The Practical Aspects of AI Implementation Part 2

March 2018

I believe that Charles Darwin, who is best known for his contributions to the science of evolution, never thought that his theories may be also applicable for AI-based solutions (with some modifications and adjustments of course).

A couple of weeks ago I shared with you the first part of ‘The Practical Aspects of AI Implementation’ in which I focused on the first step in the process: The ‘Problem Discovery’ phase.  This phase in a typical AI related project debate is in essence about finding the ‘right’ business problem to solve utilizing AI. I would now like to fast forward to the final step, which is often ignored and neglected with dire consequences.  After the AI solution has been released, there is a constant need to revise, review and evolve according to three main environmental and market conditions which we will discuss below.  This phase can be referred to as the ‘Extension & Expansion’ phase.

Releasing your AI-based Solution is only the First Step

After a solution is in place and well deployed, it will naturally begin to evolve. Don’t get me wrong. I’m not talking about products enhancing their own intelligence and autonomously building their next-generation child-product-machines and conquering the world as we know it. We are here to discuss something much more practical and refined.

The Three Forces

Extension and expansion are patterns of solution growth. There are directions in which this type of progression takes place and in reality, combinations of these ‘orthogonal’ directions are typically met:

  • Environmental – adjustment to environmental changes
  • Market – modification of the existing solution to meet new features demanded by the existing market
  • Expansion – adaptation of the solution for new markets in order to expand beyond your current scope

AI Implementation 3 Forces

Responding to Environmental Change

 The ‘Environmental’ direction which seems the most obvious one may be the most difficult. This refers to changes happening spontaneously in the environment which might affect the AI solution, within its current scope and market.  Many AI algorithms rely on training, meaning that some data sets are used to ‘teach’ and refine the internal processes of the solution. On the other hand, the actual world we are operating in is dynamic in nature and if the core technology is not flexible enough (to re-learn), they will not be able to adjust to new conditions which may arise.

For example, let us examine an insurance company using an AI-based software application which determines car insurance terms and conditions based on some automatic risk assessment of applicants and their vehicle types. The software app needs to be sensitive to changing environmental conditions – it is quite common that certain areas are much more prone to vehicle theft and the vehicle type, model and manufacturer also impact the related statistics. Of course, the car owner and related driving ‘capabilities’ are also part of this equation especially on the tendency to reckless driving. All of these parameters influence the insurance terms and unfortunately may change overtime. New models become more popular among thieves due to trends or just the cost of automobile parts. New city neighborhoods will impact demographics and may influence the likelihood of car thefts. Even the assessment of driver ‘trustworthiness’ may change not even due to intrinsic nature but for example the drastic rise in vehicles per road mile in the recent vicinity. If the AI models and algorithms used at the heart of the app are too rigid, it may set insurance conditions which are either too generous or too expensive ignoring new situations. The app needs to implement some KPIs (Key Performance Indicators) which can sense either by external means or by internal metrics the fact that something has environmentally changed and may enforce model re-adjustments and re-apply it gradually as the market changes.

Staying Ahead of The Market with New Features

The ‘Market’ direction imposes new functionalities missing with the current AI-based solution. This refers to new features you identify will be in demand or are already demanded by the existing market.  For example, the current OS (Operating System) of a company building autonomous cars expects that the terrain types for driving are either urban, highway or semi-urban (part urban part highway). If you want to introduce a new terrain type as ‘off-road’ for your 4×4 communities, then this will set some new challenges. For example, with this terrain type, the road itself is not clearly defined hence relying on some vision contrast algorithms to detect road borderlines may be completely ineffective. It could be that you will need to develop and use some path smoothness and regularity measurements instead which will determine if the current path is adequate enough or if deviations maneuvers are required.

Autonomous Vehicles

Taking your AI Solution to New Markets

The ‘Expansion’ direction is mostly taking the current AI-based solution and applying it (or some of its capabilities) to a completely different domain. This refers to expanding beyond your existing market to new sectors or verticals.  Going ‘Horizontal’ in many cases may bring you outside of your business practice boundaries. You should proceed with caution.

For example, in the previous case of the car insurance app, if you are an insurance company which developed this AI-based app, then applying the core application capabilities to a completely different industry may be of no interest. However, as this is seldom a black-and-white situation, even in this case it could be that there are other departments (within the insurance company) may benefit from this development effort. For example, evaluating the trustworthiness of new clients in general, is an essential assessment processes for other insurance related offerings yet it may have different requirements and will tend to emphasize and de-emphasize certain aspects.

While the ‘Environmental’ and ‘Market’ directions maintain more or less a similar business course as the original solution had, it is the ‘Expansion’ direction which opens up opportunities that are usually ‘off-course’. Although tempting, instead of deviating your business to unknown territories you may choose to proactively examine the applicability of your existing AI-based solutions for other departments or other aspects of your core competencies.

Only the Most Adaptable Will Survive

 The development of products and solutions that meet the right markets is a never-ending story and with disruptive, new technologies such as AI at the core it is necessary to acknowledge the possible changes and transitions and be prepared for the impact. Some changes will require new features. Some will raise questions of applicability as different markets are considered. However, at its heart, and even without any new feature request or new market quests, the AI components must dynamically adapt to the continuously varying environments. A famous quote which is mistakenly attributed to Darwin can be ‘adapted’ for this case – ‘It is not the strongest of the AI solutions that survives, nor the most intelligent that survives. It is the one that is the most adaptable to change.’

This post was written by Jay Klein, CTO of Voyager Labs. Jay spent 30 years heading technology innovation fronts of leading companies around the globe. He is a multi-disciplinary tech evangelist and thrives on hunting industry related ‘tornados’.

Share this post