2023 AI Industry Trends Predictions

2023 AI INDUSTRY PREDICTIONS

Jean-Luc Chatelain CTO Accenture Applied Intelligence*

The ChatGPT hype will be in full force for a while, and will be an emotional roller coaster for many, the ones who will embrace it too soon in the enterprise and have a weak stomach should buy some sickness bags!

Hype is not new; recently we have seen Big-data, Hadoop, AI and there will be more. ChatGPT (and its various copycats) is the fastest growing one so far and the shiny toy effect is in full force. While technically a very good accomplishment, it’s maturity, accuracy, economic model, privacy and many related front critical to the enterprise are not even embryonic, more like zygotic. My recommendation is to stay informed and proceed with caution.

As an example, below is that ChatGPT answered to my question about its top 10 2023 AI predictions:

I can certainly make some educated guesses about what might be some of the top trends in the data and analytics space in 2023, but it's important to keep in mind that predicting the future is always difficult and there could be unforeseen developments that change the landscape. With that said, here are ten predictions for 2023 in the data and analytics space:

a.    Increased adoption of machine learning and artificial intelligence in data analysis and decision making.

b.    Continued growth of the Internet of Things, leading to an explosion of data generated by connected devices.

c.     More companies will adopt a "data culture" and prioritize data literacy among their employees.

d.    Greater use of data governance frameworks to ensure responsible and ethical use of data.

e.    The use of data visualization and storytelling techniques to communicate insights and findings to a wider audience.

f.     More companies will use data lakes as a central repository for storing and processing large amounts of structured and unstructured data.

g.    Continued growth of cloud-based data storage and processing solutions.

h.    Increased use of data engineering practices, such as data pipelines and data wrangling, to manage and analyze data at scale.

i.      More widespread adoption of real-time streaming analytics for applications such as fraud detection and customer experience optimization.

j.      The emergence of new technologies and approaches, such as quantum computing and federated learning, that could have a significant impact on the field.

As one can see all the answers are motherhood & apple pie, temporally ignorant (these statements were already true 10 years ago) and some in fact quite inaccurate in 2023.

The first law of AI will remain true for the foreseeable future: BAD DATA = BAD AI

Nothing is new under the sun, as I mentioned last year and years before “If digital transformation is an iceberg, the tip of it maybe AI, but what will sink the IT boats is the huge amount of raw data that is under the waterline”. Today, the data supply chain which transforms raw data into trusted data has poor economics, too much friction & cannot keep up with speed needed for data preparation at scale. Data quality remains the long pole in the tent and to my chagrin, progress one would have expected from the players in that space is still moving at a snail’s pace. Unfortunately, most enterprises will still need to spend my time and money getting their data right before truly reaping the benefits of AI.

Adoption of Data Mesh Strategy will continue to gradually increase, some confusion will still reign, but its three supporting pillars are anchored: governance approach, data product paradigm and comprehensive metadata management.

Data Mesh does not force a forklift upgrade of all the underlying existing data stores but rather bring a federated & global governance. It allows for decentralized data ownership, shines a light of all enterprise data assets hence truly contribute to elimination of data silos and enabling data monetization within an enterprise.

Now well established as a must have, ModelOps will evolve and eventually morph into “AllOps”

The adoption of ModelOps is now well established as part of the data drive enterprise movement (if I take RFI & RFP as a reliable source). It’s natural evolution is the connection to the other “Ops” such as MLOps,DevOps, DataOps etc. and the emergence of “AllOps” integrated offerings giving a single operational pane of glass of the data driven transformation.

Digital Twins are getting out of their physical shells!

 Digital Twins per se are not new but have been, mostly, associated with the digital copy of a physical object such as a turbine, an airplane wing and other object where simulation has been key to refining, optimizing design and operations. Digital Twins & their acolytes, Knowledge Graphs, are now being increasingly used to simulate and optimize enterprise processes (e.g. procurement, supply chain) thus augmenting the decision support capabilities given to the CxOs. Ultimately the federation & connection of these individual functional twins will allow a full twin of an enterprise to emerge!

Neuromorphic computing has a good chance to be the leading technical approach for AI acceleration at the edge.

There are almost too many choices today in silicon AI acceleration, long gone are the days of “GPU is the only way to go fast in AI”. Many of these new choices are welcome as the one the side fits all paradigm does not work for all use case scenarios. The data center or the cloud are no longer the only environment where computing happens; some of the workload need to run “on the edge” where power consumption & footprint are critical, especially as one must scale tiny edge (e.g., sensor) and/or fat edge (e.g., satellite). Neuromorphic computing approach takes a radically different approach to implementation of AI algorithm, it does not brute force try to imitate the features of human brain and scale, but leverage the principles behind the brain decision making process, craft the silicon to achieve the desired goals and, in most vendors’ case, thrive to minimize the power & footprint budget for both tiny and fat edge deployments.

* The views and opinions expressed in this document are mine and do not necessarily reflect the views or positions of Accenture.