The Challenges, Successes, Progression & Failures of Processing in AI | Whitepaper

Data makes the world go round, or at the very least, is the underlying heartbeat of all AI and Machine Learning Development. Whilst 2020 was somewhat of a testing and tumultuous year, the development of AI infrastructure and the collection of data for this purpose continued, at a faster pace than previously seen.

Albeit behind the scenes, 2021 could be the year in which we see the rollout of AI in societal settings. None of this, however, would be possible without the large swathes of data collected, which in itself, could potentially be a bigger challenge than creating the models themselves.

The key components of a data science project, and the different kinds of challenges associated with them, play an important role in identifying data limitations. Availability, cost, privacy, ethics and processing data collections all stand in the way of wide-scale development at all industry levels, as well as rollout for consumer use. It is widely believed that we will not be experiencing 100 years of progress in the 21st century, but rather closer to 20,000 years. That is dependent, however, on the large amounts of data needed to be collected. Therefore it is only through consistent experimentation that the future potential of machines can be met.


The stark reality is that we have moved through a ‘generation of big data’ to daily generation, from the many mobile applications, messages sent, and the 3.5 billion daily searches. One of the key challenges to this acceleration is in the data exchange required between processors and memory, as well as data transfer and storage capabilities. Furthermore, data availability is not the same as data integrity, data retention or data reliability. While all these concepts have some similarities, they are also very different from one another. Hence, potential scarcity of usable and good quality data could create a world in which building suitable models that work cross-industry is so close, yet so far.


From ElasticSearch to Fake news and Edge-computing to Data Limitation, the following white paper takes the concept of data and addresses a variety of both accelerating and restricting factors, as well as discussing relevant industry developments and their effect on the current state of data.

“One of the biggest limitations to workload acceleration Nis the limitation in data exchange required between processors’ and memory.” Mark Wright, GSI Technology

For more information, visit RE•WORK and download the white paper.