Mechanistic interpretability is concerned with analyzing individual components in a (convolutional) neural network (CNN) and how they form larger circuits representing decision mechanisms. These investigations are challenging since CNNs frequently learn polysemantic channels that encode distinct concepts, making them hard to interpret. To address this, we propose an algorithm to disentangle a specific kind of polysemantic channel into multiple channels, each responding to a single concept. Our approach restructures weights in a CNN, utilizing that different concepts within the same channel exhibit distinct activation patterns in the previous layer. By disentangling these polysemantic features, we enhance the interpretability of CNNs, ultimately improving explanatory techniques such as feature visualizations.

All businesses are aware of the importance of utilizing AI models to speed up the digital transformation process. But, the process of understanding and zooming onto the most suitable model is complicated as well as crucial. As enterprises are discovering the benefits of AI, they also struggle to map the journey of AI's successful adoption. Many CIOs expect AI to quickly transform their business without identifying the processes that actually need AI for improved performance. In today's ideal world, one can pick any process, infusing it with AI, and then discovering the pros and cons during the journey.

Imagine meeting a personal stylist online who will make specific recommendations for clothing you need to purchase -- including matching items and accessories. Take that a step further, and your stylist will be able to parse recommendations based on your size and budget, as well as available inventory. Finally, you can easily get input from your trusted friends online, make a final decision, and then click once to purchase. With recent improvements to messaging, chatbots, and artificial intelligence technology, that scenario is entirely possible, and the stylist doesn't even need to be a real person. Artificial intelligence (AI) platforms can do much more than comb through product listings and return search results.

Voice technology is gaining mainstream acceptance thanks to the new voice based devices from the leading companies like Amazon, Google. Simultaneously Artificial Intelligence (AI) is becoming more efficient in identifying user intents. Many industries are experimenting with the devices and AI powered abstraction layers to identify areas that they exploit to deliver superior customer experience. According to Google, 20% of the searches are already by voice and they expect this to increase to 50% by 2020. Quantum leap in voice accuracy is another trend that is seen by the technology analysts over the last couple of years.

Just a few decades ago, interactions were slow. Customers sent their complaints through the mail, waiting months for a response that may never arrive. As technology advanced, businesses offered new ways to innovate and improve customer engagement -- at ever-increasing speeds. Snail mail became phone banks with advanced logical routing to subject matter experts. Then the conversation moved online, with inventions such as live chat and interactive voice response creating new channels for engaging with the customer.

While machine learning and behavioral analytics have been around for years, many industry experts are predicting 2018 will be its momentous point. According to Gartner's latest CIO survey, respondents indicated machine learning and artificial intelligence are in their top-five priorities this year, with 21 percent of CIOs from around the world testing or considering these initiatives in the short term. One quarter of the CIO respondents have medium- or long-term plans for the technology. There are many reasons behavioral analytics should be an integral part of the short- and long-term technology roadmaps, especially for financial institutions and payments processors.

Retailers are utilizing AI systems to develop better personalization, grow their audiences and analyze unstructured data. For example, the outdoor and equipment apparel company, The North Face, is using an interactive online shopping experience powered by IBM Watson cognitive computing technology. The system was developed using IBM Fluid's Expert Personal Shopper (XPS) software to create a more engaging, personalized and relevant shopping experience. XPS aims to help consumers discover and refine product selections based on their responses to a series of questions. AI plays an integral role in enhancing marketing systems by feeding "very directly into many marketing functions and processes, including ROI [return on investment] and accountability, ad personalization, voice assistants and programmatic" approaches, according to the NewBase report.

This past summer proved to be the official tipping point for voice-first shopping for many consumers. With Amazon's Echo Dot ranking as the "best-selling product from any manufacturer in any category across Amazon globally" during Prime Day 2017 and Google Home pairing up with Wal-Mart and The Home Depot, the era of AI-assisted selling officially had its breakthrough during the first half of 2017. Additionally, according to a recent Gartner study, sales of voice-activated speakers with artificial intelligence capabilities will reach $3.52 billion by 2021, signaling that adoption of voice-enabled speakers will only continue over the next few years. Though e-commerce continues to gain ground on in-store purchasing, we are collectively a group of consumers who often use our voice throughout our purchasing journeys. Whether it is asking for a different size or color, checking if our product is in stock or simply expressing how we want to pay, we are used to these interactions.

It's no secret artificial intelligence is the hottest technology in insurance. Insurtechs and incumbent carriers alike are excited for the technology's potential in underwriting, customer service and claims. However, the relatively new technology does have some shortcomings, experts say. A lack of historical data, insurers' legacy systems and AI's inability to decipher natural language beyond basic speech has stymied the technology's growth. And for all of AI's industry-wide adoption, carriers appear to only trust machine learning to carry out the same tasks as web portals or mobile apps.