Continue the course, or start over?
Artificial: Something that is made or produced synthetically rather than occurring naturally.
Intelligence: The ability to gather information, extract knowledge from information, and apply the information to adapt to the environment or context.
AI is the building of intelligent machines.
Artificial Intelligence is a field of study that relates to the study of “intelligent agents”; where an intelligent agent is a device that perceives its environment and takes actions that maximize the chance of successfully achieving its goals.
Machine learning can be applied to solve a wide variety of problems. There are three main types of machine learning techniques: supervised, unsupervised, and reinforcement.
Here are a couple real-world examples:
Being fed music partitions by Mozart, Bach, Beethoven and others, AIVA creates mathematical models representing music, and writes unique compositions. AIVA became the world’s first virtual composer to have creations registered with an author’s rights society (SACEM). For human composers, AIVA can be a useful tool in brainstorming melodies.
Click the play button above to listen.
Researchers from Georgia Tech University demonstrated that an AI can create video games. First, the computer was fed video of several retro arcade games. Next, machine learning algorithms catering to level design and game mechanics were run. To add a twist, the algorithms expanded the models to create hypothetical games that were based on but different from the exact videos ingested. One example combined wall obstacles with enemy mechanics, creating a new game, “Death Walls.”
We’re going to teach a machine to play the game Bouncy Circle using an artificial neural network.
First, let’s determine what a player would see:
1. The location of the obstructions (the walls).
2. The location of the circle on the screen.
3. The velocity of the circle.
The actions you can take: Bounce up, or not.
When not bouncing, the circle is pulled down by gravity.
Now let’s add the goal of the game: To go as far as possible.
These establish the rules of the game.
Next, let’s set up an artificial neural network:
A circle decides to bounce (yes or no) based on its position, velocity, and the location of the wall obstacles.
Now we’ll create a group of circles with different parameters to follow the rules we’ve established. Some will bounce faster, some will wait longer to bounce, etc. The first assortment will be totally random.
When all of the circles have failed, a new round will automatically start.
Now, we’ll set up a genetic algorithm, that will apply natural selection to our group. The best circles, those that go furthest, will be included in the next round.
Speed: 1
High Score: 0
All Time High Score: 0
One problem tackled by deep learning is handwriting recognition. The task is complicated by the wide variation in shape and size from one person’s writing to the next, and the similarities between numbers like 1 and 7.
You can try it here. Draw a single digit number in the first box. The machine will guess the correct digit. This neural network was trained with 60,000 labeled examples.
Importantly, machine learning is not the same as human learning. It looks for correlation in data and helps make predictions, but it does not have a full understanding of how or why those predictions happen.
Here’s an example of a strong correlation: This chart illustrates the per capita consumption of mozzarella cheese and the number of civil engineering doctorates awarded from 2000-2009.
The data aligns pretty well, but did eating cheese really affect the amount doctorates awarded? A strong correlation does not mean causation.
Exposing machine learning algorithms to biased data, or incomplete or incorrectly labeled data, will lead to inaccurate predictions. And these biases are difficult for the algorithm to detect.
There are two components to successful machine learning: good quality and quantity of data, and algorithms that scale.
Artificial Intelligence is a field of study relating to intelligent machines. Machine Learning is a set of technologies that make machines intelligent through training.
Machine Learning looks for correlations in data to make predictions, but it does not have a full understanding of how or why those predictions happen.
The two components to successful machine learning are good quality and quantity of data, and algorithms and implementations that scale.
How can AI benefit your domain?
Thank you for taking Digital Fluency: AI
Additional Resources for Further Exploration:
What is a Neural Net?
Digit Recognition
This learning was not created by a machine. Not yet.