IBM's Artificial Intelligence (AI) didn't fulfill expectations. We examine why.
Achieving success in the next stage of The Data Gold Rush
Transforming the enterprise with Artificial Intelligence and data
We are in the midst of the third phase of achieving enterprise transformation through Artificial Intelligence (AI).
Early ML Adoption
In the early stages of AI adoption, enterprises - heavily influenced by the lean startup approach by Steve Blank and Eric Reis - started building AI tools and products without any systematic approach.
In this stage, the key differentiation between clear winners and losers was access to the volumes of useful data needed to power any reasonable machine learning (ML) technology.
As organizations created enterprise data and data science teams, they needed data to make these teams effective and build advantage. Without sufficient data, no amount of investment in data science could yielded expected results.
Data Acquisition Race
As a result, many firms went into a data acquisition arms race, “hoovering” all they could to power models. This second stage was achieved by either interacting directly with the user and recording every single interaction or by simply acquiring data from data aggregators. While significant investments were made into acquiring data, not much thought was given to quality or affordability. Internet and data platforms benefited greatly from the second phase.
The only way to acquire this volume of data was through disparate sources where data really didn’t ‘talk well with one another’. There were even multiple unicorns created that offered solutions to integrate these disparate data sources. However, data science teams still struggle to realize the full potential of ML and AI as a result of poorly formatted and disorganized data. Meanwhile, the pressure is even higher on data teams to deliver ROI on all thats been invested in the last few years.
Value Extraction Challenge
This brings us to today - the third stage: The Value Extraction Challenge. The shift from text based communication to more visual, image and video-based communication has increased the existing challenge of data science teams to work with poorly compatible data sources to drive insights and knowledge extraction. These inefficiencies have created additional mistrust in the promise of AI. Increasing data volumes is adding more noise as enterprises become ‘data poor’ with more data. Here’s why:
Value extraction challenges from each separate data source
The quality of each data source is critical. Quality drives the value that gets extracted. The feature sets that can be summarized from a dataset are getting harder with the increase of raw information volume. Further, challenges such as lack of knowledge and antiquated models running on noisy raw data continue to obfuscate the realization of insights from each set of messy data.
With the increase in visual data, data sets have increasingly become ‘less dense’
As communication has evolved, we have seen an increasing amount of video data in addition to text and image data. Applying individual ML models on each data source creates a set of features that cannot be easily integrated and leveraged. As a result, the value extracted from this increase in data and new visual data sources is actually decreasing!
Inability to integrate and experiment fast
The core part of any efficient data organization is the ability to experiment with data really fast. With brute force ML approaches, the amount of time involved in processing information increases exponentially and adds to the frustration in the ability of data science teams to process the information at the pace the business needs to glean insights and gain competitive advantage
Solution Based AI is the next, and most critical phase
A fourth phase is now on the horizon, with technology such as Netra at the forefront. Solution-based AI focuses on generating the critical data required to solve unique business challenges. For example, Netra offers a consistent, total comprehension taxonomy across text, image and video data in a way that data science teams can rapidly and easily integrate all of the information using a single layer.
Netra’s API can ingest all three forms of raw data and create a common, structured understanding of content that can easily be experimented with by the data teams. The data has a common taxonomy and labeling schema that can be applied to AI and ML models to drive key functions such as optimization and analytics.
Netra’s ability to classify image and (especially) video cost-efficiently at scale provides ML and data science teams with a common framework of features and comprehension. The Netra common data schemas empower these teams to deliver on AI’s promise by organizing the Data Gold Rush into the knowledge and wisdom that each unique business seeks.