Floridi Curves

An Approach for Taming AI Problems in Industry

Artificial intelligence is transforming industries worldwide — from healthcare and finance to automotive and manufacturing. Yet consistent implementation success remains remarkably elusive. Recent comprehensive studies have documented alarming failure rates for AI initiatives, frequently exceeding 80% across various sectors. It appears that these failures rarely stem from algorithmic inadequacies, but rather from fundamental misalignments between AI capabilities and improperly structured supporting environments.

How can managers who have made or who are actively considering an investment in AI know where best to apply it to greatest effect? The author draws on the work of philosopher and Yale Professor Luciano Floridi to help managers classify what kind of problems they’re facing along what the author calls the Floridi curve: mapping problem “complexity” (requiring higher human reasoning and judgement) against problem “difficulty” (requiring increasing levels of computational power). Floridi also provides a way to understand the environments or contexts that make AI work successfully, what Floridi terms their “envelopes.”

The “aha” here is a comprehensible methodology to increase the efficacy of AI in service or product design and implementation by better understanding both the intrinsic nature of the challenges contained in the design process and then how to apply AI at its most powerful to bring those service or products to completion. The key is a systematic progression from the first minimally viable service / product and what the subsequent iterations might be as managers / designers shift a service / product down its Floridi Curve, requiring fewer human skills but more AI complexity (as well as understanding how much should remain in human hands).

The article begins with a definition of terms and concepts drawn from Floridi’s work. It walks through the classification of problems, mapping those problems against complexity and difficulty and defining a Floridi curve, and “enveloping” those problems in current and future design / product states. Successful implementation begins with these foundational steps.

To illustrate these concepts, the author takes a design challenge from the auto industry: the development of a semi-autonomous car. He maps the Floridi curve for associated problems and then describes the systematic progression from currently viable products / features to future states (along with dependencies). The article includes graphics illustrating the Floridi curve for these products / features as well as for the “enveloping” and design progression across years zero to two of the design and implementation process.

The article concludes with these concepts in actual practice with a client in the textbook industry to improve textbook forecasting. This case study breaks the application of these concepts into five steps for managers / designers to consider: conceptualizing solutions to the problem; verifying AI model and envelope feasibility; incorporating a checkpoint evaluating the economic feasibility of the approach (a critical and often overlooked step); designing necessary system components; and implementation.

The overall framework described throughout this paper provides for clients, designers, and developers a mutual understanding of the key AI models and envelopes that must be built, and when, to achieve the desired outcomes. The author believes it is a useful framework for any industry, although each industry will have its own well-honed methods for bringing the actual AI to life. Car companies will collaborate with parts manufacturers, robotic warehouse designers will collaborate with their building architects, and textbook companies will collaborate with their IT departments and data scientists.

Some of the benefits of our framework include:

Clarity on which AI models and envelopes must be built and when.
The timeframes for development (model years, etc.) which form the basis of a multiyear rollout plan.
Strategic budget development - the ability to forecast the conceptual costs in labor, skills, construction, and implementation costs.
A checkpoint in the methodology for deciding if the gain in AI improvements actually provide the hoped-for return on investment (step 3 in our methodology).
A framework that can be universally applied across industries.
A means of achieving a shared vision across disparate groups when developing the Floridi progression.
An approach that is easy for any development team to adopt yet is grounded in the strong theoretical foundations of Floridi’s philosophy of AI.
An easily understood communications framework for boards, CEOs, and CFOs.

Applying Floridi’s concepts in this approach can significantly improve the likelihood of success when implementing AI solutions for business because doing so sidesteps the biggest risk in AI projects, which is overestimating the return on AI while underestimating the necessary envelopes that must be built to make them successful.

Click here to download the full article