Finding an easier way to AI adoption
We are excited to bring Transform 2022 back in-person July 19 and virtually July 20 — 28. Join AI and data leaders for insightful talks and exciting networking opportunities. Register today!
Artificial intelligence (AI) is the most revolutionary, consequential technology to have come along in decades. And while that’s exciting, it’s also ominous. Get it wrong, and you’re sunk.
At the same time, AI poses such an implantation challenge that getting it wrong is more likely than not. It will touch every digital aspect of the enterprise eventually, which means it is rife with pitfalls, namely, in the integration, training and execution phases of the rollout and, for many it will lead to a wholesale reworking of processes and even the business model itself.
AI deployments: Danger Below
On the surface, it would appear that AI deployments are moving along swimmingly. IBM reports that more than a third of companies say they are using AI in their businesses in some way, and another 42 percent are in the exploratory stage. Nevertheless, things like costs, lack of expertise and the inability to develop workable models are hampering these efforts, and far too few organizations are concentrating on fundamental aspects of AI like building trust, removing bias and tracking performance.
It’s no wonder, then, that many organizations – even large ones with substantial resources – are caught in a kind of AI paralysis. Even those that have dipped their toes in these waters are findingfew concrete examples of success.
For this reason, AI advocates are starting to shift the focus away from all the magical things the technology can do to more practical matters like how to deploy it in an efficient, effective manner. One example is a new book by six leaders in the field called Demystifying AI for the Enterprise. The book lays out a framework to overcome key implementation issues, such as understanding AI’s limits and targeting it at key problems that it can solve. As well, organizations should understand that, unlike previous technologies, AI doesn’t come fully formed out of the box. It must be trained to a level of maturity before it can provide effective support to key processes. And perhaps most importantly, AI works best when it allows people to become better at what they do.
The way forward with AI
A closer look at organizations that have successfully deployed AI shows a number of common themes, most of which involve looking past the technology itself to the data environment as a whole. For example, Jonathon Wright, chief technology evangelist at Keysight Technologies, notes that developing explainable use cases ahead of time greatly improves deployment and implementation, as does an overarching strategy of using AI to augment existing processes and filling gaps in the human workforce. As well, avoid an all-at-once, forklift upgrade to AI and instead plan for a smooth, nondisruptive transition from AI-ready to AI-capable and finally AI-enabled.
Every journey begins with a first step, however, so getting AI right on the first try, even in a limited fashion, can go a long way toward fostering future success. Naturally, this means putting AI to work on easy, proven tasks, which the editors at eWeek have identified as chatbots, image classification and price prediction. Intelligent chatbots are already revolutionizing customer support across many organizations, and they are fairly easy to implement using natural language processing (NLP) and other proven means to mimic conversational speech.
Meanwhile, AI is highly adept at scanning images to a greater degree than previous technologies, and then identifying anomalies and automating classification. And since most prices are subject to a wide range of influences that can only be effectively tracked and measured in an intelligent fashion, AI has become the solution of choice in this area and is already producing significant contributions to revenues and profits.
It will be the rare organization that experiences no setbacks or hiccups in the process of putting AI into production environments, but that doesn’t mean steps should not be taken to keep failures to a minimum. And even failure can be beneficial if it leads to greater understanding of how the technology should work.
At this point, it is hard to see AI not producing the revolutionary changes to systems and processes in the enterprise. The only question is whether organizations will take the easy way to that point, or the hard way.