So far, Artificial Intelligence has yet to live up to the enormous expectations which have always surrounded this technology.
Its failure to garner mainstream adoption during its inchoate inception in the 1960âs and 1970âs kept it out of the headlines until relatively recently. The surging interest around cognitive computing and self-service platforms such as IBM Watson has set the stage for more demands than ever, perhaps culminating in the multitude of prognostications from Gartner, Forrester, and others regarding AIâs ability to transform the data sphere in 2017.
With an array of self-service machine learning platforms, cloud-based data science solutions, and significant advances in both storage and compute power, machine intelligence is certainly more accessible to the enterprise that it has ever been.
But in practice in the business world today, AIâs moment is still impending, and not yet realized. According to Cambridge Semantics VP of Financial Services Marty Loughlin, approximately 90 percent of the attention focused on AI within the finance industry is media generated, versus only approximately 10 percent practical usage. âThe interest is definitely high, but itâs a little bit ahead of reality I think,â Loughlin commented.
Theory Versus Practice
Although AI includes everything from basic Natural Language Processing capabilities to some of the more advanced options for virtual reality and augmented reality, it is perhaps most practical to the enterprise via the automation capabilities of machine learning and deep learningâwhich in turn fuel efforts towards speech recognition and other forms of pattern recognition. The potential of these AI components to drastically improve analytics, data preparation, and even elements of data modeling is considerable. Traditionally, their utility to the enterprise was girded by:
- Time Constraints: Organizations can spend more than six months perfecting the predictive models for machine learning or deep learning.
- Data Quantities: The predictive capacities of both of these facets of AI require exorbitant amounts of training data, contributing to the lengthy times required to fine-tune them.
- Data Scientists: Data Scientists are arguably the chimera of the 21st century, that rarest of commodity with illimitable demands on their time.
- Infrastructure: Costly hardware was previously required to accommodate the massive data quantities necessary for training machine learning and deep learning models.
Self-service SaaS offerings which have automated critical aspects of data science prerequisites for these manifestations of AI, along with accelerated processing speeds, have rendered most of these concerns obsolete. Nonetheless, actual use cases for AI in transformative roles impacting business valueâas opposed to rudimentary forms of machine learning in comprehensive platforms for analytics, transformation, or data preparationâare limited due to more modern concerns.
Consultancy: Redressing the Data Science Shortage
Exploiting AI to effect competitive advantage requires leveraging it in the most vital business tasks, as opposed to simply quickening portions of everyday data management. Oftentimes, finding that niche requires more than a simple cloud-based deployment and necessitates substantial consultant work–which may simply occur with SaaS and PaaS providers functioning as consultants while aggrandizing costs. The Senior Vice President of Strategy, Research and Analytics at Shapiro+RajÂ (which specializes in assisting clients with Bayesian machine learning methods) Dr. Lauren Tucker remarked: âPeople who can do the modeling and so forth have very little understanding of how to connect the dots to craft a story around the inisghts to come out of those models. But then you need people who understand how to present them, how to get that information originally from the client, and how to do the model assessment. It takes a village, and that village is more often found in firms that focus on that type of business rather than holding them in-house.â The implications are that the temporal and financial costs for using AI are often more than the rapid self-service cloud offerings may initially seem to be.
Unstructured and Structured Data
One of the principal usages for AI is in assisting organizations with the abundant amounts of unstructured data they are accounting for, which is a direct consequence of the normalization of big data to the enterprise today. âThe amount of data that you actually put through the ETL pipeline and structure is only the classic tip of the iceberg,â indico Chief Customer Officer Vishal Daga said. âThereâs so much data out there thatâs of the unstructured variety.â Analyzing unstructured data has its own innate peculiarities, which may vary according to industry. âA lot of that analysis, if you can use machine learning or AI you can get into something where you can actually do the interpretation,â Biotricity CEO Waqaas Al-Siddiq said. âBut itâs a different data type. Itâs what IBM Watson is always talking about. Youâre not looking at traditional data, youâre looking at unstructured data.â
However, there are many organizations which are still struggling with conventional structured data. Mastery over this data domain could very well take precedence over that of its unstructured counterpart, and could partially explain why AI technology adoption has yet to become more pervasive. Loughlin spoke about this reality in the financial services vertical, stating that despite the plethora of unstructured data required for analysts and trading purposes, âI see more talk than action there. I think right now the financial organizations have such a monstrous problem on their hands with structured data that unstructured data is deferred. Itâs coming. The use cases are out there, but itâs still early days for unstructured content in the financial services world from what Iâve seen.â
Loughlinâs sentiment seems widely applicable across industries as well. What is unequivocally changing is the inclusion of basic AI capabilities in platforms which can hasten numerous time-sensitive processes. Classic examples are found in data preparation and aggregation prerequisites prior to analytics in which AI and NLP âcan recommend looking at the data and say oh, I see you have a customer data set and a demographic data set,â Paxata Co-Founder and Chief Product Officer explained. âLet me tell you how to bring these two data sets together so you can accelerate to get to the point where someone can get the data that they need.â The full scope of AI, however, which is why it so highly anticipated to alter the data landscape this year and beyond, involves accelerating core functions of business processes and, perhaps even displacing job positions due to expedited automation.
This perception may function as somewhat of an unspoken caveat restraining the full advent of AI upon the enterprise, as vendors continue to push the rhetoric that its technologies are not displacing laborers but allowing them to concentrate on âmore profoundâ problems. Regardless, unveiling this potential of machine intelligence requires a sizable allocation of resources in terms of cost and expertise denoting just how it can make core business functions more efficient. This phase is followed by implementation and assumes, (quite incorrectly, in some instances) that organizations have already mastered the fundamentals of structured data and their requirements. Thus, for better or worse, AI largely remains an intriguing idea, and one which is currently actualized only at a fraction of its full potential.