I was on a call with IBM research this week talking about how IBM is deploying artificial intelligence (AI). The company remains one of the most knowledgeable vendors in their class.
Earlier this week, I was advising several governments at a conference that you should start with a vendor with substantial experience when looking at advanced technologies. This approach is so the vendor has learned on someone else’s dollar and not your own.
Let’s talk about what IBM said about their automation vision this week.
Learning Over Building
The typical approach to any technology deployment tends to lead with the product and let the customer figure out how to use it. This approach works when the products are mature, the buyer is knowledgeable, and the related risk is relatively low.
That isn’t the situation we have with AI. The technology is new, the vast majority of vendors in the market inexperienced with it, experienced internal practitioners are increasing but still below critical mass, and the technology has yet to reach its expected general-purpose maturity stage.
At least initially, until you build internal competence, this means that experience is the most important value in a vendor working to help you deploy AI. Their value is less about the hardware they offer and more about what they’ve learned about the limitations of the technology, the most successful approaches to deploy it, and how to avoid expensive problems commonly experienced by buyers.
One common problem is moving to deployment before the solution has been fully understood, resulting in a deployment failure.
More Learning, Less Building
IBM has focused on their effort in the IBM research organization that is storied in the market. IBM research remains one of the largest and best-funded private research organizations globally, and they have been working on and with AI for decades. They didn’t just focus on creating AI technology but were missioned to discover the most successful ways to provide, train, emulate, and deploy the final solution.
One of the interesting comments that came out of the presentation this week was that you need to focus more on learning about the problem than building the solution. This focus on understanding the problem you are attempting to solve with AI is critical to successfully deploying a solution.
With initial AI deployments, buyers of the technology were shocked to learn that the hardware cost was trivial when taken against the cost of the training necessary to get it to work. If the problem isn’t adequately analyzed, then the training will be inadequate and the solution much more likely to fail.
Currently, the predominant problem that AIs are being asked to address is pulling information from unstructured data. This data flows in at ever more significant volumes as cameras and other sensors are brought online in our ever more Internet of Things (IoT)-driven world.
But even here, arguably before the first sensor is placed in the solution, issues, like compatibility, interoperability, assuring the reliability of the data and thinking through what the need is to make sure the sensors are capturing the critical data, all need to be determined to avoid a later problem with the result. And the goal of assuming the timeliness and accuracy of the information provided with the least human involvement should be built into the plan long before the hardware is specified and purchased.
IBM still stands out in terms of experienced AI providers for enterprise-class deployments.
Yes, they have Watson, but their actual value is their decades of experience building and deploying AI systems. They may initially seem expensive, but the cost of a failed AI project is mostly waste and the embarrassment of needing to start over or massively reinvest in the project to correct the problems resulting from poor planning. What IBM brings to the table is the aim that this won’t be your result if you use them. Their benefit is to help you build up your own core AI expertise, so, eventually, you roll out solutions like this with your experience guarding against failure.
In the end, IBM’s value isn’t their hardware or software. It is their experience, experience that forms the basis to help your AI deployment do what was intended and not become a costly embarrassment.