Who’s afraid of practical AI? What businesses need to know
Even as nearly every major industry embraces AI, many companies remain wary. Their concern? Well, as Tom Cruise famously put it in Jerry Maguire, “Show me the money!”
For corporate CIOs and CEOs, the decision to undertake an AI project is usually driven by ROI. Real economic consequences are at stake, pressuring them to realize the potential of AI technology, represented by a neural network widely called deep learning (DL), while also confronting its limits.
Who’s doing what, so far
Practical AI applications that solve problems can be found in many industries. Financial services are at the forefront of incorporating neural net solutions largely because they possess massive amounts of data in the form of financial transactions. For instance, JP Morgan implemented multiple AI programs, including LOXM, to perform equities trades in real time, and COIN, a system for reviewing commercial loan contracts — an effort that previously took loan officers 360,000 hours of work each year and which can now be done in a matter of seconds.
Healthcare also has witnessed explosive growth in AI implementation. Robots are gradually working their way into the operating room, and a growing number of companies use AI to perform early diagnosis, expedite drug discovery, and increase patient engagement.
Retail — including good old brick and mortar stores — has undergone an AI-driven makeover, too. The industry applies AI in customer behavior analysis to improve merchandising and advertising, boost customer satisfaction, and more efficiently manage supply chains.
While AI holds much promise for these sectors, there remain several common obstacles that all businesses should consider.
AI is consuming the world
Marc Andreessen of Andreessen Horowitz stated in an interview with Vox that the theory of how AI would take shape used to be that large tech companies like Google and Facebook, with unmatched amounts of data and capital, would be best positioned to offer AI-based products to large corporations. No longer. Startups are taking this bet and companies like Element.AI, provider of AI-as-a-service solution, or Clarifai, provider of image recognition systems, and many others are already working with big enterprises across an array of industries. The financial, banking and retail industries, for instance, faced with build or buy decisions, often choose third-party AI services for tasks like security, customer support, and customer behavior analytics.
The reasons to buy instead of build are many. One of the biggest challenges for enterprises is taking their large amounts of raw data and turning it into knowledge. It’s a process called “data cleansing” and it can be time consuming, i.e. expensive. It involves tasks like ensuring consistency of data — dates in six-digit formats or eight-digit formats, and labels for each bit of data, too.
Another limitation is lack of skilled professionals. C-level executives in large corporations don’t comprehend the difficulties of building deep learning models like feature selection and other typical technical challenges, while technical teams are having a hard time dealing with those issues. Now AI is a club of 10,000 people. The lack of talent and expertise comes to the forefront.
Among other challenges, some large enterprises remain unconvinced by the value of replacing reliable rule-based practices with AI; it’s an old-fashioned mindset of innovation resistance. Idan Tendler, CEO of Fortscale, a startup that provides data analytics to identify insider threats, points out that “at the end of the day, companies just want effective solutions for their problems and don’t care where the solutions come from.” The key is to understand that “AI technology is in a dynamic stage of development and can’t always perform ideally…big companies should take this into account, but understand that once they adjust their internal business process with the technology, they can enjoy much better performance.”
Obstacles to solve
Regardless of whether companies take the leap of faith into AI, the technology itself is hardly off-the-shelf software. It comes with many issues yet to be addressed. One vexing problem: the “interpretability” of deep learning models, meaning that it is sometimes difficult to determine exactly how a neural network reaches a decision. This poses a number of risks, including hidden biases, lack of verifiability, and margin for error.
Ultimately, the challenges for enterprises lie with harnessing with a large amount of data, and turning it into knowledge, as well as a lack of talent, and in some cases, an underlying resistance to innovation. But they can all be solved.
Already, new high-level infrastructures like Keras and ONNX allow less-skilled engineers to work with deep learning models. Broader and deeper data science and ML engineering programs at universities and in online offerings will help with the talent shortage. Active research into new approaches to deal with unlabeled data, and positive ROI results from DL projects offer persuasive evidence to large corporations in nearly every industry. There’s no longer any reason for any business to be afraid of practical AI.