What is AI?

The definition of artificial intelligence (AI) is broad. AI refers to a branch of computer science that enables machines to perform tasks that mimic human behavior, which may include the capacity for knowledge representation, planning, learning,  judgment, adaptability, decision-making, intentionality, contemplation and, to an extent, creativity. Continue reading to learn how AI works and the different types of artificial intelligence.

How Does Artificial Intelligence Work?

So how does AI work? AI uses large inputs of data to create environments for machines to create and process (think) while also reacting and responding (behaving) in ways that parallel human intelligence. Essentially, AI sometimes operates by processing data through advanced algorithms. Creators expect patterns to form from artificial intelligence, meaning that the data insights may be useful or part of the design.

A bootcamp in data science may help lay the foundation for those interested in exploring artificial intelligence. Another option is artificial intelligence (AI) courses.

AI courses commonly include:

• artificial intelligence strategy.
• artificial intelligence in health care.
• artificial intelligence in robotics.

Some examples of artificial intelligence include:

1. Siri and Alexa: use machine learning to better predict human language.
2. Tesla: all the car’s self-driving features use AI.
3. Netflix: the platform uses AI to recommend movies based on a user’s viewing habits.
4. Pandora: uses AI to recommend songs you might like.
5. Nest: uses behavioral algorithms to predict heating and cooling needs. 

The online Artificial Intelligence (AI) program from Columbia is designed for professionals and leaders who are seeking a deeper understanding of AI.

Learn more from Columbia University

Sponsored

What are the different types of AI?

There are a number of ways to consider creating AI, depending on your approach. Two common types of AI are:

• Narrow AI.
• General AI.

The main difference between narrow AI and general AI may be in the goal of the acquired intelligence. General AI seeks to closely resemble humanity, creating machines that have all the mental powers that humans have, including consciousness. Meanwhile, narrow AI sometimes seeks to outperform human intelligence in one narrowly defined task.  

Narrow AI

A common narrow AI definition is that it exists to perform single, goal-oriented tasks. Narrow AI, artificial narrow intelligence (ANI) or weak AI, is the most common form of artificial intelligence and the only one fully realized to date. Using specific parameters and constraints, narrow AI may do one task very well, like facial recognition or natural language processing (NLP.) Narrow AI may either be reactive, which may mean it responds to a stimulus in real time, or have limited memory, where it may track large data sets to have deeper learning, for instance when your entertainment platform offers suggestions based on past usage data. Narrow AI sometimes uses limited memory to accomplish tasks. 

General AI 

Another type of AI is typically called general AI, or artificial general intelligence (AGI). Also known as deep learning or strong AI, general AI has yet to be fully realized. It may rely on machines to do a broad array of tasks and solve any number of problems using cognitive abilities that make it indistinguishable from human intelligence. This type of AI may require machines to understand and perceive thoughts, behaviors and attitudes of other living entities. It sometimes uses a theory of mind framework but may be limited by the scientific lack of understanding of the human brain. 

What Are the Uses for Narrow AI?

Narrow AI is often used to handle singular or limited tasks. One of the most available integrated forms of AI is considered to be Apple’s Siri virtual assistant — it brings the functions of machine learning to the iPhone. 

Here are some other common uses of narrow AI:

• Spam functionality in email.
• Chatbots.
• Image/facial recognition used by Facebook, Apple and others
• Disease mapping and prediction tools.
• Virtual assistantslike Siri and Alexa.
• Self-driving or autonomous cars.
• Content monitoring for dangerous content on social media.
• Entertainment suggestions based on past usage data.

What Are the Uses for General AI?

General AI is meant to closely resemble human intelligence. Though it’s not clear when or if this type of AI will be actualized, in theory it would allow machines to fully understand and mimic human thinking, cognitive function, behavior and decision-making. Eventually, it may even outsmart humans and automate certain tasks or functions, though human instinct, decision-making and relationship-building are considered to be challenging to replicate in machines. 

Other uses of general AI could include:

• Rapid number processing.
• The ability to perform human tasks or jobs for people.
• Robots and robotics.

Machine Learning and Deep Learning

The next step in artificial intelligence for many is whether machines can complete tasks that only humans could do previously. This would rely on machine learning where machines learn by doing and acquire skills without oversight by humans. Machine learning sometimes crunches data into meaningful patterns to make predictions. Deep learning includes neural networks – human-brain inspired layers of algorithms called neurons that stack and prioritize data and outputs until a particular action is “learned” by the machine. 

When this process is carried out across many neurons, deep learning occurs, where machines process enormous amounts of data to create meaningful responses, behaviors and actions. In this way, the machines are both thinking and acting autonomously. This is the basis of the deep learning tasks we see in today’s most common AI applications like speech recognition. In practical terms, some startups consider machine learning for a variety of applications, from better tracking and predicting customer preferences to automating certain human tasks. 

Types of machine learning include:

1. Supervised learning: this uses labeled data to scour huge amounts of inputs. For example, the label might be “fire truck” and the machine would scan thousands or millions of images to find all the images that match the label. Increasingly, fewer data sets are needed and this may eventually be called semi-supervised learning. 

2. Unsupervised learning: algorithms look for similarities in data to form meaningful patterns. The algorithm looks to group data sets with similarities, for example, the weight or color of an object. 

3. Reinforcement learning: through the process of trial-and-error, an algorithm gets rewarded for landing on the best outcome and learns to repeat it in the future. 

Why the Resurgence in AI?

There are multiple reasons for the resurgence of AI.

One reason is economic. In 2014, Silicon Valley pumped a half-billion dollars into AI. A part of this funding may have to do with the promise of what machine learning could open up for companies now and the potential for its future. 

Another factor may have to do with accessibility — what once had to be done on a supercomputer can now be achieved in the cloud at a fraction of the cost. 

Another factor driving AI may have to do with the sheer capability of machine learning platforms. The combination of big data and the growth of machine learning make it more possible than ever to realize human intelligence through machines. 

Conclusion

¹QS World University Rankings (2020)

AI capabilities have led to numerous innovations, including self-driving vehicles and connected devices in our homes. Personal assistants like Siri and Alexa and streaming recommendations on our entertainment platforms are other AI examples. 

If you are interested in learning more about AI, a bootcamp in data science or courses in AI might be a good next step for you.

Last updated: November 2020