Artificial intelligence (AI), sometimes called machine intelligence, is intelligence demonstrated by machines, in contrast to the natural intelligence displayed by humans. Leading AI textbooks define the field as the study of “intelligent agents”: any device that perceives its environment and takes actions that maximize its chance of successfully achieving its goals.
Colloquially, the term “artificial intelligence” is often used to describe machines (or computers) that mimic “cognitive” functions that humans associate with the human mind, such as “learning” and “problem solving”. Learn where our future is headed by understanding AI’s past. This article provides an in depth overview of the past, present, and future of AI.
AI is everywhere in tech right now, said to be powering everything from your TV to your toothbrush, but never have the words themselves meant less. Since the mid 1900s AI has taken many different forms.
This includes everything from automata to linear regression and perceptrons to decision trees and eventually neural networks and deep learning. As progress is made and the public becomes aware of what this “AI” truly is, the methods very commonly make their inevitable shift from an intelligence to a statistical technique.
AI often revolves around the use of algorithms. An algorithm is a set of unambiguous instructions that a mechanical computer can execute. A complex algorithm is often built on top of other, simpler, algorithms. A simple example of an algorithm is the following (optimal for first player) recipe for play at tic-tac-toe.
- If someone has a “threat” (that is, two in a row), take the remaining square. Otherwise,
- if a move “forks” to create two threats at once, play that move. Otherwise,
- take the center square if it is free. Otherwise,
- if your opponent has played in a corner, take the opposite corner. Otherwise,
- take an empty corner if one exists. Otherwise,
- take any empty square.
Learning algorithms work on the basis that strategies, algorithms, and inferences that worked well in the past are likely to continue working well in the future.
These inferences can be obvious, such as “since the sun rose every morning for the last 10,000 days, it will probably rise tomorrow morning as well”. They can be nuanced, such as “X% of families have geographically separate species with color variants, so there is a Y% chance that undiscovered black swans exist”.
Learners also work on the basis of “Occam’s razor”: The simplest theory that explains the data is the likeliest. Therefore, according to Occam’s razor principle, a learner must be designed such that it prefers simpler theories to complex theories, except in cases where the complex theory is proven substantially better.
Well, the biggest advantage of making use of AI is the most obvious: you never have to actually program it. Sure, you do a hell of a lot of tinkering, improving how the system processes the data and coming up with smarter ways of ingesting that information, but you’re not telling it what to look for. That means it can spot patterns that humans might miss or never think of in the first place.
And because all the program needs is data — 1s and 0s — there are so many jobs you can train it on because the modern world is just stuffed full of data. With a machine learning hammer in your hand, the digital world is full of nails ready to be bashed into place.
Where we are Today
The current state of the art in language processing has effectively merged recurrent and convolutional neural networks, while creating new methods within. All of this centers around Transformers, self-attention and word embeddings.
These concepts all help to model relationships of words in massively parallel CNNs or FCNs that can effectively understand full corpuses.
A transformer does exactly what its name lends it to — transforms a set of word embeddings into another set of word embeddings or similar structures. This is particularly effective for machine translation, word generation, or vector creation for classification.
This is a great in depth look at transformers, but in short they are basically a pairing of encoder and decoder networks that are trained to accomplish the tasks above. A powerful concept that lyes at the heart of the transformer is self-attention.
Self-Attention was recently developed by Google as a method to model the recurrent and spatial nature of languages in a single network pass. To do this, complex mathematical methods are performed on each input sequence using a query, key, and value matrix. These values are tuned to model spatial relationships of all words in a sequence.
Where we are Headed
The opportunities are endless, as massive changes in methods can be brought to life by any unique mind, debates on AI and data ethics will continue, and businesses will rely more and more on these methods as their most valuable resource.
Taking the time to understand where we came from and where we are going can allow everyone to develop their own vision of the future. The global matrix of these unique human visions is what will lead us into a bright future with AI at our side.