AI - How to start thinking about it

Business executives today hear a myriad of buzzwords every day both at the office and home – AI, machine learning, analytics, etc. Like any other wave, they are trying to discern what it means for them, their business and their vision. It is becoming certain that there is some truth in this buzzword. Intelligence in systems is growing and its impacting business performance.

Given this, when an executive wants to implement AI, he/she is faced with a tremendous challenge in understanding what can be done, the feasibility of options, technology choices, and approaches to execution. This challenge is driven by rapid growth in technology, futuristic headlines and articles, novelty and lack of prior knowledge. Both buyer and seller of services face similar problems of defining scope & expectations. The expectation mismatch between a business executive and technology executive is huge and driven by a lack of understanding of boundaries, dependencies, time-warped understanding of AI (futuristic vs real), which makes it difficult to separate fiction from fact.?

There are two extreme points of views –

1. AI can solve every problem
2. AI cannot solve any problem and is not applicable

Executives, both sellers/doers, and buyers have their realities at multiple points between these two extreme views. The success of any business AI implementation is driven by informing those positions with an accurate understanding of your organization & its problems, business goals, combined with knowledge of AI methods.

Based on my experience spanning two decades in this industry, I take a simple framework of learning constraints to determine what a business executive can do to understand the potential for AI in their organization.

A simple model for intelligence development is -

Data + Algorithm = Intelligence

Key factors which determine intelligence outcomes are –

1. What data is getting generated & how?
2. How much task-specific knowledge is required to train the algorithm? Is this knowledge very specific to individuals or is widely available?
3. What is your optimization goal -

• Human task replacement
• Human + Machine output maximization

If a business executive has unconstrained data generation, widely available task knowledge and limited specificity to organizational context, he/she is can build unconstrained intelligence limited only by scientific progress. Typical examples of these problems are replacing human functions (cognitive/sensory etc) by computerized algorithms. There are no constraints on intelligence development. The limiting factors are infrastructure capacity, scientific progress, etc. Examples of such problems are computer vision, computer-generated speech, voice translation, etc. The executive approach, in this case, should be to focus on adopting technologies from large ISVs.

If the task requires both human cognitive/functional abilities and process-specific knowledge, a typical approach is to adopt AI-driven intelligent automation. A successful approach is to sub-divide automation goals and find methods to replace individual chunks of tasks using AI. Examples in these cases are automating document processing, reading customer reviews, routing emails, etc. Business executives need to pay serious attention to this area as it will become one of the most important cost reduction levers for organizations.

The third group of AI applications utilizes data generated from production systems. These data sources are structured/unstructured, fragmented, siloed and scattered in different parts of organizations. Business executives need to understand the limitations on usage of Advanced AI in these applications. These apps are constrained by efforts required in data management, complex task needs, need for organizational context understanding, etc. A successful approach in building intelligence is by adopting a use case driven build-out of data, analytics, visualization, and AI capabilities. Typical requirements in these cases are not fully known and organizations need to adopt a discovery led requirements creation approach. The executive focus in these cases should be to increase machine + human intelligence and not on substitution.

It is very important for business executives to differentiate their needs and set the right AI vision. For questions, please reach me at [email protected]