How to spot an AI Imposter
Today, we’ll discuss what Artificial Intelligence actually is, some noteworthy real-world applications of AI, and how to know if the tech vendor pitching you is an AI imposter.
First, what is Artificial Intelligence?
Elon Musk, the creator of PayPal and Tesla, sees so much relative advancement in AI that he worries we will eventually lose control of the technology (perhaps a bit extreme, but his perspective nonetheless). What many don’t know is that there are multiple types of AI, most of which are not focused on creating human-like consciousness in machines. Instead, the Artificial Intelligence that exists today is meant to help people with focused, specialized services.
Artificial Intelligence, in simple terms, is a system or machine that learns to make decisions on its own without being explicitly programmed to perform a specific task. The purpose of AI that we’re discussing today is to automate the processes and software functions that don’t require abstract thought (and in no way create a “Terminator” scenario).
So what are some real-world applications for different types of AI?
Odds are, you’re familiar with some of the functionality Artificial Intelligence provides.
Last year, chatbots were incredibly popular – and while it’s unclear if chatbot technology will last in the long term, it’s an interesting use case for Artificial Intelligence. Chatbots like Microsoft’s “Tay” utilized a form of natural language processing to learn from and interact with users on twitter, which unfortunately had a less-than-ideal result (because of internet trolls). While “Tay” didn’t work out as planned, B2C companies are looking to chatbots in the future to assist in functions that typically involve human interaction: customer service processes or chat functions to order a pizza on the Dominoes app.
Another interesting use case for AI is in agriculture, where farmers utilize machine learning algorithms to increase the yield of their corn crops. The agriculture industry has a very limited window of opportunity to plant seeds. There’s no shortage of data about which seed strains grow well and what the ideal conditions are; the challenge for these corn growers lied in analyzing the thousands of data points they’ve collected without generating custom algorithms and hiring full time data scientists. Machines, however, are very good at detecting and learning patterns, and don’t have to be explicitly programmed to look for specific patterns. They make inferences themselves without the need for human programming. By feeding an AI with agricultural data and field and crop images, the machine quickly provided decisions that were easily readable for non-technical staff.
While it’s still relatively new in medicine, AI is also having a large scale influence within the healthcare industry. Machine learning is designed to recognize and analyze patterns and create associations that would ordinarily take a human a great deal of time and effort to accomplish. By consuming data in Medtech, AI is helping doctors diagnose and create highly targeted drugs to a specific illness or disease, and even discover new strains of diseases and ways to cure them, again all without providing concrete instruction set to the machine.
These are just a few applications of AI in the world today, but there are countless others – here’s a short list of some AI applications that are a bit unconventional.
How do you know if you’re speaking with an AI impostor?
So how do you recognize an AI Impostor? How do you know if the tech company pitching your team has real AI technology or is just jumping on the AI bandwagon?
The various use cases mentioned above share a very important characteristic that’s necessary for any true Artificial Intelligence – none of the technologies require any human intervention or pre-programming to produce results. All of the examples above represent a machine that consumes information, automatically recognizes patterns, and makes the connections necessary to complete a resulting action.
With that in mind, recognizing an impostor is easy:
Any company that mentions “business rules,” “rule-based automation,” “rule-driven information,” “pre-programmed responses,” or “pre-defined strategies,” is not utilizing AI. These systems require humans to create rules that map user actions to specific pre-programmed or pre-computed responses.
For example, there is a class of marketing technology solutions focused on A/B testing that also incorporate “personalization” as part of their stack, where marketers can program rules for serving banners and general content. If users has browsed section X, serve banner A; if user has browsed section Y, serve banner B. These types of solutions are not AI. They require humans to feed them rules, and simply execute those functions.
The same concept applies to email triggering solutions. These solutions require marketers to create conditional rules like “if a user visits my site but does not buy, send him a message 24 hours later with the browsed product and a set of related product recommendations.” None of this is AI. This is a machine executing a pre-programmed set of instructions.
Even the concept of cross-sell and up-sell recommendations based on collaborative filtering is questionably AI. Yes, recommendations are mined using a form of machine learning (mining correlated items that are frequently bought or browsed together). However, the set of recommendations is typically pre-computed ahead of time, and is the same for all users who browse the same product – meaning everyone who fulfills a certain action sees the same pre-programmed response. This violates the definition of AI where the machine makes an independent decision without being explicitly programmed to make it.
Certain vendors boast that they offer hundreds of different strategies mined through collaborative filtering – but aren’t “strategies” just another word for a set of pre-programmed rules? How is that AI?
What if “Tay” utilized the same method as these B2C marketers? Instead of an AI that learned to develop “thoughts” based on the world around her (is it a technically a her?), Tay would have pre-programmed messages that would trigger based on the messages she was sent. That would mean the team that built Tay would have to program each and every message, create associations mapping incoming messages to pre-programmed responses, and build a way to fire those messages out automatically. Since Tay had hundreds of thousands of twitter-ers messaging her immediately, the pre-programmed messages would have resulted in a robotic, non-sensical string of thoughts that only related to the most recent outside influence. (On the bright side, if Tay utilized pre-programmed messages, internet trolls may not have ruined the experiment).
On the other hand, Jetlore’s Prediction Platform was built from the ground up as an Artificial Intelligence engine that collects information and creates complete responses from one singular learning model. Marketers can ask “what is the best content” and immediately receive the correct answer, without pre-programming or testing hundreds of interchangeable algorithms.
To learn more about Jetlore’s brand of Artificial Intelligence, and how we differ from the AI imposter, click here.