Artificial Intelligence and Machine Learning – hype or reality?
Artificial Intelligence (AI) and Machine Learning (ML) are increasingly used by large technology companies to produce new consumer applications and drive improved business outcomes. At the same time, many industrial companies are struggling to get a clear picture on what AI can do for their enterprise and are uncertain of how to transform their general interest in AI/ML into real sustainable business value. So, what does it take for the industry to join the forerunners in the field and get competitive advantages from AI/ML?
AI/ML is not new
When reading a technology newspaper today, one could easily think that AI/ML is a completely new area. However, AI/ML is in fact not new at all; it has been around as a research field at least since the mid-1950s. The interest in AI/ML has come and gone in a cyclic manner over the years. Promising early results have propelled the interest into a hype that has then faded into an AI winter when the outcome failed to meet the hyped expectations. What is important to remember, though, is that no matter where we are in the cycle, the technology around AI/ML is developing continuously and we can do a lot more with it today than we could yesterday. However, this time, there may also be some interesting circumstances that will make the future different.
Today, it is not only AI/ML as a technology that stands at the doorstep of a breakthrough. Instead, it coincides with technology shifts in many other areas, such as biotech, transportation, energy and industrial digitalization, to name but a few. All these areas interact with, and in some cases need, AI/ML to be able to make their own breakthrough. In that sense, AI/ML is becoming an integrated part in several other parallel ongoing technology shifts. This time around, we also have easy access to pretty much everything we need in terms of increased computing power, powerful development platforms and the availability of large datasets.
What are current bottlenecks?
What is missing to make AI/ML applications an integrated part of every industrial enterprise? Let’s first conclude that just because all the tools are available, it does not mean that AI/ML would be simple; on the contrary, using AI/ML to add sustainable business value in an industrial enterprise is complicated. It requires not only an understanding of how this fast-evolving technology can help and what the best places to deploy it are, but also a genuine understanding of how such a system needs to be constructed and behave to achieve long-term industrial-grade quality and reliability.
Finding enough people with these competencies can be a bottleneck for the western world’s highly effective industries and slimmed-down organizations with scarce resources. In quite a few companies, there might simply be no free human capacity available to evaluate their own business in terms of AI/ML suitability, and in many cases even less resources available with the competence and mandate needed to change the company strategy to include AI/ML as an integrated component in the enterprise.
Previously, industry has often been able to solve similar problems by investing in proven use cases, solutions and systems with good references that have proved their value at other companies. Within AI/ML, though, this is not as easy. Implementing an AI/ML system is fundamentally different from a traditional industrial software project in that the system performance will always be heavily dependent on external data. In other words, a successful reference is primarily an indication that the external data used in that specific installation contained important information. An AI/ML reference says very little about the likelihood of succeeding elsewhere since the information content of the external data will be different. Hence, identification and validation of use cases with business value is one of the bottlenecks.
AI/ML provides value
Having said everything above, however, we should be careful not to overcomplicate AI/ML either. It is not easy, but it is perhaps not as hard as some might say either. Many companies have come quite far and found ways to overcome these hurdles. In a recent report from Deloitte Insights (October 22, 2018), 1100 IT and line-of-business executives from US-based companies were asked about the results from their investments in the AI/ML field. Their answers indicate that something has happened:
82% have gained financial return from their AI investments
17% median return on investment
37% have gained competitive advantage
42% claim that AI/ML will be of significant strategic importance for them
Although AI/ML is not the solution to all problems and although there are impediments to overcome, more than 8 out of 10 companies in this investigation have gained financial return on their investments. And what is even more promising is that 37% say that they have gained a competitive advantage from their AI/ML investments.
How to improve chances of gaining value?
So how do you maximize the likelihood of becoming one of the enterprises that will get financial return from your AI/ML investments? Even if an AI/ML system is fundamentally different from a traditional industrial software project, in many aspects, some things are still very familiar. For example, a solid evaluation of the business case should still be made (ideally, the business case is one of many that have emerged from a company-wide AI/ML strategy). For some reason, this seems to sometimes be forgotten when it comes to introducing new technology. If you aim at nothing, you most likely will get just that from the investment.
Another important aspect to include in an AI/ML strategy is how to maintain AI/ML solutions over time. As mentioned earlier, AI/ML performance is dependent on external data. If you have a production process that over time will change how data is being generated, the performance of the AI/ML system will also change.
Expectation management is equally important. Think in terms of what business value the strategy needs to create in the overall perspective for you to get that competitive advantage for your enterprise.
As with all new technology, there comes a point when the cost of not using it, and falling behind, is greater than the cost of failing when trying to use it. Gaining competitive advantages in industry is never easy, but AI/ML is offering industrial forerunners the opportunity to do just that if the technology itself is combined with a sober level of expectation, a solid AI/ML strategy, a good method for evaluating AI/ML business cases and a good partner that can cover areas where internal resources are not available.
Johan Bäckman, the author of this article, is leading three projects within Sweden’s strategic innovation program for ProcessIT and Automation with focus on AI/ML for industrial applications. During the last 20 years, he has worked as an engineer, project manager, product manager and business unit manager within the field of industrial IT and production-critical software solutions.