Artificial Intelligence – what is it, really?

By Douglas Duncan

As a technology leader, it is my job to make technology understandable, actionable, and within reach of my organization. With thousands of technology providers competing for a piece of the market, all business partners get hit every day with materials touting Artificial Intelligence, or AI, solutions. Answers to your analytical problems! Make better decisions faster! Lower costs! How can our top business leaders understand the value of AI if you cannot give them a clear definition of what it means? I needed to sort this out for myself before I could make it clear to someone else.

One viewpoint I immediately rejected was that AI was intrinsically complicated. I could not accept a magic box. Perhaps the results are complex, but surely the components that lead to them can be broken down and understood? At the root, AI is about questions and answers, and critically, the decision to select which answer to use. While either the question or answer can be simple or complicated, the process of deriving answers for questions is rooted in logic, and logic can be reduced to basic building blocks. Answers that require a value system can also be modeled logically, and any view of AI must take this into account.

Logic Gate – “Boolean” AI

I am at heart a reductionist, and to me the core capability of A.I. must be something very simple. What is the most basic questions and answer decision? A coin flip.

Coin flips, and compound logical operations like if/then statements, are the basis of both elementary computer operations and elementary AI. They are one in the same, and the first dimension of A.I. is simply the logic gate. By adding iterations and efficiency, we are able to get calculators, standard computer operations, and much of what technology has to offer over the last 50+ years. This approach is easy to scale and is very linear. It does not necessarily answer questions that could not be answered by humans, but it does so much faster and with a high degree of accuracy. To the extent that human intelligence is a logical decision and the inputs are well known, this is indeed an artificial version of it.

Data Search – “Seek” AI

A quick geometry quiz: when you have a single dimension, like a line, how much “more” do you have when you add another line on the same plane, making it two dimensions? It is infinitely more, at least from the perspective of the original line. For AI, that next dimension is the concept of “matching” – the ability to match a question with a finite set of correct answers.

“Google” has become synonymous for information retrieval. Before Google, searching on the web needed to be fairly precise, and the hits were often based on the name of the file, not the data embedded within it. While Google’s secret sauce includes many things like intensive processing power, web content crawls, and access to personal data, what made it the Overlord of data search was its ability to match things together in a meaningful way.

Pattern Matching – “Synthesize” AI

The next dimension of A.I. that makes it infinitely more powerful than number crunching logic, geometrically going from a plane to a space, is this ability to find relationships between the question asked and the possible answers available. Matching success is strongly related to the power of the system, but the ability to organize the information through sophisticated indexing, organizing of data categories, and optimized processes is critical. Matching will rarely give a single correct answer, but it can provide a range of answers that are most likely to be correct. Like the logic gate, the ability to make matches augments human intelligence. AI of this form still is a shadow of organic intellect; a servant, impressive yet always docile as no decisions or values are yet in play.

IBM Watson had its day in the sun when it competed on Jeopardy in 2011 and soundly trounced two human champions. It took a brute force approach by combining powerful search algorithms with the ability to identify the most likely question for the answer given. Watson was not “thinking” in the traditional sense; it was certainly not “remembering” anything. How Watson succeeded was by being a super matcher, or correlator, of data. Still lacking, however, was originality of thought.

Being original means to introduce something new. A machine can be said to be approaching AI when it comes up with something we did not know before-hand. Beyond answering math problems, I would propose we have achieved a level of AI when a future outcome is predicted. It is important to draw a distinction between calculating the likelihood of different futures, such as an odds table for rolling a couple of six-sided dice, and making a specific prediction about an upcoming event. Admittedly these action are on the same continuum, but anyway that is my thesis… AI is not a specific point, it is an emergent activity.

Sophisticated weather prediction models, heuristic fraud analysis routines, facial recognition systems… these are current examples of AI taking probability analysis beyond building odds table to actually producing real-world results with a high degree of accuracy. These activities are not intuitive, they still require a mathematical relationship between the data points and must have a model to tie them together. But a key differentiator is that the “programmer” of the system is not necessarily dictating how the routine uses the data, nor how to correlate it. There is a learning element involved that lets the system be more than the sum of its parts. The model on which this activity takes place remains critical, and the closer it matches the particular reality you want to understand, the better aligned will be the results.

Future Modeling – “Cause & Effect” AI

What higher dimension of AI remains once you have achieved model-based prediction of the future? Let’s add two more concepts: persistent memory and value judgement/fuzzy logic. Persistent memory provides the capability to modify behavior based on concurrent activities in “real-time”, and to evaluate new activities based on what happened in the same session. Rather than drawing from a database of historical outcomes, a persistent memory constantly incorporates what is happening right now. The use of value-judgement, or fuzzy logic, means that not only can current observations influence how decisions are made, but the actual rules used to make decisions can also change on the fly, albeit within some framework.

A good example of this approach is a true self-driving car, one who is not on a pre-defined course and must deal with an almost infinite combination of environmental variables. Real-time voice chats, particularly those not restricted to a specific topic might fit this area, and it is not implausible to think about autonomous battlefield killer robots as needing this sort of capability to truly be effective in modern warfare.

The next dimension of AI?

Future trends are notoriously hard to predict, but I will venture to guess what comes next will first leverage what we have, only more so. I strongly suspect it will have an element of evolution in its design, and will be more cybernetic than purely artificial. And it is always safe to say it will be “quantum” in some way! Joking aside, however, it is certain that we have not yet reached the pinnacle of AI advancement.

Author’s Bio:

Doug Duncan is CIO of Columbia Insurance Group, a mid-sized Commercial Property & Casualty insurer based in Columbia, Missouri. Previous roles with Swiss Reinsurance and General Electric included Global Head of Workflow & Document Management, Head Applications Development, and I.T. Operations Leader. He is a Six Sigma Black Belt, and has the Project Management Institute PMP designation since 2001. His M.B.A. is from Colorado State University.

Doug’s writings cover multiple topics, including: Leadership, Project Management, Service Management, Insurance Analytics, Electronic Content Management, and Data Visualization. His philosophy and passion is in building a sustainable operating model and developing the next generation of leaders. He believes this requires fostering a culture of collaboration between I.T. and the business to provide the maximum in terms of leadership, business value, and operational effectiveness in the ever-accelerating business landscape.

Feel free to reach out to Doug on LinkedIn at: www.linkedin.com/in/douglas-duncan