Last year everyone began talking about AI but when will the talk turn into practice in the real world? There are a lot of prophets out there talking about how AI is disrupting every industry and beginning to replace employees and take over our jobs, while on the flip side, ordinary people are still struggling to make Alexa turn on the lights at home. Clearly there is a disconnect between expectation and reality.
So, the question is: what’s today’s AI reality and how will artificial intelligence or rather the sub-segment, Machine Learning, impact business over the next few years?
To set the record straight, when discussing AI replacing employees’ jobs, this typically refers to the automation of specific job tasks. Recent examples of this include:
Dominant medical companies such as GSK, St. Luke’s University Health Network and Kaiser Permanente are using orplanning to implement AI to transform Healthcare by demonstrating how Machine Learning can perform repetitive measurement tasks, clinically validating quantitation of biomarkers, and utonomously triaging emergent findings – freeing up the schedule for Radiologists.
Facebook and Google both switched from recommending manual audience targeting to leveraging Machine Learning (AI) to identify the target groups and buyers. Simply put, the machines are now better at identifying and targeting customers than humans are. This has not only affected their internal productivity, but also impacted marketing in general, because advertising becomes more efficient, as results are measured more effortlessly. Ad agencies have already seen their shareprices go down over the past year due to the dramatic change in the industry… and this is only the beginning.
Smart email categorization works to prioritize the inbox, categorizing e-mails by primary, social and promotions. This has resulted in cutting down the amount of time spent reading and responding to e-mails by 6%. Additionally, as much as 10% of mobile responses on Google’s Inbox app is made with their AI-driven Smart Reply feature.
Finance & Insurance
The decision-making process for determining how much insurance a person is qualified for and at what rate currently performed by insurance brokers and underwriters can now become more accurate through the use of machines.
Retail & Consumer Packaged Goods & Services
- Using predictive analytics for demand and supply forecasts, product selection, placements and pricing.
- Digital assistants such as Alexa, Google Assistant, Siri and Cortana that are starting to play a key role in the smart home.
What’s Suitable for Machine Learning (SML)?
Machine Learning requires well-defined problems where the input data can reliably map the output predictions. For example, in medical diagnostics medical records go in and diagnoses come out. Another way of looking at this is the classification of things, such as the breed of dogs from images. On the other hand, there are some areas where Machine Learning is not a good fit, including anything where past data is a poor indication of future events, such as the sales forecast for a brand-new product launch.
Leveraging Machine Learning
Another way of understanding how to leverage Machine Learning is by understanding the difference in innovation vs automation:
Innovation implies you are identifying completely new business opportunities by using Machine Learning. Whereas on the other hand, automation primarily means taking an existing task/service and choosing to automate it.
How to approach a Machine Learning project?
It’s possible to look for opportunities where machines could automate existing tasks and processes. One way to go about this is to look at the main categories of Machine Learning in practice for today as referenced in our last blog:
- Image/Voice/Video Recognition/Classification
- Conversational Interfaces (NLP & Chatbots)
- Other R&D type project requests that fall outside of the categories above
Once you’ve identified the list of potential tasks and/or processes to reassess, the next step is to get a deep understanding into the tasks being replaced prior to searching for a solution. Too often we find that technical solutions are being applied to tasks where there is a poor understanding of the capabilities, resulting in long and expensive projects that fail entirely. Before applying a technical solution, it’s imperative that one tests the concept among small amounts of data to figure out if it’s even a feasible solution. It’s important to remember that many common tasks have already been solved by others, so ensure you leverage these resources and out-of-the-box solutions for Machine Learning problems.
Stay tuned for the remaining parts of this series to find out how these six trends are expected to shape success in 2018.
- Conversational interfaces trump chatbots
- Low code and no code
- Data security and privacy
- Apps replacing office tools
Chief Innovation Officer