Artificial intelligence (AI) is a term to describe a branch of computer science that is dedicated to creating intelligent machines that would learn to work and react like humans. The field of Artificial Intelligence is not limited to creating only high levels of machine learning, but also to utilize data input in a manner required to produce a needed result. AI has multiple varieties and characteristics. Some of these types are noted as follows.
- Reactive intelligence – As the term implies, reactive intelligence is the type of artificial intelligence that only reacts to certain input. Reactive AI does not have the ability to form memories or use past experiences in order to make a decision. On other words, they are incapable of learning in the traditional sense. Instead, the AI is programmed and will react with a predefined set of parameters. One of the more famous examples is IBM’s Deep Blue chess computer that was able to beat Garry Kasparov in 1997.
- Limited memory intelligence – This type of AI is an improvement over reactive AI in the sense that it is capable of using memory to study past data and then make a decision based on those choices. The type of memory this form of AI uses in general is short-lived or temporary memory. One of the more famous example of a limited memory AI is a self-driving car. The AI that is built into the car can use sensors and then call upon its previous memory in order to identify pedestrians or traffic signals. In this way, it is able to make quick decisions that can reduce accidents occurrence. An important thing to add is that machines that operate on a limited memory principle have a set of values pre-programmed and are unable to change those main parameters. They can adapt to new changes by modifying the original program, but they cannot change their behavior unaided.
- Theory of mind – In the examples we have seen so far, we have identified machines that are intelligent to a certain degree, but cannot really derive much from their defined set of values or libraries. Theory of mind AIs are different in the sense that they will be able to socialize and understand human emotions. In other words, they will be able to better understand human behavior and interact with us just like any other human and not as a tradition learning machine. Theory of mind AI computers are currently in development by various companies around the world and as of yet to be built. The main contention for these type of AIs is that in order for them to be truly self-aware, they must understand emotions and be able to change as well as adapt to human behavior and interactions.
- Self-aware AI – Self aware AI machines are what most people imagine when thinking about artificial intelligence as seen in the movies. Machines built on a self-aware principle will be super intelligent, sentient and have a certain level of consciousness. The self-aware AI concept represents the holy grail of artificial intelligence, and we have yet to reach the level of technology needed to make machines truly self-aware.
Additionally, self-aware AIs also present moral and philosophical challenges as we continue to debate about what makes human/machine interactions truly reciprocal in action or influence.
From a technological standpoint, we are yet to determine how to replicate and incorporate the neural patterns in the human brain that allows us to think and analyze ideas while incorporating our emotions. Another question which arises with truly self-aware intelligence is simply “Will it be any good?” not good in a relatively moralistic sense, but truly beneficial in its purpose. This calls into question what it means to be human, and because we make mistakes, will those errors creep into the decision-making process. Many believe the simple reason that humans make mistakes is based on their ability to process emotions and because of this, there is a chance that a human designed machine, no matter how intelligent, can be prone to issues because of its human-based programmed emotional biases. With our current level of technology, self-aware machines are still only a concept at this point, but it will be interesting to see what the future holds.
Achieving Artificial Intelligence
So, now that we understand what types of machine intelligence or semi intelligent machine we want to build, what would be our next step? How will the machine learn?
The main method of teaching a machine intelligence, or in other words, teaching a machine how to gather and process information is a concept known as machine learning. Machine learning is achieved by using complex algorithms that discover patterns and generate insights from the data that they are exposed to. An important subcategory of machine learning is called deep learning. Deep learning is a method of creating a simulated artificial learning ability for the machine, which allows it to mimic a human brain using an artificial neural network to make sense of patterns, noise and other sources of input within the confusion of large amounts of incoming data.
How Does Deep Learning Actually Work?
The most basic model of deep learning works by mimicking neural networks by creating the multiple layers of input. The first layer would be called the Input layer and it would serve to feed weighted values into the machine.
The next is called the Hidden layer. This layer would be responsible for mathematical computations or feature extraction based on human inputs that resides in between an input layer and an output layer. As more and more hidden layers are added, additional processing of deeply complex data can occur within that region. The simplest way of explaining this is that within a hidden layer, values are assigned by human input (our data) and then the machine will begin connecting the values we set with data we provided. For example, let’s imagine we have a set of dog pictures within a hidden layer. We can add values like bread, color, size, etc. and the machine will assign those values with input data. The more hidden layers we have the more accurate the predicted output will be. This process can be more finely tuned through a process called backpropagation.
The Output Layer is the last layer and it will provide us with results that are given from the hidden layer.
Where to Begin?
When we talk about various types and forms of artificial intelligence, an average user might ask where should I begin?
Luckily, there are tools specifically developed and designed for the purpose of teaching your AI. The most popular frameworks being used today are TensorFlow and Theano. Both of these learning platforms are open-source and have their specific advantages and disadvantages depending on need. TensorFlow is more user-friendly and is currently used more often. Theano has recently lost some of its popularity but still retains a strong online community. Theano leads in usability and speed, but TensorFlow is better suited for deployment and therefore is used more often by researchers and data scientists. There are other frameworks that you can use depending on your need so its best to research all the available options and then select the one suited for your specific needs.
While these frameworks have their own respective libraries for deep learning, a popular way of improving them is to use Keras. Keras is a high-level neural network API written in Python and capable of running on top of both TensorFlow, Theano or CNTK. For more information about what Keras is and how to install it, you can review our Knowledgebase article on Keras.
The domain of artificial intelligence, smart machines and neural network learning is far more advanced today than in any point in our past history. Advances in AI continue to move us forward in search of methods that take better advantage of machine-based reactions that mimic humans behavior. Utilizing machine learning or AI to improve or better us is not a new idea, but due to the technological limitations in the last century, many advances were mostly science fiction. With the dawning of the 21st century, faster computers, more powerful programming languages, the rise and increasing speed of the internet and multiple other technological advancements have made it possible to bring these ideas from the realm of science fiction into a new reality of the possible.
There are still many challenges needing to be overcome in order to achieve a fully self sentient AI or super intelligent self-aware machine, but as computer science, hardware technology and advanced programming languages progress become available, the possibilities continue to grow. While we are sure that we will not see machines achieve sentience anytime soon, our children or grandchildren very well may experience these benefits sooner than we think. Many large companies like Facebook, Amazon, Google, Apple and Microsoft continue to utilize these methods in our daily lives to improve our ability to progress in the search for the creation of a truly artificial intelligence.
Should we be afraid of this advancement? At this point in time, artificial intelligence is currently in use in many applications like Siri, Cortana or Google Assistant as well as services like Netflix that use machine-based learning to provide movie choices based on prior viewing habits which pose no real threat to humans since they still operate within defined parameters set by their creators. The goal of achieving the level of self awareness and intelligence in a machine that can create independently is still out of our reach and it would present other philosophical and moral questions upon its creation.
Finally, an advantage of the knowledge we currently possess in this arena is that artificial intelligence is no longer reserved for large corporations or research universities. As open-source machine learning platforms and frameworks continue to improve, anyone can explore, create and expand upon this knowledge. Artificial intelligence and smart machines are still in their infancy but continue to grow as a promising area of problem solving.
Would you like to know more about how to take advantage of this technology?