What is Machine Learning

Machine Learning is making machines intelligent.

People have dreamed of creating intelligent machines since long before the modern understanding of a computer, the computer introduced the first machine which was possibly capable of realizing this dream. As long as there have been computers there have been those trying to make them intelligent.  There is a long history of computer engineers attempting to create smart machines, from early neural networks in the 60s through the artificial intelligence breakthroughs of the 80s to today’s focus of the statistics focused machine learning.

In fact in the 18th century a machine was created called the Automated Turk which could play chess! Who knew such a difficult problem was solved in the age of steam engines – hundreds of years before computers. Amazing!

What does it mean to be intelligent? This is an incredibly difficult question that seems easy at first! In the most general sense it is simply the capability to take in data, make connections, and draw conclusions with respect to goals.

Intelligence doesn’t exist in a vacuum.

There is no way to generally define intelligence without respect to goals being achieved or problems being solved. The most intelligent systems we know of – be it genius individual people or powerful networks of computer systems – are considered so because they are able to solve problems. Consider two intelligent systems – the renowned genius Albert Einstein and a cat. How could we test which is smarter? If we gave them both mathematics problems to solve Einstein would certainly do better but if Einstein and the cat were both placed on their own small island with no supplies it is very likely the cat would achieve better results. Because of this difficulty in measuring general intelligence we instead look at circumstantial intelligence.

Intelligence is much easier to measure against a specific task. For the limited scope of a problem systems can each be measured against each other with confidence. If computer A is better at playing chess than computer B we could say computer A is a smarter chess player. Even within a limited scope though there could be different measures of intelligence. The broader the scope of problems a system is good at the more intelligent it is.

There are many types of intelligence.

While most people consider intelligence the domain of big brains and deep thinking it is most always this case. Often a dull system can still behave very intelligently.

Consider a self driving car attempting to cross a desert. The desert is treacherous with many obstacles such as high wind, shifting sand, and large hills and the car must cross many miles of this in order to achieve its goal. There are two car systems, each with different strategies.

The first is based on the most modern machine learning techniques. This car has a suite of inputs including speed sensors, obstruction sensors, and environmental sensors as well as a huge repository of knowledge in the form of detailed maps, understanding of soil conditions and how they effect wheel slippage, and more. The car also has a powerful on board computer which takes all of this knowledge and data and interprets it to calculate the best path, carefully avoiding steep slopes and areas it thinks contain a lot of sand. In the end it makes the trip over several hours.

The second car system is a car with giant tires and a big engine. It doesn’t have a lot of sensors and doesn’t bother to turn but it’s physical capabilities makes it so the car can drive over all of the problems it comes across. It makes the trip in a few minutes.

The first car takes a classical approach to intelligence, attempting to understand and adjust to all conditions. The second car takes an approach called “embedded cognition” where the physical manifestation of the system embeds the problems solving tools required for the task.

Machine learning is a large field with many subtleties.

Machine learning is a statistics based approach to creating intelligent machines focused on creating systems able to solve difficult tasks with quantifiable results.

Leave a Reply

Your email address will not be published. Required fields are marked *