in

Artificial Intelligence Is Not as Intelligent as You May Think

AI Isn't That Intelligent

If you think science has become so advanced that we will soon be living in a world where everyone has a personal robot who takes care of things, you are sadly in a dream. Although Artificial Intelligence has progressed immensely from its humble beginnings, science is still taking baby steps in this field. And although a lot of the applications we use today are only possible thanks to AI, the truth is, these applications are programs or algorithms. They cannot accurately be called ‘intelligent.’

Before we discuss some of the reasons why Artificial Intelligence isn’t intelligent (yet), let’s take a brief journey back in time.

A Brief History of AI

The possible link between human intelligence and machines was first observed in the late 1950s. Norbert Wiener was the first to theorize that intelligent behavior was caused by feedback mechanisms that could be adopted by machines.

The next step was taken by Newell and Simon in 1955 when they designed The Logic Theorist, which is accepted as one of the first Artificial Intelligence programs. However, the term “Artificial Intelligence” was coined by John McCarthy in 1956. Since then, research in AI has intensified, and the program called The General Problem Solver was created.

Here is a brief timeline of how Artificial Intelligence developed into what it is today:

  • 1950- Scientist Allan Turing proposes a test for machine intelligence
  • 1955- the term “artificial intelligence coined by John McCarthy
  • 1964- Chatbot named Eliza crated at the MIT AI Lab
  • 1989-neural networks used to steer autonomous vehicles
  • 2002- development of an independent robotic vacuum cleaner named “Roomba.”
  • 2017- AI AlphaGo beats world champion Ke Jie in a problematic board game

What Is Intelligence?

Researchers and philosophers continue to attempt to define the concept of intelligence. Despite long years of debate, no unified definition has been set. We usually think of intelligence as the degree of cleverness in an individual. For instance, someone who can solve complicated mathematical problems could very well be called intelligent. This being the case, a machine designed to do the same task is also smart, right? Well, it depends on how you view intelligence.

Sure, any computer can out calculate an average human, but does it really understand what it is doing? Does it even know what Mathematics is? Does it know the practical applications of what it is calculating? No!

No matter how efficient and fast a machine is, it is functioning under a predetermined set of rules. It either has been programmed to do a task or can absorb large amounts of data and learn from it. Still, some may argue that the fact that AI does have the ability to crunch large numbers or beat a human at chess makes it ‘intelligent.’ Well, here are just two examples of

Why AI Is Not as Intelligent as You May Think.

Artificial Intelligence Does Not Understand Human Language

The field of Natural Language Processing or NLP has taken significant steps in creating machines that can generate passages at the touch of a button. These advances are thanks to what is called “deep learning techniques,” which detect statistical patterns in word usage and argument from millions of texts. Still, according to new research, machines do not understand what they are reading or writing.

Researchers used a test called the Winograd Schema Challenge, which is used to evaluate the common sense reasoning of NLP systems. The test consists of more than two hundred questions that come in identical pairs except for one word.

This ‘trigger word’ changes the meaning of the sentence by using opposite pronouns. For example, “large” is changed to “small,” “high,” is switched with “low,” and so on. The machine must figure out which of the two options the changed pronoun refers to. The result? The latest deep learning models can reach up to 90% accuracy, but there’s more.

Another data set was created called the WinoGrande. This test had 44,000 of the same types of problems. The difference with the test is that pronoun references weren’t easily deductable using word associations. With this new problem set, the machines’ performance fell to 59.4%.

What does this mean? Since machines seem to answer correctly, the bottom line is they are following set patterns. Take away those patters, and they fail. Unlike humans who can understand broken grammar and put together the meaning of a few words, machines cannot.

AI Can Be Fooled Easily

The danger of adversities machine learning was proven by the cybersecurity firm McAfee on Tesla Cars. Tesla cars run on autonomous driving systems and use a camera system called MobileEye. This camera system read speed limit signs and feed this information to the car’s different features like automatic cruise control.

To test how easy this system could be hacked, researchers stuck a tiny sticker on a speed limit sign. This small alternation caused the camera to read the 35-speed limit as 85 and increase speed to 50 miles per hour. Other research conducted by UC Berkeley used stickers to fool a self-driving car into misinterpreting a stop sign for a 45 mile per hour speed limit sign. A year ago, another Tesla car was fooled into veering into the wrong lane, which messed the cars learning algorithms.

Although the purpose of the research was to educate consumers to the possible flaws, the writing on the wall is clear-machines can never out beat humans, at least for now.

Alexa and Siri Fooled by Software That Swaps Words

You probably have spent a few minutes entertaining yourself with Siri or Alexa and discovered how seemingly smart they seem. Well, a software program TextFooler can cause NLP systems into misunderstanding text. By merely replacing certain words in a sentence with synonyms, the accuracy of these state-of-the-art systems reduced was drastically reduced.

The software developed by an MIT team replaced essential words with a synonym that humans would typically use. Merely rearranging the words in a way that wouldn’t make any difference to a human, caused a lot of confusion to an AI. The experiment showed how easy it is to fool an NLP system like Siri, Alexa, or Google Home. It doesn’t take a lot to see how limited Siri, Alexa, and Google Home really are. Change your accent, speak with broken English, and you’ll have a very confused AI.

Artificial Intelligence Slowly Seeping into Our Lives

Machines may not yet be taking over our lives, but it is affecting how we live daily. From the smart voice-activated computer systems like Siri and Alexa, AI has become part of our lives, whether we like it or not. Here are just 5 of the most popular applications of AI in use today:

Netflix

Netflix offers accurate predictive technology, depending on your reaction to movies. It does this by analyzing billions of records to come up with a list of films you might like based on your previous choices. As the data set grows, the program continues to become more and more accurate.

Amazon

The transactional AI used by Amazon is behind the success of the online business giant. Each year, Amazon’s algorithms continue to be refined, and the company continues to coerce buyers with what they are interested in. Sometimes, the product suggestions are just too tempting to resist. Before you know it, you are ordering something you never intended to buy.

Tesla

Considered as one of the coolest, highest-tech cars ever made, Tesla is the future of automatic driving. The car’s features include self-driving, predictive capabilities, and much more. The car’s system continues to upgrade with the over-the-air updates.

Boxever

Boxever is a company that is dedicated to creating learning Artificial Intelligence to improve customer experience in the travel industry. The platform uses data and AI to help brands generate smarter customer interaction. The AI analyzes customer profiles and then automates how to personalize communications on every channel.

Pandora

This is considered to be among the most revolutionary AI systems active today. The so-called musical DNA is based on 400 vocal characteristics that analyze each song manually. The music streaming automated music recommendation app plays songs that have similar musical traits.

The Future of AI

Scientists continue to develop Artificial Intelligence that can imitate human intelligence, but they are far from perfecting the process. AI can create music, paint, imitate human voice and speech, read and even write, but they still follow an algorithm. Whatever they do is because they are programmed to do.

Also, if they seem to have the ability to ‘learn,’ this too is an algorithm that is limited and can easily be fooled as we have seen. Although AI does continue to amaze us with impressive feats, it is far from taking over the world or evolving into an intelligent, aware being.

The future of Artificial Intelligence is bright as scientists continue to discover more ways to create the perfect machine capable of speech, understanding, and emotion. Still, this dream is further than you think. AI won’t be taking over the world anytime soon, and unless they become genuinely “intelligent,” they never will.

Written by Siddhesh Jain

My name is Siddhesh Jain, and I'm the fastest man alive (when it comes to learning, that is). I've been acquiring knowledge and learning new strategies for years, and I'm here to share my vast experience with you. What's my goal? Simple, I want to help every young entrepreneur succeed. Follow me on my journey!