Where Are They?
Where are they, where are the AI assistants like JARVIS, or where are the droids like C3PO or R2D2?
If you are a sci-fi movie or series enthusiast like me there is a probability that you at least once in your life ask yourself the same question.
It has been years that sci-fi movies and tv-show have been bombarding us with the promise of having one day AI with human-like intelligence roaming the street and living in harmony with us. But none of those promises seem to come to a realization.
I acknowledge that we have made a good part of the achievement toward this utopian dream of mine as a matter of fact we have Amazon Alexa as a simple substitution for Jarvis and robots have begun to appear more in normal life , and not only for industrial use.
But given the many articles praising the revolutionary advancement and improvement in the field of AI. I would have thought we would have been more technologically advenced to make those proposals true. To characterize the feeling I am having right now I would tell you that it feel like when you are installing a new app on your laptop and the installment goes super fast for the first 90% but the last 10% take forever to finish.
But fortunately, as I a beginning to see more self-driving car in the street all around the world I begin to have hope again because I am sure that if we achieved the creation of a level 5 autonomous car will we have the key to achieve the implementation of those long awaited promise.
Depending on how much a self-driving car is autonomus we assign is a number from a scale that begin at 0 wich is it is not autonomus and 5 wich is that it is fully autonomus.
The field of AI is a subfield of computer science that tries to answer the fundamental question that is “can AI have human intelligence?”. It is found on the concept that human intelligence can be so precisely described that even a machine would be able to simulate it.
This broad field can most of the time be divided into two categories the first one being narrow AI and the second one being Artificial general intelligence.
Let’s begin by diving into the field of narrow AI it would serve us as a base and will help us to better understand Artificial general intelligence.
Narrow AI:
This type of AI is usually called weak AI and is described as a type of artificial intelligence that produces a limited simulation of human intelligence it is focused on performing a single task well. A narrow AI can seem to be intelligent but it operates under more limitations than even the most basic form of intelligence an human can have.
Most of the time when we are referencing the narrow AI field we are either talking about one of its two most popular subfield machine learning and deep learning
Machine learning consists of feeding a lot of data to an algorithm to allow it to learn from those data and give a prediction for new unseen data based on the insight that it had on the previously seen data. To summarise the goal is to train a machine learning model which is the AI algorithm on knows input so that it can predict never seen input.
On the other hand deep learning tries to mimic how our brain works by running input through a biologically inspired neural network and allow them to learn from data with algorithms that mimic the way the brain process information.
The term Artificial intelligence, machine learning, and deep learning are often used interdependently and this may cause some confusion to some people. To help better understand those 3 terms you can look at the picture below, as you can see ML is a subfield of AI and Deep Learning is a subfield of machine learning.
Even though there is no clear path for doing a Machine learning project we can generalize the step that someone would take to solve a problem with machine learning as defining the problem\gathering data, fitting the model, evaluate the model\predict.
Defining the problem and gathering data.
Most of the time you begin a machine learning project by determining what is the problem you are trying to solve and gathering the data that will be needed to solve it.
When defining the problem you will need to choose what type of machine learning system you will use this allows you to know what type of data you will need to gather.
There is mainly two types of machine learning system you can use is can either be a supervised system or an unsupervised system
- For a supervised system, you will need label data this means that the desired solution is in the data set that will be used to fitting the model.
- For an unsupervised system as the name might suggest the data is unlabeled so you do not have the desired output in the data that will be use to fit the model.
Fitting model:
The second step consists of fitting a machine learning model to the data that you gathered in the first step.
Depending on what type of task you are doing the algorithm or model you are going to use will vary but there is 3 main type of algorithm that you can use for the task which are regression algorithm, classification algorithm, and clustering algorithm.
- Regression algorithms are used to predict an continuous quantity or value for exemple the price of an home.
- Classification alogrithms are used to predict an discrete class label for exemple tell if an email is spam or not.
- Clustering algorithms is used to find a natural grouping in unlabel data this is why it is usually use in unsuperviesd task.
Beware of under and overfitting.
Sometime when you are fit your model you may evaluate to closely to your fitting data this might seem contradictory most of the time when something is really good we should be happy about but you have to keep in mind that you do not want the model to predict the testing data but you want it to predict new and never seen data.
But you also have to take care of it not fitting well on the data that you should strive to keep a balance between over and underfitting.
Evaluate model.
After fitting the data you need to know how well your machine learning model did.
This is why we usually split your dataset into to different part one for fitting the model and one for evaluating how well you model actually did at predicting the output.
Predicting the result.
A last but none least this is the step that you have been waiting for with the model behind trained to predict from unseen new data.
now that we have covered all the narrow AI we can get into what I considered as the fun and exciting stuff artificial narrow intelligence also know as AGI.
Artificial general intelligence.
Artificial general intelligence is the type of ai that we see in the movie it is also the type of ai we are referencing when we are talking about rd2d and c3p0 those AI, unlike narrow Ai the imitate only a small unit of human intelligence have a more general type of intelligence and it has the capacity to apply intelligence to solve any problem that same way that a human will do.
As of now this type of ai is only hypothetical. But there a possibility that autonomous cars could be the key to attaining this type of AI.
How solving the autonomous car problem can be the key to the AGI door.
Autonomous cars have been starting to appear in some of the techiest cities in the world like silicon valley, Toronto, and much more.
Unfortunately, I do not think that people know how big of a deal this is in fact or they are excited for the wrong reason. In fact, most people when talking about autonomous cars will be more excited about not having to drive to work than about the fact that if we are succeeding at creating a fully autonomous car we will have also achieve the completion of an AI-complete problem.
So that you understand how big of a deal completing an ai-complete problem is you should know that in the field of AI the most difficult problems to solve are characterized as ai-complete or as AI-hard and when you solved an ai-complete task it is basically the same as implementing an AGI.
Because if you solve it you resolve every problem in AI but to solve it you need to solve every problem in AI.
Why car and not something else?
Autonomous cars are the perfect example of it because it as to have a wide range of human intellectual property, like reasoning, it should include computer vision, it should be able the have natural language processing and should deal with unexpected circumstance while solving a real-world problem.
The reason it is so difficult to make self-driving cars and why it is considered as an ai complete task is because self-driving cars have to deal and interact with humans. If it wasn’t the case self-driving car could easily behave the same way that warehouse robot behaves by already having a precise map to follow and communicating with each other so that they do not pump and have a coherent organization but unfortunately, humans are not as coherent as robots so to consist with human the self-driving car as the be able the think like and human and also communicate with the human.
But still, where are they?
You might be wondering this is not enough. and still wondering “Where are they?”, and to respond to this I will tell you they are near. We have made such amazing advances in the field of AI take for example AlpahFold by Deepmind that can predict how a protein will fold a challenge that we were trying to solve for years. Do not lose hope as we know in all technology-related fields a simple discovery can have tremendous outcome this concept can be more defined by Moore’s law.
We only need to find one key concept and the implementation of AGI will become extremely mundane.
