15 Artificial Intelligence Facts That Every Business Person Should Know

The business impact of AI is real. It’s propagating through industries right now, and it’s probably already finding a way into yours. According to McKinsey, 70% of companies will use some form of artificial intelligence by 2030. It will result in an additional $13 trillion of economic activity on a global scale.

Companies are rushing to be the first ones to realize that potential. And it pays off immensely. Deloitte says that 82% of companies that adopt AI today already see a positive ROI trend, with 17% as the median return.

The potential for these returns is horizontal, across many industries, from retail and manufacturing to energy and healthcare. So if you’re still considering whether it’s worth to explore an artificial intelligence project, remember that your competitors might already be doing it.

Will AI kill jobs? Are we all just going to bow to our silicone overlords? The truth is, we don’t know for sure. Yes, AI might eliminate jobs due to automation. But at the same time, it will create jobs elsewhere. It all depends on the industry and the level of its automation. But there are industries somewhere in the middle, like the postal service. Email is here. Postal service is here. There’s an equilibrium in many other industries to be yet discovered.

We wanted to focus on one of the limitations of AI because it’s a critical one. As a business, you can’t expect ROI from artificial intelligence, when you don’t have the "fuel" for it. Just like you wouldn’t expect a car to run without any gas. AI, just like humans, learns from experience. This experience in case of AI is the historical data that you’re supplying it with. AI-enabled systems use machine learning, which is a process by which the software learns through data. Many use these terms interchangeably.

You can’t expect good results from a system that’s been taught on bad data. That could be anything from bad quality of data to lack of data. To quote Doug Bewsher, CEO of Leadspace, "Data is AI’s Achilles’ heel."

So if AI is in your purview for the next couple of years, you'd better start thinking about improving your data quality and data management practices (architecture, metadata, privacy, and other elements).

Deep learning (DL) possesses a certain mystique. Many business leaders hear it being thrown around and think that it’s some sort of a magic pill for business problems. But this statement couldn’t be further from the truth.

First, deep learning is machine learning. It’s a subset of machine learning techniques. It sounds cooler; we’ll give DL that. But just like with any other machine learning approach, it has a ton of limitations. One of them is the fact that deep learning is a black box. It provides you with output, but you won’t be able to easily explain how it got to it. So if you work in a bank that’s building risk models, deep learning is a very bad choice, as it can’t be that easily interpreted for the regulators.

We talked about data before. Bias is another problem with data. Data is generated by humans. And we’re biased. Intentionally or unintentionally. Let’s take a real-life example. Tokyo Medical University recently lost its accreditation because it unfairly treated women who were applying for a degree there. They were disqualifying women because the consensus was that men would be better doctors, as they won’t take maternity leave and will be less likely to switch or quit to raise a child.

Now, imagine we tried to train an AI that predicts the best potential students for our university. If we took the data from Tokyo Medical University, the algorithm trained on it would pick up that men get accepted more, as it would just see it as an important factor for people being accepted. It would treat male gender as a positive factor and disqualify women. All because somebody was biased somewhere down the line.

There are plenty of examples like that. Even the biggest players, like Amazon, screw up with AI bias all the time. That's why you need to scan for these things, both before and after the system is operationalized to deliver insights, predictions, etc.

Yes. We talked a lot about data in the post. But guess what? Sometimes you don’t even need data. You can make it up. Sort of. It has to resemble real data as if it was generated naturally. It’s called synthetic data. And companies are using it in industries where data is lacking or is expensive to acquire.

We touched upon the topic of black box AI. This is a system that delivers output, but it’s hard to explain why specifically that output was generated. Why might this be important?

If you’re faced with a business problem, it’s very useful to know why data behaves in a certain way. For example, if AI provides you with a credit risk score for loan applications, you might want to know what data influenced the prediction of the score.

Sometimes, you are required to explain your output like in the case with regulations that we mentioned before. Federal Reserve Board, as such a regulator, requires banking institutions to explain their models.

Fortunately, there are many ways of making it happen for non-deep learning models through additional model/data manipulations. There are AI tools that already support interpretability features by default, making the process less technical and less time-consuming.

Sorry for the mouthful, but we wanted to keep it simple. If your team doesn’t understand how AI works, your AI project is likely going to be a failure. Many business leaders already understand this. That’s why, for example, Nokia’s Chairman made it a part of the company’s culture to learn about machine learning, the cornerstone of AI.

If you’re reading this article, then you’re on the right track. But you also need to deliver the message about the potential benefits of AI across your team and the organization.

If you’re looking for an experienced AI partner - reach out to Iflexion. We understand what it takes to operationalize an AI system and how an organization’s culture needs to change to accommodate such a huge technological shift.