FEBRUARY 17, 2019 — 9 MINUTES READ
These days, artificial intelligence (AI) dominates the headlines of leading business publications as well as the discussions of C-Suite conference calls. As a result, enterprises and startups are both racing to bring this disruptive technology into their fold.
But for many companies, developing AI and deploying it are proving to be more challenging than originally anticipated.
Updating an Age-Old Art
Since the introduction of synthetic ingredients in the 1880s, making perfumes and colognes hasn’t changed too much. Achim Daub is an executive of Symrise AG, one of the biggest names in fragrances. He took notice of the lack of innovation in the industry and decided it was time to inject some into Symrise’s process with AI.
So he hired IBM to design an AI system capable of digesting immense amounts of information like existing fragrance formulas, regulations, consumer data, and more. It’s called Philyra (the name of the Greek goddess of fragrance). So far, progress has been promising; Symrise is set to release two new Philyra-concocted fragrances in Brazil next June.
Daub does emphasize that it took almost two years to get to this point, though. Initially, Philyra’s suggestions left much to be desired. But after some intensive training by Symrise’s perfumers and costly IT upgrades to connect the company’s disparate data to the system, things became better.
Some work still needs to be done. Currently, only a couple of the company’s 70 perfumers are using Philyra. Daub expects that all of them will be using it in due time but also admits that implementing AI was more arduous than he originally thought. “It’s kind of a steep learning curve,” he explains. “We are nowhere near having AI firmly and completely established in our enterprise system.”
A Marathon, Not a Race
The fragrance industry is far from the only sector that has encountered roadblocks in adopting AI. Sure, we hear stories of deep learning neural networks trouncing Go grandmasters at their own game. But utilizing AI and machine learning to incrementally improve a business is easier said than done. And the results are often modest at most.
At the moment, the majority of companies aren’t seeing substantially more output from AI than from what they’d usually get from regular employees. Today, large productivity gains are reserved for the richest companies that can actually afford to invest heavily in the AI talent and technology needed to make things happen. To see this in action, look no further than the recent work of giant San Francisco developers like Google, Apple, or Facebook.
Now, this doesn’t necessarily mean that AI’s overhyped. In the future, it’s still quite likely to disrupt many aspects of society as we know it. It’s just that the technology as a whole is still limited in terms of what can be done. Pattern recognition algorithms only account for so much when it comes to how business is conducted today. And even when AI is implemented, it currently requires an absurd level of patience and meticulousness to get it right.
Data scientist and SkipFlag CEO Peter Skomoroch aptly described AI implementation in a recent tweet: “As a rule of thumb, you can expect the transition of your enterprise company to machine learning will be about 100x harder than your transition to mobile.” While it came off as a joke, many companies chimed in to say the tweet rang a little too true.
Skomoroch, whose company transforms a business’s internal communications into employee knowledge bases, wasn’t surprised. “I think there’s a lot of pain out there—inflated expectations. AI and machine learning are seen as magic fairy dust.”
Read more: Unlocking the Information Advantage
Connecting the Dots in Disparate Data
As far as implementation obstacles go, none have proven to be as troubling as unifying a company’s records and information. This was the main problem that Richard Zane, chief innovation officer at UC Health, encountered. To better assist patients reaching out through phone or Internet, the network of hospitals spanning Colorado, Wyoming, and Nebraska rolled out Livi, a chatbot that utilizes natural language processing.
Because of IT hiccups in connecting Livi to a plethora of information like health records, insurance data, and other hospital systems, it took a year and a half to deploy it. Today, Livi helps patients in a plethora of ways, and Zane is pleased with the results. But this is just the beginning of AI entering the health industry. How hard will it be to implement more sophisticated initiatives?
The healthcare sector isn’t the only one that keeps information siloed off. Many industries do. It’s why the most common use cases of AI today often revolve around one aspect of a business that has tons of data to mine from, like financial fraud detection or cybersecurity.
And even after the data is united, there’s still the issue of making it understandable and meaningful for the AI to use. Genpact is an IT services company that helps organizations deploy AI. According to its chief digital officer, Sanjay Srivastava, “10% of the work is AI. Ninety percent of the work is actually data extraction, cleansing, normalizing, wrangling.”
How to Accelerate AI Adoption
The two steps above make AI look easy when you examine Amazon, Netflix, Google, and other tech titans. But that’s because their businesses revolve around making sense of digital data. They literally have entire teams composed of PhDs in data science, computer science, and other fields relevant to AI.
For smaller businesses and organizations not as focused on data, developing and deploying AI remains a daunting challenge. This is partially due to the shortage of AI experts currently available. It’s not easy to find engineers and developers who are adept at seeing a business’s process and building AI to improve it.
With the current hype surrounding AI, this will certainly change in the coming years. More students, techies, and people from other industries see the potential in this technology and have started pivoting toward a career in it. But for now, businesses can do a few things to make themselves AI-ready when the time comes.
First, it’s important to open up information silos. Gone are the days of closed doors. Modern organizations are connecting the information they have at hand to reap tremendous benefits. Any company that fails to do so soon will feel the repercussions in the near future.
Second, businesses should establish a clear-cut, organized way of gathering and organizing data. This will save enormous amounts of time down the line. Rather than having to take a step back and tackle this issue when it grows into a bigger problem, companies that take a proactive approach will be equipped with some of the most vital tools they need to make their AI initiatives successful.
As with many revolutionary technologies, the hype surrounding AI currently precedes the capabilities it offers. But it’s only a matter of time before its benefits are fully realized.