With its increasing popularity, AI (artificial intelligence) has received tremendous media coverage. We have been bombarded with messages hyping AI’s capabilities. We are repeatedly told that AI will replace many jobs. It is even being said that AI may eventually eliminate almost all jibs, making human labor essentially unnecessary.

As a result, more and more people wanting to keep up with the latest technology have been trying to implement AI in their work. Many of these people know little about AI, but they do knew it is an up and coming technology that they heard so much about and feel they must start using.

So, implementing AI is exactly what they do. And, because they’ve heard so much about AI’s wonderful capabilities, they have high expectations for AI. Unfortunately, instead of getting the expected phenomenal results, they may end up disappointed. This occurs because they are trying to use the AI for something it is not capable of doing, since that’s not the type of data that the AI was trained on.

Unlike humans, AI cannot think. So, although AI systems can identify patterns in the data, AI is not able make inferences. If the AI hasn’t been trained on the type of data associated with the problem, it won’t have the knowledge to come up with a good solution. That’s why AI often works so much better when it is combined with human input. And, that’s why AI results can be disappointing when the system used was not trained on the kind of data needed to accomplish what the AI was asked to do.

A good example of this was discussed in a recent Wall Street Journal article about Toyota’s efforts to use AI for automobile design. According to the article, Toyota was using an AI system that was built to design art and images. Toyota tried to use this AI system for automobile design. Toyota asked the AI to design an automobile with a sleek and masculine image. This request may have seemed reasonable since they were using an AI system built for art and images. Unfortunately, Toyota ended up being disappointed. Toyota’s reaction to the AI design was that they’ll have to train the AI in aerodynamics.

The lesson here is that AI can only produce what is related to the data it was trained on. The AI system Toyota used was trained on data for designing art and images. Although automobile design does require artistic capabilities to create the desired image, it takes far more than artistic skill to design a well-functioning vehicle. Automobile design also requires skills in engineering areas, such as aerodynamics. But, the AI system Toyota was using had not been trained on data related to engineering. Thus, the AI Toyota used was not able to design a well-functioning automobile.

As I said in a previous blog post, AI is just like other types of data analysis in that the kind of data that goes in is the kind of data that comes out. It doesn’t matter whether the data analysis entails AI, or an Excel spreadsheet, or a consumer survey, or a placebo controlled trial, or any other kind of data analysis. If you have data about white males, the results won’t apply to women or to non-whites. And, if you have data about art and images, the results won’t apply to engineering.

That’s why it is so important to pay attention to what types of data an AI system has been trained on and to use AI systems trained in areas that fit what you want the AI to do. Sometimes, however, it may be necessary to apply trial and error to see what the AI system is capable of. But, paying attention to what the AI is trained on can help prevent disappointing output.

So, It’s important not to get too wrapped up in the hype we are hearing about how AI can replace most humans. Granted, there is a possibility that may eventually happen in the future. But, we are not there now. Today's AI systems are not capable of doing everything. Thus, it is important to think about how the AI you’ll use was trained, and whether or not it is an appropriate tool for what you are doing. AI can be useful as long as it is a system trained for what you are asking it to do.

As an ending note, I’ll point out that I am writing the above as someone with considerable experience in data analysis. Early in my career, a position I held was that era’s version of what today is called a data scientist. Additionally, a peer reviewed data analysis article I wrote for a market research publication was cited in at least one academic textbook on AI.

Please contact us if you would like an introductory presentation on AI.