Truly Understanding AI

The concept that a mechanical system may be able to perform tasks that exhibit what we call "intelligence" in humans and animals is not new. Scientists, philosophers, and innovators across different countries and cultures have come to the conclusion that a non-living entity can still accomplish this feat. So how does one understand artificial intelligence?

This is not a new question by any means, and it's one that has been asked numerous times over the course of human history. Ancient Greek thinkers such as Aristotle and Plato considered whether or not a mechanical system could exhibit “intelligence.” Stop in at various points throughout the 18th, 19th, and 20th centuries, and you find famous intellectuals like Alan Turing, John Searle, and Marvin Minsky posing this question as well.

Today, the question still persists. In fact, people are now asking it with a different word: "artificial." It is no longer enough to ask if a non-living entity can truly understand artificial intelligence; rather, we now probe into "how far can we take this capability?" A question with very serious implications. But there's another conversation that needs to happen—particularly in the global marketplace.

Do we truly understand AI?

Breaking Down Artificial Intelligence

Looking at it from a research perspective, artificial intelligence is a subfield of computer science.?It deals with the study and creation of non-living systems that, as Marvin Minsky defined, “...do things that would require intelligence if done by men.” This distinction is important to understand when we talk about artificial intelligence because it appears in a lot of research papers and product releases alike.

For example, IBM's Watson platform, infamous for besting the world’s top chess professionals and Jeopardy! Contestants, leverage a lot of AI techniques for voice recognition, natural language processing, and machine learning. Another large AI implementation occurred back in 2012 when the Obama Administration created the Big Data Research and Development Initiative (BDRDI) — a system designed in the hopes of solving complex national issues. From situation-specific systems to Big Data analytics tools to the classic conception of a machine with something that resembles human consciousness, there’s a wide spectrum of artificial intelligence applications that needs cognizance by corporate leaders.

The Spectrum of Artificial Intelligence

AI can be broken into large sections, depending on the type of problem it's trying to solve. Machine learning (ML) and artificial intelligence are often used interchangeably despite their conceptual and practical differences . ML refers to a system’s ability to learn from data and experiences without explicit programming, making it one subset of the artificial intelligence arsenal. However, even at the higher level of AI, there are major differences — particularly when it comes to Artificial Narrow Intelligence and Artificial General Intelligence. This distinction is important because it captures the spectrum of AI research.

Artificial General Intelligence

On one end, we have artificial general intelligence (AGI), which aims to create intelligence in machines that mirrors the cognitive structures and abilities of the human mind. Just like the construct of intelligence is applied generally in psychological assessments, AGI refers to non-living systems that can perform sensory perception, natural language processing (NLP), and other complex tasks without the guidance of programming. This type of intelligence does not yet exist and many top researchers still believe that vast improvements in neural networks and deep learning are needed for true realizations of AGI.

Artificial Narrow Intelligence

Artificial narrow intelligence (ANI), sometimes known as weak AI , is a tool currently being applied for real-world execution. ANI is narrow because it is designed to exhibit intelligence within the confines of a specific context or singular task. While the output of that singular task may be complex, highly valuable, and require thousands of hours of work were to be performed by people alone, it still doesn’t encapsulate the “strong” intelligence experts theorize AGI is capable of. The earlier examples of IBM’s Watson and the UG government’s BDRDI are both qualified as artificial narrow intelligence.

For enterprises and business leaders not at the cutting edge of AI R&D, ANI represents a much more applicable system thanks to two powerful core functionalities.

The Core Functions of AI

There are two high-level functions that these systems perform particularly well: categorization and prediction .?

What's the difference between the two? Well, think of it like this: we're all familiar with having a hundred songs in our pocket and needing to pick out just the right song to match our mood. When you think about it, that's categorization—this is also what your phone does when you ask it to pull up any number of genres or artists. Businesses can also use these systems to create a more specific outline of their consumer base, breaking buyers into categories based on different demographic and behavioral information.

Prediction is the much more touted application of contemporary AI systems. Predictive AI is able to make estimations that previously needed human-level cognition to decide upon. These systems generate predictive analytics that can be used by both businesses and individuals for any number of activities. From forecasting stock prices and supply chain costs to more reliable public health assessments, there are many reasons why business leaders and entrepreneurs are interested in this function.

?AI in the Commercial Realm

From a commercial application perspective, enterprise stakeholders need to be aware of the different types and applications of artificial intelligence discussed above. Profit motives, common use cases, and marketing perspectives are all driving a certain conceptualization of Artificial Intelligence that often sets enterprises up for disappointment.

For example, by viewing AI as "enterprise-ready," we miss some key limitations that exist in the application of these intelligent systems. For example, there is still a significant issue surrounding the training of these systems with real-world data. We must understand the shortcomings of AI systems, particularly ANI, to develop real solutions that can help us further democratize these technologies.

The second issue is that an entire set of tools surrounding issues like privacy, bias, and fairness are not yet mature enough for unfettered adoption by enterprises. Healthcare providers and insurers often run into issues of inequitable output results from enhanced ML and ANI systems that are leading to material differences in patient outcomes. To avoid these issues, companies need to take a more dedicated approach to properly training and implementing their AI systems.

Structured Implementation of ANI for Organizational Success

Despite all the doom, gloom, and concern surrounding AGI and ANI, they still offer lots of upside for enterprise companies. Categorization and predication of various enterprise data inputs have the potential to save companies massive amounts of capital when properly implemented. ANI outputs can reduce the cost of supply chain disruption, provide advanced warning of emerging market opportunities and help companies truly understand their consumer base.

As development within the field continues, and companies adopt the digital structures and protocols to accommodate the benefits of ANI, these systems will become a core component of successful enterprises. To learn more about what AI truly is and how it can be implemented within your digital infrastructure, reach out to the experts at Aviskaran today!

?

?