed data. Each AI implementation requires a continuous feed of raw data and acts accordingly. Google’s AI-Powered maps, commercial planes AI Autopilot, Email Spam Filters, Grading and Assessment tools, Banking Fraud Prevention, Facebook facial recognition. These are some of the common examples of how AI has infiltrated ordinary day-to-day activities.
Almost every primary online shopping application like Amazon analyzes the user’s social trend, search history, and there are theories these AI bots even listen to the user’s voice and show up the ad’s accordingly.
AI Sample Code
Despite theoretical content, the base of AI is the data, its analysis, and the output. Python leads the development when it comes to AI. The Below sample code shows how AI interprets the sample data and provides output post applying the Binarization technique.
// Importing the useful packages import NumPy as np import sklearn.preprocessing //Defining sample data Input_data = np.array([2.1, -1.9, 5.5], [-1.5, 2.4, 3.5], [0.5, -7.9, 5.6], [5.9, 2.3, -5.8]) //Applying Binarization technique: This technique is used to convert numerical values into Boolean values. data_binarized = preprocessing.Binarizer(threshold = 0.5).transform(input_data) print("\nBinarized data:\n", data_binarized) //OUTPUT Binarized data [[ 1. 0. 1.] [ 0. 1. 1.] [ 0. 0. 1.] [ 1. 1. 0.]]
Machine Learning (ML)
In a layman’s term, ML starts by getting it wrong then with experience; eventually, it produces appropriate output. ML is referred to as an approach that implements AI. The software application evolves as more and more data is passed through it, and the ML algorithm predicts the outputs based on historical data. Arthur Samuel defines ML as a “Field of study that gives computers the capability to learn without being explicitly programmed.”
ML algorithms differ from traditional programming as it gets feeds from data, the data gets trained, and the machine enhances itself to interpret the data for the next cycle better. ML integration in day-to-day usage looks like.
- Voice assistant like Alexa, Google Now, Siri, which is implemented using Neuro-linguistic programming (NLP).
- Predictive Machines that Google Maps implements to estimate routes based on traffic.
- Video surveillance.
- Social media services.
- Search engine result refining.
With the advent of online payments and non-cash transaction, a lot of hackers attempts to get through the financial details of customers. However, these are prevented by Fraud Detection Systems to ensure secure cyberspace by pattern recognition that distinguishes between illegitimate and legitimate transactions.
There are Experiences that are taken into consideration against any Task. The Performance increases as more and more data get fed into the system. As per definition, a program falls under ML if it improves its Performance P from Experience E over a Task T.
ML has four primary categories based on learning experiences.
When a model gets trained with a Labelled dataset, the resultant learning is called supervised. An example could be like giving a child 100 stuffed toys that have ten animals of each kind. The child learns each toy one by one by its characteristics, color, and shape. As he progresses, the next instance, when a toy repeats, the kid would know what it is.
Unsupervised learning is the opposite of supervised learning, where the learning machine does not have a labeled dataset. The system should be smart enough to learn on its own.
A mix of supervised and unsupervised as some data columns are labeled while others are not. Google photos are a good example where when we mark a person with a label, it learns through its face recognition ability and traces the picture of the same person from the remaining gallery.
Reinforcement learning has less human involvement, as the machine gets exposed to a scenario where it has to act accordingly as per the processed data. With each action, there is either a reward or penalty, resulting in the system updating its policies. This cycle continues until the most optimal action against the situation is available. Walking Robots are a primary example of using Reinforcement Learning. When a robot takes the wrong step, it updates its policies and avoids the wrong step for the next time.
ML represents a form of AI where ML learns and evolves in the best possible manner to achieve Narrow AI to categorize data, recognizing faces, detecting frauds, or predicting the future based on the available datasets. The goal of researchers working on AI / ML is to attain technology that learns and effectively enhances itself to ensure the predictions and outcomes ultimately help the end-users with the least possible failures.