Getting the most out of AI

Are you unsure about how to benefit from AI? You should be!

We have all heard the stories of “AI” being used to diagnose cancer, drive cars, and even detect fake news. But despite all the hype that’s been generated by the big names in the industry, it isn’t always easy to see how these technologies might help your business.

At the same time, there are significant technological breakthroughs hidden behind all that is lumped together as “Artificial Intelligence”. In particular, these include advanced algorithms that perform self-configuration by adjusting their own rules (e.g. decision criteria, or statistical characteristics such as thresholds, weights, coefficients, etc.) during their processing. This relieves us of the need to define and specify all the detailed rules before they are executed, and just as well, as the extent of the full set exceeds human capabilities. The downside is that these self-adjustments reach the point where we no longer know how they work.[1] Without upfront specifications and detailed rules, there is no way to “reverse engineer” the obtained results.

In addition, let’s bear in mind that as long as intelligent algorithms are designed by human beings and not self-generating, they are still just “mechanical calculating systems” with predictable limitations[2]:

• It is impossible to guarantee the absence of bugs in a computer program, whatever its algorithm!
• As tempting as it may be, it is misguided to believe that insight or invention can “emerge” from any pseudo-random number generator.

And this will remain true, despite the predicted increase in machine performance and volumes of data. We assert that AI is just the natural continuation of improvements to Information Systems with new algorithms.

However, human beings, unlike machines, being capable of thought and knowledge, don’t have any such limitations. Humans are able to:

• Invent new solutions to existing problems, by thinking outside the box;
• Be “lucky”, coming upon a breakthrough by accident;[3]
• Be creative, putting together seemingly unrelated ideas to solve intractable problems;
• Produce models of reality at various levels of abstraction and express tacit knowledge within any professional domain.

Moreover, intelligent algorithms do not exist in a vacuum, but are always part of some larger system including people. In such a system, the people and the machines interact in an integrated way. The purpose is, by definition, to help humans (including customers, staff, suppliers, etc.) better perform their tasks through the use of advanced technology. It would be a serious mistake to consider people and machines as “interchangeable parts” of such a system.

In today’s world, implementation of several successive generations of IT systems has left very few purely manual tasks in most business environments. Most tasks are mixed, either human-centered aided by machines or machine-centered overseen by humans. This requires us to focus on User Interfaces (UI, often broadened to a concept called UX for User Experience). Regardless of the underlying technology, the UI must always be controlled and driven by the human!

Specifically, when implementing a natural language UI (e.g. chatbot for textual input or intelligent virtual assistant contained in a smartphone or home speaker for vocal input), it is essential to begin by eliciting and specifying the key tacit knowledge of the business domain in order to delimit the applicable conversations (e.g. can’t order a pizza from a medical service), as well as to customize the dialog (e.g. a medical app recognizing that “temperature” and “pressure” most likely pertain to the human body and not to the weather). This will provide true added value to the human users and contribute to implementation of high performance people-machine systems.

A final note about an AI project: regardless of the algorithms involved (e.g. process automation/robotization, conversational assistants, robots, self-driving vehicles, etc.), the standard project practices remain valid, even (or especially!) in “agile” or “lean” modes.

[1] For example, see MIT Technology Review, “The Dark Secret at the Heart of AI” https://www.technologyreview.com/s/604087/the-dark-secret-at-the-heart-of-ai/

[2] For more details, refer to the work of Turing, G?del, Chaitin, and Kolmogorov.

[3] 9 such inventions are described here: https://www.inc.com/tim-donnelly/brilliant-failures/9-inventions-made-by-mistake.html