Three Barriers to AI Adoption.
Some organizations are still questioning the business impact and benefits of this technology. It is key that businesses understand what Artificial Intelligence or Automated Intelligence AI can and cannot do – there are several main barriers to AI adoption which exist today.
Skills is the first of these barriers. Business and IT leaders alike acknowledge that AI will change the skills needed to accomplish AI jobs. Currently, for example, AI can evaluate X-rays like human radiologists. Similarly, Machine Learning ML also has a learning curve.
As this technology advances beyond research environments, radiologists will have to shift their focus to consulting with other physicians on diagnosis and treatment, treating diseases, performing image-guided medical interventions and discussing procedures and results with patients.
The fear of the unknown is another factor delaying AI adoption in some cases. According to Gartner’s 2019 CIO agenda survey, 42% of respondents do not fully understand the benefits of implementing AI in the workplace.
Quantifying these benefits poses a significant challenge for business and IT leaders. While some advantages of AI adoption, such as revenue increase or time savings, are well-defined values, notions such as customer experience (CX) are difficult to quantify or define with precision.
Any success of AI can only be determined by taking both tangible and intangible benefits into consideration. By 2024, 50% of AI investments will be quantified and linked to specific key performance indicators to measure return on investment.
The third significant challenge creating a barrier to AI adoption is full data scope, or the data quality derived from AI. Successful AI initiatives rely on a large volume of data, from which organizations can draw insights and information about the best response to any given situation. Businesses know that without sufficient data – or if the situation encountered is at odds with past data – then AI falters. Others are aware the more complex the situation in question, the more likely the situation will not match the AI’s existing data, leading to perceived AI ‘failures’.
Data volume is not the only or most important factor to consider, however. Many successful Use Cases UC can be achieved using a reasonable amount of data, provided that the dataset is good quality (i.e. normalized, complete and diversified). A lack of volume can be compensated by a reduction in project scope, while a lack of data quality will invariably lead to Proof of Concept PoC failure.
It is important to remember that “a reasonable amount of data” will mean different things, depending on the AI technique being considered. Probabilistic reasoning techniques such as Machine Learning rely heavily on data to deliver insights. This is where data quality problems are most acute. The same is true for Natural Language Processing NLP systems.
Beyond data quality and completeness, CIOs must also understand the sustainability of that data. For example, are the sources of that data anecdotal or systematic? Can the data be obtained on a sub-second basis, daily, weekly? This is a crucial consideration for the potential scalability of the PoC.
The more organizations that implement AI, the more jobs it will create. These jobs will fall into two categories: jobs directly related to implementing and developing AI in the organization, and secondly jobs created by the opportunities for scale provided by AI.
Overall, AI will not eliminate jobs. Whether “Automated” or “Artificial” Intelligence or “Machine Learning” ML & AI is set to become a net-positive job motivator by next year, eliminating 1.8 million jobs but generating 2.3 million. Organizations should be aware of the opportunities and challenges posed by AI if they are to overcome the existing barriers and deploy AI successfully.