Seemingly overnight, machine learning became the hottest thing in artificial intelligence with impressive feats such as IBM Watson schooling Jeopardy and AlphaGo beating the world’s best Go player – a game where players must think on their feet and not simply memorize moves. Recent successes of smart bots is starting to scare even geniuses like Elon Musk who are growing concerned about a dystopian future where we’re serving the robots. Although entertaining, the hype around robots ruling the world has distracted us from what machine learning is and how it can be useful. In this blog, we’re going to breakdown machine learning: what it is, how can it help you, and what is the master algorithm.
Why does understanding machine learning matter? Machine learning and their algorithms currently shape much of our daily lives and will continue to grow in their influence. Think about the last time you searched something on Google. You most likely clicked one of the links on the first page, and these links were put there by machine learning. What about your last Amazon purchase? Did you follow-up with any of the suggested or related items? If so, machine learning played a significant role in your purchasing decision. What about the last time you checked your email? Your spam-less inbox is thanks to machine learning!
We begin to see how much influence machine learning has on our lives. It’s important for us to not only understand the implications of machine learning, but also, how it can be utilized to improve our daily life, businesses, and world.
So what is the difference between machine learning and artificial intelligence?
We can answer this simply: all machine learning is artificial intelligence, but not all artificial intelligence is machine learning. As defined by Merriam-Webster, artificial intelligence is the capability of machines to imitate human behavior. Machine learning is simply a specific way machines learn and achieve this intelligence. It’s characterized by the act of algorithms guiding a smart computer to learn and predict future outcomes. How does the bot learn? Cue big data. Large datasets allow for computers to parse huge amounts of information, identify trends in the past, run their own simulations, and see their own patterns. The implications of such a learning robot are tremendous including new advancements in cancer research and innovative companies.
What guides machine learning?
The short answer, algorithms. Machine learning algorithms are actually appear rather simple when compared to much of the code used in large companies. Learners strength comes from its ability to continuously digest data and begin to see their own trends in the world. Since algorithms are the guiding force of learners, the power is in the hands of the human controlling what variables the learner should choose to focus on or ignore. With this in mind, we humans have developed a few standard algorithms that are best used specific scenarios. In his book, The Master Algorithm: How the Quest for the Ultimate Learner Will Remake Our World, Pedro Domingo introduces the tribes of machine learning. These are the scientists with differing views on which type of learning and algorithms are best suited for artificial intelligence. Let’s take a look at the various tribes.
Symbolists
Symbolists are use pre-existing information and make inferences about related unknowns. Their approach is often explained as If-Then. For example, if all professors are human, and Emily is a professor, then Emily is also a human.
Bayesians
Bayesians are at home in statistics. Using previous data and viewpoint on their world, Bayesian learners create sets of truths that adapt if new data or simulations prove them untrue. With the updated data, the learner will create a new truth and so on. Our beloved spam filter operates this way. They create a truth of what a spam emails look like and then determine how statistically likely the incoming messages should be labeled spam or not.
Connectionists
The connectionists focus on reverse engineering the brain. They view the brain and its neural connections as the epitome of innovation and create artificial neural networks to resemble learning similar to that of a human brain – just with huge amounts of data and energy.
Evolutionaries
As their name alludes, evolutionaries rely on evolution and genetics as the basis for their machine learners. The key for these algorithms is constantly optimizing for the best fit like survival of the fittest.
Analogizers
Analogizers look at the world of data for similarities. Following the “Nearest Neighbor” principal, analogizer learners go from one scenario to the next always looking for how they relate. Finding shared characteristics is what makes these learners good for recommendation features found at Amazon and Netflix.
From the examples above, we get an idea of how each tribe of algorithm works and their impact on our daily lives. These machine learners are incredibly powerful but suffers from limitations where other algorithms perform better. It is for this reason that Pedro Domingo argues that there is a need for a universal learner.
One Algorithm to Rule Them All
Machine learning is currently limited by time, memory, and usability of algorithms.
There is too much data in the universe for a machine learning bot to digest it all in a reasonable timeframe. This is easily solved by adding constraints and selecting variables that limit the data but not so much so that the learner cannot accurately predict future outcomes.
Memory constraints are quickly growing obsolete as Moore’s Law continues to chug along.
That just leaves us with algorithms to tackle. Each algorithm school is limited to developing solutions for specific problems. A “master algorithm” would function as a universal learner that could present solutions to all problems.
Why is this important? A master algorithm is a holistic approach to machine learning. Learners following a master algorithm will have a complete picture of humanity and the world we live in. In one go, this algorithm would be the source of new theories of life, develop a cure for cancer, and suggest a new Netflix series worthy of your binging. The power of a universal learner is why some people call artificial intelligence the last invention humans will ever make. It’s hard to know whether these claims are driven by too many Terminator movies or real concern. What we do know is that machine learning has incredible benefits and will revolutionize business, medicine, and other fields in ways we cannot predict.
Sources:
https://techcrunch.com/2017/01/30/is-a-master-algorithm-the-solution-to-our-machine-learning-problems/
https://www.mckinsey.com/industries/high-tech/our-insights/an-executives-guide-to-machine-learning
https://learning.acm.org/webinar_pdfs/PedroDomingos_FTFML_WebinarSlides.pdf
Pedro Domingo, The Master Algorithm, 2015
Machine Learning Enthusiasts 