Artificial intelligence: Learning from insights

At one of my latest presentations on artificial intelligence (AI), a member of the audience asked me about the difference between business intelligence (BI) and AI. As BI and AI use the same statistical techniques, why can’t we use BI where AI is just hype?

And frankly, he’s right. BI, data analytics, and AI use the same foundations: statistics. AI relies more on correlations than the other, but that’s not the crux of the matter. When we use BI to build predictive models, the outcomes are probabilistic. You get outcomes like: “With 80% certainty, 90% of my customers will like my new ACME product.” When only 70% of my customers wind up liking them, it still fits the predicted outcome. Statistically speaking, that is.

And AI uses, broadly speaking, the same type of reasoning. It uses the same insights that can be derived from BI methods. But AI does more. And that’s where the differences between BI and AI come to light.

Cirrus Shakeri, CEO at Inventurist, uses three dimensions to define the capabilities of AI-systems: autonomy, learning, and reasoning. I quite like this model, because it points to the direction in which AI is evolving. When a system scores high on all three dimensions, we truly have an intelligent system.

Business intelligence can gain some leverage on the reasoning dimension. By using advanced techniques, we can establish a reasoning model using BI, for example, “when the sun shines, we sell more ice cream.” This is, of course, a very basic predictive model, but it can help retailers stock more ice cream when the weather forecast predicts lots of sunshine.

I won’t discuss the statistics behind these kinds of models. The baseline is, when the model works it’s all right. And the same is true for AI models. The block box nature of most AI systems doesn’t give us insight into how it actually works. BI models are much more transparent than AI-based models.

Where AI should excel is on the learning and autonomy dimensions. From my point of view, learning is currently the most important dimension to focus on. With BI, the statistical models will always give the same output when using the same input. With learning models, that output will evolve due to the learning process. We all hope the learning process will improve the accuracy of the predictions. But since we can’t guarantee that, we should expect the unexpected when using AI at full depth.

Ron Tolido, CTO at Capgemini, predicts 2019 we’ll see a rise of autonomous AI. Notwithstanding the current discussion around the ethics of AI, I believe autonomous AI will take off because of the immanent economic advantages of autonomous systems. We currently see the autonomous behavior of AI in robots, self-driving cars, and so-called lethal autonomous weapons.

But learning AI has another kind of autonomy. We can’t really control the way AI learns and how it obtains insights from the data we feed it. Many AI applications, such as the Microsoft bot Tay and Amazon’s recruitment system, failed because they learned things we didn’t find acceptable.

So, that leaves us with the final question. Should I use BI or AI to create predictive models? That depends. Let me start with the statement that old-fashioned statistical models, like the ones used in BI, can render good results. When these results are good enough for your business case, there’s no obvious reason to start a long and winding AI learning process.

For AI, the business case is of utmost importance. AI should really fit the case. Don’t try to squeeze AI in because it’s fashionable. If you don’t have the business case, stick to the well-proven algorithms and statistics. Moreover, your organization should be able to work with AI. To put it boldly, the learning characteristics of AI should fit the culture of your organization. I strongly believe that a learning organization can get more benefits from AI. The basic idea is that an organization that learns can consistently transform itself to gain a competitive advantage. Learning organizations can cater for the need for continual learning of AI by feeding the AI with new data, giving feedback on the results of the AI system, and adapting and training the AI for new situations – and, most importantly, actively acting upon the outcomes of AI within your business processes. Only when you can fully exploit the learning and autonomy of AI, will you reap the benefits of AI.

If you want to use AI, you should not only look at the technology. Your business, your organization, your company culture amongst others should be involved in your investigations. You should ask yourself what, how and when will I use AI? But before you answer these questions, pose yourself three other questions: “What is AI, can I use AI, and can I handle AI?”