At BOXARR we spend all of our time innovating, planning and executing ways to make BOXARR indispensable in solving the world’s most complex problems. A significant element of this innovation involves harnessing and embodying ideas, experiences and technology from Artificial Intelligence (A.I.) into our software platform. Traditionally, the term "A.I." has been difficult to define, but one way is to focus on what it fundamentally enables humans to achieve. In that sense, A.I. is any technology that enables humans to successfully cope with, and solve, “thinking” problems that are beyond human capacity without assistance.
Based on this definition, BOXARR’s approach to A.I. is distinct and innovative. Consider for a moment what I believe are the four fundamental, real-world approaches to delivering A.I. (as defined in this way). All of these approaches are valid, and each has its own merits to answer certain business, social, economic and engineering problems, but we believe BOXARR’s is the most powerful and unique.
There are three typical existing approaches that build one on the other:
The “Black-Box” Approach: This involves a powerful computer with clever software that receives a problem and delivers a hands-off answer. The underlying complexity of the problem remains only “machine-readable”, and the solution is delivered while the problem’s complexity is kept beyond the reach of human capacity. This approach offers useful answers to all kinds of automation opportunities, such as some varieties of image labelling, medical diagnoses, optimizations, etc. The pros of this approach to A.I. are that it automates an entire process. Its cons are the need to totally trust the solution and ignore any oversights or biases that may have been (accidentally) built into the black-box itself (note that these defects are inevitable, particularly in complex problems). What’s required to overcome those cons is a human in the loop, as in…
The “Big Data” Visualization Approach: This involves taking very large datasets, doing some “Black-Box” processing, presenting insights on (usually historical) trends graphically, and then (through the human’s perspective on those visualizations), predicting what’s not in the current data (usually potential futures). The pros of this approach are that they produce useful human insights. Its cons, in addition to those of the black-box approach, are that it relies on the idea that given data predict unforeseen data: in the historical context, that past behaviour predicts future events. While this can be a valuable rule-of-thumb, as any investor knows, it isn’t always true. So what’s required to overcome this is…
The “Clever Consultants” Approach: Ship in a room full of very clever consultants to extract data and develop insights (most often using secret tools that may be “Black-Boxes” and supplemental “Big Data” Visualizations, which the consultants know well enough to usually overcome the previously noted difficulties). The pros of this approach are that it often delivers quality insights. The cons are that it is usually very expensive (and when you stop paying the day rate the consultants leave, taking their secret tools with them). If after the consultants have left, and your problem changes in some way that aggravates the previously noted cons, you have no way of solving the issue (other than calling the consultants back to restart the effort afresh).
Then there is The BOXARR Approach, which is enabling what we are calling “Human Readable Complexity”.
Our mission is to enable the modelling of the most complex interdependent systems and rendering them into visualizations that are “human-readable”. Note that this is more than simply providing a black-box driven (usually statistical) visualization of an underlying complex system. It is providing tools (including A.I. and machine learning tools) that can be used to adaptively formulate visual summaries, driven by easy-to-implement-on-the-fly algorithms, that can operate on the objects in those visualizations, and can be continuously re-adapted and probed by any user. Allowing and encouraging the humans-in-the-loop to understand, interpret and transact on the complex system is the real enabler for effective A.I., which overcomes the inevitability of biases in black-boxes, the assumptions of big-data-based predictions, and the segregation of expertise inexpensive external consultant gurus.
The pros of this approach are that it keeps humans (as many humans as are needed) truly in the loop. The cons are a need to invest in the modelling, but that itself is mitigated by human readability, in that key people can be effortlessly enlisted into the model, since productive modelling becomes more effortless and collaborative. Those people then become hugely more productive and valuable to the organization. Delivering much more value with less resource.
We believe that striving for effortlessness in modelling is the key to real A.I., under the definition given: that A.I. is about giving humans super-human capacity in coping with, and solving, “thinking” problems of enormous complexity.
We are proud of contributing to the advancement of the state of the art of A.I. (and machine learning) with the different and innovative approach of Human Readable Complexity. We believe that the technologies we have delivered, that are continuing to develop, and the patents we are filing, will contribute to helping the world deal with this emerging and hugely challenging new complexity, bringing the real promise of A.I., its promise to expand human ability, to fruition.
Written by: Alasdair Pettigrew, CEO