Businesses today run on a multifarious mix of information technology resources, cloud-services and data-centric workflows that all (hopefully) coalesce to provide a connected unified IT layer upon which to do business.
We tend to refer to an IT installation as a ‘stack’ for good reason i.e., a set of tools, functions and services that only really work properly once they all reside on the same foundation stone.
Key among the elements in the modern stack is course data. But relax, this is not some hackneyed or laboured attempt to remind you (dear reader) that data is the new lifeblood of business – don’t worry, we all know that part.
What we want to concern ourselves with now is the application of artificial intelligence (AI) functions upon an organisation’s datasets and – crucially – the need to provide AI model and engine operations (Ops) controls in a more automated way to ensure that AI keeps working 24×7 and deeper into perhaps-unknown virtual time zones.
The Rise of ModelOps
This is where ModelOps comes in and we can make use of analyst house Gartner’s simple definition to grasp a core appreciation of where this technology works: “ModelOps is focused primarily on the governance and lifecycle management of a wide range of operationalised AI and decision models, including machine learning, knowledge graphs, rules, optimisation, linguistic and agent-based models.”
Why are we concerned with ModelOps? Because recent research suggests that 92% of companies are spending more on data science (within which we can reasonably enshroud AI), yet only 12% of them are doing it effectively. The research in question here is New Vantage Partners, 2021 Big Data and AI Executive Survey, authored by Tom Davenport and Randy Bean.
We have been here before. Back in the 1960s and 1970s hobbyists used to build ‘kit’ PCs from scratch at what was still a comparatively experimental time for desktop machines. At that point, we could still justifiably spend time inventing and reinventing a wheel that had yet to be fully formed, forged and forked.
The same dynamic is happening in data science. Enterprise organisations are largely still shouldering the full responsibility of building and maintaining their own data science estate. Admittedly, this practice is rather more advanced when compared to hobbyist kit development of any kind, but it still places an over-arching responsibility on the enterprise itself to handle the Ops element of AI intelligence development.
Cloud-Based AI Model Management
Aiming to fix some (if not all) of the issues associated with this predicament is enterprise data company TIBCO Software. The firm’s latest release of TIBCO ModelOps is designed for organisations to deploy AI models faster, across a broader reach of IT systems and devices and at scale.
This is cloud-based scalable data analytics model management, with additional (and, if you accept the above, much needed) AI model deployment, monitoring and governance functionalities.
“To help organisations realise the value of their AI deployments, we’ve designed a system that puts self-service access to data science firmly in the hands of teams, including business users,” says Mark Palmer, SVP, engineering, TIBCO. “This allows decision-making teams to choose the algorithm they want, work from any cloud service and run it safely, securely and at scale. This is a bold step to enabling business users to take AI out of the lab and out on the road.”
How Does ModelOps Work?
Given that we have set out what is hopefully a rationale and validation point for ModelOps and explained how it now comes to the fore within enterprise software platform vendors like TIBCO, just exactly how does it work?
In this case, the ModelOps function works to address an organisation’s requirements for AI deployment speed; because a huge weight of the functional mechanics are taken care of, a business can confront critical deployment hurdles like ease-of-applying analytics to applications, identification and mitigation of bias and transparency and manageability of an algorithm’s behaviour within business-critical applications.
The solution enables businesses to deploy and manage model pipelines into production environments efficiently, and in robust ways. The TIBCO ModelOps solution is format-agnostic, supporting all common model formats, including API-based models in any cloud service or on-premises.
TIBCO will now be working to finesse the AI use case deployments it oversees across its UK customer base which includes a number of well-known brands including mobile network provider Three UK, Save The Children UK, Dartmouth College, casino operator Genting Group, Skipton Building Society and the AA.
AI Ops Is Easy Now Then, Right?
If we drink the Kool-Aid here and accept the AI ModelOps gospel according to TIBCO, does that mean that AI operations are now a piece of cake?
A 2022 survey of TIBCO customers confirmed that it’s no longer uncommon for a typical business to manage hundreds (or indeed thousands) of analytic models and workflows. The TIBCO ModelOps service claims to be able to allow any authorised business user, data scientist, analyst, or IT user to manage and deploy thousands of models in production with complete governance and management capabilities.
There is, undeniably, something of a gap between some of those stakeholders i.e., the space between authorised business user and data scientist is not inconsiderable and these are still early days for AI automation.
While we can see that TIBCO users are able to deploy AI models in the cloud or on-premises, highlighting model performance through built-in customisable dashboards – which TIBCO itself supplies the engineering for with its Spotfire AI data wrangling and data visualistion service – we are some way off plug-and-play AI engineering how ever many tools are on offer here.
The company insists that with TIBCO ModelOps, customers can move past the worry of unintended negative consequences of failed automation because of complex or poorly managed AI or rules-based models, making it safer to automate based on validated and secure AI models.
None of which is to say that this technology doesn’t automate and alleviate. It clearly does. All we want to say is that citizen data scientists are probably inevitable, but they sometimes need to be treated with the same caution as citizen lion tamers. Just because you’ve got the hat, doesn’t mean you know the job.