Introducing authentic AI: Making artificial intelligence effective
Artificial intelligence is being celebrated as the innovation that will change the world. And while it undoubtedly has a multitude of applications and uses, it’s worth remembering that one size does not fit all. When a company looks to deploy AI technology, there are many business-specific challenges, so making the right choices can be tricky.
For example, just recently, yet another AI-related breakthrough was announced: A robot dog learned to open a door to allow another robot dog to walk through it. While it is well-acknowledged that the invested research and development for this mission was huge and the commercial potential for some applications is enormous, it is somewhat unclear how this specific innovation or the core models and algorithms of it can serve other industries and verticals. Herein lies the problem.
Gauging AI success in one field in many cases can be meaningless for another. To make things worse, even when trying to go deeper into the technology and attempting to evaluate, for example, which machine learning algorithms are utilized by the product, or what are the number of layers in the deep neural network models mentioned by specific vendors, in the end it will be possibly pointless as it does not directly reflect the technology deployment “success” implications.
Nevertheless, it seems that the market ignores this reality and continues to evaluate AI-based products by buzzword checklists using familiar and related AI terminology (e.g., supervised, unsupervised, deep learning and so on). While checklists are an effective tool for comparative analysis, it still requires the “right” items to be included. Unfortunately, what typically is absent are the items which are important to the customer, from a problem-solution perspective.
Introducing authentic AI
Given all of this, there is a need to change the narrative around AI technology to something meaningful and authentic that reflects the real-life challenges and opportunities that businesses are facing. This is the time to introduce authentic AI.
The Merriam-Webster dictionary defines authentic as both “worthy of acceptance or belief as conforming to or based on fact” and “conforming to an original so as to reproduce essential features.” This is not about fake to be contrasted with real; it’s about the essential features of AI which need to be acknowledged, and hence, redefine the “checklist.” Often, these essential “authentic” features are hidden and only surface when a CIO or CDO is faced with a new problem to be solved. This is seen especially when the AI aspects of a proposed product are fully explored by asking questions such as:
- Is the AI technology utilized by the product aimed specifically for my problem, optimally (e.g., performance, cost, etc.)?
- Is it capable of addressing the complete problem or only a part of it?
- Can it be assimilated into the existing ecosystem without imposing new demands?
- Can it address the compelling environmental conditions of the problem space?
These issues can be grouped into three different “classes:” original, holistic and pragmatic.
Original — How innovative is the solution? This can be quantified by assessing the following:
- The invention of new algorithms or even new models;
- The use of complex orchestration techniques; or
- Through the capability to handle complex data formats and structures.
While there is no need to reinvent the wheel repetitively for any problem, there are distinctive characteristics which require optimizing.
Holistic — How complete is the proposed AI technology? It takes into account the capability of handling the end-to-end aspects of the offering, the competence of harmonizing the operation of the various AI components of the technology and the ability to adapt to ever-changing conditions of the AI application.
Pragmatic — Can the technology solve real-world problems in their actual and natural space in a commercially viable way? This means that, for example, the data sources can be processed in their most native format (unstructured or structured), as well as provide insights or results matching the pragmatic needs of the specific market expectations. In addition, the ability to be quickly deployed and rapid to act are assessed.
All of these elements should be used to systematically assess and evaluate AI-based products and technologies to assess their authenticity and therefore effectiveness in specific use cases.
For example, many home loan mortgage evaluation and recommendation systems utilize a somewhat isolated machine learning-based applicant classification method, one of many processes included within the product. The AI in this system cannot be considered authentic AI to a high degree as it scores low on the original and holistic classes as it isn’t innovative enough (from an AI sense). In addition, the AI component itself does not cover on its own the end-to-end aspects of the technology (hence, affecting the overall performance and precision). It could be considered to be pragmatic to some level if it can handle the required data sources of financial institutions or the customer applications natively, and if the technology’s “outputs” are the explicit results required as a specific recommendation (e.g., loan conditions). However, the deployment timeline (time-to-market) and commercial aspects need to be evaluated as well. This is just one example of many others, covering all kinds of variations.
So, in the case of the door-opening dog, although many people heard about it, its application is fairly limited — in fact, you could say that its bark is definitely worse than its bite.
All IoT Agenda network contributors are responsible for the content and accuracy of their posts. Opinions are of the writers and do not necessarily convey the thoughts of IoT Agenda.