AI as a service to solve your business problems? Guess again

This post was originally published by Annie Saunders at TechCrunch

SaaS, PaaS – and now AIaaS: Entrepreneurial, forward-thinking companies will attempt to provide customers of all types with artificial intelligence-powered plug-and-play solutions for myriad business problems.

Industries of all types are embracing off-the-shelf AI solutions. According to industry experts, global AI software revenue — most of it online artificial intelligence as a service software (AIaaS) — is set to grow by an astounding annual rate of 34.9%, with the market reaching over $100 billion by 2025. It sounds like a great idea, but there is a caveat — “one-size-fits-all” syndrome.

Companies seeking to use AI as a differentiating technology in order to gain business advantages — and not merely doing it because that’s what everyone else is doing — require planning and strategy, and that almost always means a customized solution.

In the words of Sepp Hochreiter (inventor of LSTM, one of the world’s most famous and successful AI algorithms), “the ideal combination for the best time to market and lowest risk for your AI projects is to slowly build a team and use external proven experts as well. No one can hire the best talent quickly, and even worse, you cannot even judge the quality during hiring but will only find out years later.”

That’s a far cry from what most online off-the-shelf AI services offer today. The artificial intelligence technology offered by AIaaS comes in two flavors — and the predominant one is a very basic AI system that claims to provide a “one-size-fits-all” solution for all businesses. Modules offered by AI service providers are meant to be applied, as-is, to anything from organizing a stockroom to optimizing a customer database to preventing anomalies in production of a multitude of products.

There are several companies that claim to provide AIaaS for automated industrial production. Most of the successful data presented by these providers is based on individual case studies, with problems involving limited data sets and limited, generic objectives. But generic AI solutions are going to produce generic results.

For example, the process to train algorithms to detect wear and tear would be different for factories that produce different products; after all, a shoe is not a smartphone is not a bicycle. Thus, for “real” AI work — where intelligent modules actually managed and changed production in response to environmental and other factors — the companies developed customized solutions for their clients.

Many customers who were “burned” by bad experience with AIaaS will be more hesitant to try it again, feeling it is a waste of time. And use cases that did require heavier AI processing did not yield the results expected — or promised. Some have even accused the cloud companies of deliberately misleading customers — giving them the impression that off-the-shelf AI is a viable solution, when they know very well that it isn’t. And if a technology doesn’t work enough times, chances are that those who could potentially benefit from real AI solutions will give up before they even start.

The objective is to standardize a solution that performs well almost immediately and does not require extensive know-how. AIaaS’ success so far has been in enabling researchers to run complex experiments without requiring the services of an entire IT team to figure out how to manage the necessary infrastructure.

In the future, AIaaS will hopefully enable individuals who are not AI experts to utilize the system to get the desired results. That said, online automated AI services even at their current levels can greatly benefit industrial production — if it is done right.

AI properly done could provide great benefits for industry. Instead of giving up on AI, companies should do a deep dive on the AI services they are thinking of utilizing. Does the solution provide for customization? What kind of support does the service provide? How is the algorithm trained to handle data specific to your use case? These are the questions that companies need to ask when shopping around for AI services. Providers that can furnish substantial answers — and back up their claims with real data on success rates — are the ones companies should work with.

Like all new developments that enhance business activity, AI applications require a high level of expertise. The engineers who work for the big cloud companies indeed have that expertise — which means that they could be providing much more value for customers by helping them develop customized solutions. Whether that can be done “as a service” needs to be examined — but the system in place right now is not the answer.

Spread the word

This post was originally published by Annie Saunders at TechCrunch

Related posts