The Enterprise Adoption Of Artificial Intelligence

Tamal Dutta Chowdhury, Senior Vice President - Artificial Intelligence, Course5iCourse5i is a global analytics & market insights solution provider offering Marketing Intelligence & Effectiveness, Digital Analytics, AI & Data Sciences, and Market Research Transformation services.

Artificial Intelligence (AI) is an umbrella term for multiple breakthrough technologies like Machine Learning, Natural Language Processing, Computer Vision, Motion & Manipulation, Knowledge Representation, Speech Recognition, and Machine Translation. These technologies, either on their own or in conjunction with one another, are driving a lot of the technological discourse, public interest and investments today. Despite the hype, studies have shown that many AI programs do not achieve their intended objectives, and the general AI adoption rate by global enterprises has been slow.

The Evolution of Artificial Intelligence
AI has been around since the 1950s, and has gone through several up & down cycles. AI has witnessed two major winters and some smaller ones, and several periods of hype and progress over the years. Most AI efforts till the first decade of this millennium were largely limited to academic and corporate research. It was only in the early part of this decade that AI started to emerge-out of the confines of research labs, and flourish in the world of enterprise adoption.

Many enterprises, especially large corporations, have started to invest significantly in AI initiatives. The phenomenal rise of enterprise AI, particularly since 2014-15, can be attributed to four key factors: (1) the availability of massive amounts of digital data; (2) a phenomenal in-crease in computational capabilities and the development of massively parallelized systems; (3) the widespread availability of advanced computing technologies; and (4) a significant increase in corporate spending on AI-centric theoretical and applied research. The Diffusion of Innovations theory posited by Everett Rogers explains, the Technology Adoption Life-cycle (TAM) Model where adopters of any new innovation or technology can be grouped into five categories: (1) Innovators, (2) Early Adopters, (3) Early Majority, (4) Late Majority, and (5) Laggards.

The current enterprise AI adoption is largely limited to the first two categories (viz., Innovators and Early Adopters). Companies that are in the Innovators category are generally `AI producers' (i.e., they develop AI technologies, systems and solutions), and to a certain extent, `AI Consumers' (i.e., they implement AI programs for their own internal use). Top technology organizations like Amazon, Microsoft and Google, and startups in niche AI areas (like computer vision or NLP) are good examples. Companies in the Early Adopter category are generally `AI Consumers' that have started implementing AI programs in select pockets within their organizations. Forward-looking companies in various domains like Banking and Financial Services, Healthcare, Telecom, and others generally makeup this category. In the past year or two, many large and mid-sized companies have started to join this Early Adopter category.
Best Practices for Enterprise AI Adoption
The global AI ecosystem has learnt several lessons, while strategizing and executing corporate AI programs. Here are a few key ones: first, companies must develop a formal AI Charter and an AI Strategy Roadmap before embarking on organizational AI programs. A key element is the AI resourcing strategy, which should include three key components: (1) the core AI research and development team, along with an extended engineering and domain expert team, (2) the physical and digital infrastructure needed for implementing AI solutions, and (3) the broader AI ecosystem comprising of suppliers, partners, outsourced vendors, and academia. All projects must be in alignment with this defined charter and strategy.

Second, modern AI systems are generally built using Deep Artificial Neural Networks (ANN/DNN) that consume significant amount of data. One of the biggest threats to successful AI implementations is the absence of adequate labelled data for training AI models. This aspect of AI is often not fully understood even by seasoned business leaders. AI adopters must focus on initiating systematic processes of capturing, integrating and harmonizing data during early stages of AI planning. In many cases, companies may also need to procure third-party data sets.

As AI technologies further develop and mature with time, they can enable new feature development, and greater granularity and accuracy levels that would not have been possible today

Third, AI is much bigger in scope, complexity & impact, than Data Science or Automation/RPA. Unfortunately, many companies fail to grasp this idea. Some make the classic mistake of hiring premium AI & Machine Learning talent only to assign them projects related to advanced analytics or traditional modeling. This is similar to buying a high-end automobile only to use it to drive to the neighbourhood grocery shop. AI is way more strategic, and can really elevate businesses to new levels through proper planning and execution. Hence, it is the key that companies utilize top AI talent for `genuine AI projects'.

Fourth, AI programs seldom get implemented in a linear fashion. For in-stance, there may be periods of low productivity and output, but followed by sudden periods of remarkable progress. Companies need to understand this aspect of AI implementation, and not take knee-jerk reactions with short-term failures. A long-term view is needed for most AI projects, and these can seldom be delivered in short-time frames. Moreover, companies should focus on developing and deploying reusable engines and models that can be utilized across multiple projects with minimal modifications. This will help reduce the overall timelines for project execution. While businesses do understand this idea, they rarely follow this in practice.

Finally, several corporates make the classic error of exploring AI technologies in isolation, rather than in conjunction with other technologies. Many use cases do not need the wheel to be reinvented, or things to be drastically changed for innovation to happen. For example, some automation problems may not even need AI solutions, but can be addressed with general software engineering.

Corporates need to consider two important factors to extract optimal value from their AI programs. First, they should aim at designing, developing and implementing `cognitive' or `intelligent' systems that can be used in the future as well (and not just today). Many companies tend to limit their AI systems to mere predictive capabilities and/or basic automation of operations as they exist today. Second, AI is an evolving technology. This means that AI systems should not be designed for static development and use. The idea is to not to stop after a particular AI solution is built, but to constantly improve on for the next one to three years through model tuning, upgrades and training. As AI technologies further develop and mature with time, they can enable new feature development, and greater granularity and accuracy levels that would not have been possible today.

As an emerging technology, AI is still evolving and is currently grappling with several issues that will take time to address. Some of these are: incorporating automatic continual learning capabilities into models; explaining algorithmic decision-making; consistency of predictive prowess, especially in dynamic environments; securing AI models and systems (apart from training and inference data); and domain-specific AI challenges. Hence, it is important for companies to initially narrow-down their focus and pick on projects that can be delivered with the least amount of challenges. That will lead to greater success in AI programs, thereby leading to greater and more efficient enterprise adoption.