Artificial Intelligence is a buzzword of which its meaning isn’t always entirely clear to everyone. In this blog we elaborate on the meaning of this concept, including a brief history and its relation to Machine Learning.
Artificial Intelligence and Machine Learning are both part of the field of data science. Like data science, they originate from the three scientific fields of mathematics, statistics and computer science. Artificial Intelligence and Machine Learning both have been around since the joint existence of these three fields, of which computer science being the youngest. Since the existence of computers, academics have been fascinated by the possibilities to make it perform tasks that might supersede human intelligence. Therefore, the concept is not entirely new, although Artificial Intelligence is perceived as an upcoming trend.
The fifties preluded the era of Artificial Intelligence pioneers. Algorithms that enabled computers to win in checkers or ‘speak’ English were programmed. Because of this promising scientific breakthrough, these scientists predicted that within twenty years, AI would be able to perform any task a human can perform. However, today we know this statement wasn’t true. There are still many typical human skills AI cannot mimic (yet). A major difference between the fifties and now is that seventy years ago, mostly academics were interested in AI. While nowadays, AI finds many applications in business and other areas. We now live in a world where we are less limited by the boundaries of computational power, and we have access to vast amounts of data. The potential value of this data is starting to show, as numerous applications are already making our lives easier and/or help enhance our decision making. We have the belief that this is only the beginning of a new revolution.
Definitions of AI will differ among experts, and depend on the definition of intelligence. In the spirit of creating clarity, let’s pay a visit to our favorite free digital encyclopedia, Wikipedia:
“Artificial intelligence is intelligent behavior by machines, rather than the natural or organic intelligence of humans or other animals. In computer science AI research is defined as the study of intelligent agents: any device or algorithm that perceives its environment and takes action that maximizes its chance of success at some goal.”
Since AI is inseparably connected to intelligence, and our perception of what is intelligent changes over time, people get excited when new levels of AI are achieved. Sometimes even astonished or scared. However, when time passes our perceptions adapt, and we get used to the part AI plays in our daily lives. We call this the ‘AI effect’.
A good example that illustrates this effect can be found in the game of chess. For a long time, humans have considered this as a typical sport to challenge the skills of the human brain, in which the more intelligent or strategical opponent often wins. However, in 1997 the algorithm ‘Deep Blue’ was able to win from the defending champion Garry Kasparov, this was the cause for a lot of commotion in the media. Nowadays, one might not find this algorithm particularly exceptional. This is especially true for ‘Generation Y’, which grew up with technology, myself for example.
I remember, somewhere around the year 2000, my grandfather was playing chess against an algorithm on his electronic chessboard. Because I grew up with this technology I didn’t find such an algorithm to be exceptionally intelligent. But my grandfather couldn’t get his head around the fact that a computer could play chess so quickly and at a, for him, competitional level. In this case, I am a typical example of the AI effect.
I can imagine that the exceptionality - and hence the perceived intelligence - of recommender systems or speech recognition algorithms will also decrease over time. So, although the definition of AI may remain the same, our perception of what is – or what is not - intelligent may change over time.
Deepblue playing against world champion Garry Kasparov
To determine the relation between Artificial Intelligence and Machine Learning, we must first understand what Machine Learning is. Let’s consult Wikipedia again for a definition:
“Machine Learning is a field of computer science that gives computers the ability to learn without being explicitly programmed.”
The words ‘explicitly programmed’ are very important in this matter. A pocket calculator is explicitly programmed to do what it should, i.e. it has no freedom to search for optional routes to the correct answer. A Machine Learning algorithm is ‘implicitly’ programmed. This means, under the hood, every Machine Learning algorithm is implicitly programmed to optimize a specific goal function. However, to some extent, it bears the freedom to determine which route to take to optimize this goal function. Data scientists call this process ‘learning’.
To review, one could state that:
Machine Learning (ML) is not unlike the human or animal learning process (think of positive/negative reinforcement). Let’s illustrate this statement by solving the problem of being hungry.
When hungry, your brain transmits unpleasant signals, which will make you feel unhappy. In response, you try several things to make this unpleasant feeling go away. Eventually you will try to solve this problem by eating, let’s say a cheeseburger, which resolves your hunger. As a result, your brain transmits pleasant signals, and your feelings of happiness increases to its previous level (or in case of a cheeseburger it might exceed it). This causes your brain to rearrange the structure of its synapsis. As a consequence, the next time this feeling of hunger occurs, your brain will be able to find the solution to this problem much quicker: eating a cheeseburger.
AI as a simulation of human intelligence
One could state that AI and ML are attempts to simulate human intelligence with computers. In the metaphor above:
Many AI applications continuously get fed with vast amounts of real-time (and hence more relevant to current situations) data. What happens then is the algorithm is changing its underlying model. The algorithm concludes that the current predictive model does not optimize its goal function at the most efficient level possible and it tries to find new relations in the data to do so in the future. This results in a different predictive or prescriptive model with different relevant variables and parameter values that are able to better optimize the goal function.
Suppose the Spotify algorithm finds out a user doesn’t like a song anymore, the user used to like. Suggesting this song over and over might annoy the user. Luckily, Spotify understands these signals and changes its recommendations. In this case, we speak of an evolving model or continuous learning.
Unlike the Spotify algorithm, many intelligent algorithms in production are static, i.e. they do not change their underlying model of calculations. A lot of AI purists believe that such an algorithm is not AI because it does not continuously adapt to its environment and input signals. They are just pieces of code and are (in essence) not much more intelligent than a simple pocket calculator.
Artificial Intelligence and Machine Learning are here to stay, and have already proven their value to both businesses and consumers. AI applications will develop fast in the foreseeable future, and will climb the ranks of businesses and help make better business decisions on the highest levels, creating significant impact.
This blog was the first part of a series about Artificial Intelligence, in the second blog we will explore the implications of AI for business and society. Will the technology we created ourselves take over the world? Stay tuned and find out in our next blog about Artificial Intelligence.
The future of AI offering you a helping hand