Every few years a new-hype technology becomes the shining star. It sucks all the oxygen out of the room and becomes the headline darling for a while. The list of technologies that move the needle forward on managing the business better or help to grow a business are impressive. They provide great gains, but sometimes have a shelf life. This might be due to an eventual lack of impressive results. But more often, they are displaced by a newer, shinier technology that simply does their job better, faster and stronger. Business management software is but one example market that is no stranger to, and beneficiary of, new tech, and the gutter is littered with market losers that failed to adapt.
Harvard Business Review recently published an article that points to the peril businesses will face if they don’t adopt AI sooner than later. This initially sounds like fear-mongering journalistic headlines, but they have a few very fine and accurate points. They aren’t claiming AI is the end all be all, but are a natural progression. All the latest shiny tech tools have matured over time to build the AI tools that exist today. They are natural accretions instead of massive replacements. On the flipside, these risks could be disruptive, dilutive, or outright company ending.
Are you ready to start with AI now? Or are you planning to wait?
My recommendation? Start now.
Now, let me tell you why.
First, AI systems are accretive. One doesn’t wake up one day with a PhD, completing Ironman Kona, or implementing fully successful AI systems. These things take time. There is a life cycle similar to the crawl, walk, and run analogy of how companies can move from introducing AI, implementing successful models, and then building a fully successful scaled AI implementation. Scaled AI implementations aren’t a few Data Scientists with a handful of GPUs that sporadically create successful models that implement one off results. There is a holistic environment where Data Scientists and Data Engineers can label and wrangle data, visualize results, create models, and review efficacy of those models at scale. This might be multiple models trained off of large data sets per day for each Data Scientist and constant wrangling by Data Engineers.
The latest #RapidsAI announcement from Nvidia found here targeting GPUs for Data Science and Engineering is certain to provide yet another disruption in the changing environment. It is another game-changing evolution that is modernizing Data Science and Data Engineering tools. The efficiency of accelerating and scaling Data Engineering tasks with GPUs will take time to roll through environments, however. This is part of the accretive nature of building the large environments. Like Rapids, there will always be a next new innovation.
Second, Governance doesn’t happen overnight. Do not confuse governance of AI and advanced analytics with regulatory related oversite. They are complimentary, but not exclusive. Good governance of advanced analytics includes reviews for bias (or the chance a model has some inherent incorrect assumptions and affiliations), for outdated models (where the demographic or historical data set on which is was built has materially changed), and for alignment (where the model and results are in line with the companies mission and don’t broadcast an incorrect statement about the company).
In my work with the automotive industry around ADAS (Advanced Driver Assistance Systems), most of the major model systems for the OEMs and Tier 1 ADAS manufacturers are built for a daily recalibration. To be more precise, they have created an SLA and environment where they can re-train the entire simulation of a model in a single day…. This is an example of an industry where regulatory governance (of safety) is currently limited (but coming). By anticipating the need for fast turn-around, to quickly address any sudden safety issues, their AI governance is ready for any future changes any government may mandate. The ability to get products to market sooner is a bonus. This is what a mature AI governance looks like.
Third and probably the most under thought of is the investment path into AI. Regularly I am approached by technology managers struggling to insource their AI endeavor. All to often these companies rushed onto the AI bandwagon without fully mapping the path. These companies all too often relied on a 3rd party’s fully embedded AI technology to bootstrap their effort. This accelerated their first steps into the AI environment and bought them speed in the first bullet above. Unfortunately, this often comes at the lack of any intellectual property (IP). At the end of the day, all the IP is owned by the 3rd party – leaving the company’s product differentiation controlled someone else’s technology. This is often strategically challenging and often leads to issues around cost or around how the company wants to inevitably mature the technology into its own offering. They cannot as they do not own the IP. This is the difference between a platform as a service (PAAS) model where you consume the infrastructure and own the technology and a software as a service (SAAS) model where you only own specific models built on someone else’s technology. It is important to understand the difference and the impact to your business.
