We've been hearing a lot about artificial intelligence, or AI, of late. 

AI is nothing new. 

AI's Got a Lot of Baggage

We've seen portrayals of AI in science fiction for over 100 years. It has alternatively been depicted as solving all of the world's problems or as an existential threat to mankind — causing some people to try to rebrand AI as Cognitive Computing. 

Artificial intelligence carries a lot of cultural baggage. 

Real research into what we call AI, especially natural language processing (NLP), visual and audio pattern matching, machine learning and processing text for context and meaning, has been underway for over 40 years. 

From the perspective of the software industry, that is a very long time. Proven research typically turns into products in roughly 10 years. The research on relational databases from the 1970s turned into commercial RDBMs by the end of the 1980s. The ideas about hypertext from the 1980s were the internet’s World Wide Web by the 1990s. 

It’s rare that software takes an entire human generation to become feasible and relevant.

A Cloud Solution to Missing AI Skills

AI took a different path from other software technology for two reasons. First, it’s very hard. It requires understanding patterns in language, video and audio. Even harder is training a program to automatically recognize these patterns, without hard-coded rules. Getting a computer to make up the rules as it goes along — the way humans do — is devilishly difficult.

Worse, the average computer science major won't typically walk out of an undergraduate program with the skillset needed to create AI. That's the realm of a Computer Science PhD degree, which very few people in the workforce possess. 

However, the current cloud computing environment changes the availability of this technology dramatically. IBM Watson, for example, contains a host of AI services related to text processing, NLP, video processing and other AI. So does Google Cloud Platform, Amazon Web Services and Microsoft Azure. These cloud computing companies have made AI available to the masses.

AI Finds Its Purpose

More difficult than building AI software has been finding a reason to use it. Conventional software techniques easily meet most business needs without AI-type processing. When a certain degree of complexity creeps into business systems, that's when the need for AI surfaces. 

With the emergence of social media — and its constant streams of text information — in the mid-2000s, employing AI systems began to make sense. The volume was too high for human analysis, so businesses brought in AI to automate the process. 

Similarly, as computer systems be too for conventional user interfaces did NLP start to make sense. Video processing has been spurred on by cheap smartphone cameras and machine learning by the vast amounts of data being generated by Big Data systems and, in the future, streaming data from the Internet of Things (IoT). 

AI now has a serious purpose — driving complexity for the end user out of computer systems and data.

More importantly, AI, by virtue of its ability to detect patterns in data, helps businesses to determine possible future outcomes. The ability to find patterns in both historical and current data allows a business to map out a set of possible futures to inform decisions. This is incredibly powerful, but not easy without AI augmentation of human predicative capabilities.

Small and Focused Wins the AI Race

Now that AI is much easier to deploy it is tempting to build it into everything. That would be a mistake. AI is not an overarching intelligence like a robot brain. It is a discreet set of functions that can enhance existing business processes. 

Typical use cases include:

Finding meaning in social media

Social media is raw and unfiltered stream of information from customers, potential customers and the community at-large. AI helps analyze social media data to see how online chatter affects a brand, surface customer or technical problems with a product that have not come up through the usual routes, and to identify emerging trends.

Determining location from visual information

With smartphones, the ability to shoot video is now ubiquitous. Visual processing allows an app to determine location from videos of the surrounding area. That may mean finding where someone is in a store or where they are standing in a city in order to offer suggestions of things to do or buy.

Providing enhanced recommendations

Machine learning plus audio, video and text processing can create better recommendations for music, movies, books and other products that might appeal to an individual based on whatever they are listening to, watching or reading. This requires finding patterns in the actual data and not just matching metadata.

Processing IoT for events

The real market for IoT is industrial machinery and vehicles, large consumer products such as cars, and smart buildings. Sensors attached to these “devices” will produce lots and lots of information — but only discreet events will be interesting. Text processing and machine learning can be used to determine meaningful events in these large streams of data.

Predicting if and when someone will buy something

Machine learning is key here since patterns of behavior in customers need to be teased out from existing structured data and external data sources such as news. Each customer is different and environments can be fluid. Machine learning can help automatically pick out activities for accounts based on what super salespeople do intuitively.

There are many more but all have one thing in common: they are small, focused features that are part of a bigger set of functionality. Sales aren’t going to be performed entirely by a sentient computer. The human salesperson, on the other hand, will have an augmented experience with AI-driven sales guidance. Human listeners can listen to music someone likes and suggest other songs that might be enjoyable. That doesn’t scale however, so an automated approach needs to happen through AI.

AI is already making its way into the business world because businesses need help managing the complexity of data. They also now have access to AI components that weren't available in the past because they were so hard to build. AI is best used to deal with focused use cases that augment humans, not replace them.

So, don’t worry. The robot apocalypse is not happening … at least not yet.

Title image by Sam McJunkin