AI: Top Down vs. Bottom Up

When we think about technology, one of the first topics that comes to mind is artificial intelligence. Computer scientist Alan Turing, responsible for cracking the Enigma code used by the Nazis during World War II, formalized the search for authentic artificial intelligence (AI) in 1950. He did so by developing a test meant to determine whether a machine was exhibiting signs of real intelligence or not, a test he thought would be easily passed in time for the new millennium.

Throughout his career, Turing published articles and manifestos regarding the topic of AI where he suggested the use of two different approaches we refer to as top-down and bottom-up. But what do these approaches mean? What are the main differences between top down vs. bottom up AI? And what is the Turing test, in detail? That’s what we’ll try to cover in this article.

What is bottom up and top down AI?  

Artificial intelligence seeks to emulate human thinking through two basic approaches: top-down AI and bottom-up AI. As we’ve mentioned, the definition of top-down vs. bottom-up AI goes back to Turing’s manifesto of 1948.  

In simple terms, the top-down approach (also referred to as the symbolic approach) can be understood as breaking big problems into smaller ones that are easier to solve. This is based on previously established knowledge and relies on symbols or rules (hence the symbolic classification). 

On the other hand, the bottom-up approach (or connectionist) relies on adaptation and more realistic behaviors,  combining simple models and systems that build up to form more complex ones. Bottom-up AI tries to build structures that emulate the human brain (thus being referred to as connectionist), and is based on models of interactions with environment, instead of symbolic descriptions of these environments, used in the top-down variant. 

What is (and how to pass) the Turing test?

To measure this, he proposed a sort of game in which a group of humans have a text conversation with a judge. The judge can’t see who he or she is speaking to, and, at one point, one of the players is replaced by a machine. If the judge can’t tell the difference between the computer and the human players, then the Turing test is passed. In that sense, the objective of the test would be for the human evaluator to be able to distinguish the machine participant based on its responses. Failure to do so would mean the machine participant passed the test, being able to demonstrate intelligence equivalent to that of a human being.

Today, we have big players who seem at the verge of passing the test: Apple’s Siri, Amazon’s Alexa and 4-Time Loebner Prize winner Mitsuku. Many even regard Google Duplex as a Turing test victor.

The issue is these bots have built-in mechanisms to digress from conversations they don’t understand, putting into question the validity of their performance.

However, even if the Turing test hasn’t been passed, it doesn’t mean the advancements in AI have not been groundbreaking and valuable contributions to society up to this point. The applications of AI in our daily lives currently range from process automation to customer service, and with its constant evolution we can expect much more to come.

AI is all about perception

To get to where we are, many computer scientists have spent countless hours thinking about how to create true artificial intelligence.

Simply choosing which approach to follow is a loaded argument.

As AI seeks to emulate human thinking, the debate doesn’t only center on top down vs. bottom up AI, or how to best solve a problem, but on figuring out how the brain works. For computer scientists, it seems the most essential aspect of the human mind is its ability to perceive or process information.

According to the Oxford Dictionary, perception is “the ability to see, hear, or become aware of something through the senses,” but it is also “that in which something is regarded, understood, or interpreted.” The dichotomy between interpreting and becoming aware is precisely what has created the divide in approaches, but also what has given us varied and rich results still being tested and perfected today.

Top-down vs. bottom-up AI

When facing a problem or equation, you can find a solution in one of two ways. You can either process the problem in your head by matching it to previous information or you can let the equation present its variables to you without adding any context.

The first describes a top-down approach, used by those who prefer applying previous knowledge to educate their perception. This method, also called “neats,” suggesting a focus on logic, order and data, is preferred when dealing with high-level tasks like neuro-linguistic programming.

The opposite approach, bottom-up, is based on the belief that development should part from a stimulus. In other words, what drives our perception is that which we sense. This method, also called “scruffies,” denoting dynamism and functioning on an ad hoc basis, works better with lower-level tasks such as robotics and speech recognition.

In the end, no approach is better than the other, but both of them continuously offer good results.

Perhaps one day we will achieve true artificial intelligence, but as Siri and Alexa have proved, maybe humanity doesn’t need to create sentient computers to make artificial intelligence work for them.

Ready to send money with Ria? Download our app for iOS or Android today to get started!