There is a huge whitespace waiting to be filled by the tech companies that recognize the power and potential of messaging.
Roughly 63% of people prefer to share information on “dark social,” or closed, private messaging environments like Facebook Messenger and WhatsApp. However, the experience on these platforms remains painfully circuitous. In order to share a single piece of content within a conversation, users typically have to leave their active chat, open a new window to locate and copy the file, then re-enter the original chat to paste and share.
So there is a big opportunity in providing more intelligent ways to share content on messaging – whether that content is a funny animation, a dinner reservation, or the directions for getting somewhere. In fact, once AI is successfully applied to this area, we could see a whole swath of brand new capabilities emerge in messaging, from intelligent recommendations for nearby restaurants or retailers, AR enabled communications, elevated video and audio messaging, live updates for upcoming trips or events, or other convenience-focused features.
Simultaneously, the application of AI in messaging specifically has the potential to propel the entire AI market forward, as it can serve as a breeding ground for experimentation and innovation to solve some of the current challenges within AI. Once AI is fully functional within messaging environments — that is, once it can fully understand what is being expressed in a conversation, or what is desired – the technology could be applied to many areas and could open the door to new industries that do not yet exist.
However, before we discover this exciting new world on the horizon, we must first face a unique set of challenges in applying AI to the billions of online conversations happening worldwide.
The tech challenge
It seems simple: Build a sound artificial intelligence model, train it, and apply it to the everyday conversations happening on the messaging apps consumers adopted years ago. After all, we’ve figured out how to leverage AI to identify new life-saving drugs, build self-driving cars, and make hyper-personalized product recommendations. But regardless of recent advancements in technology, messaging remains one of the most challenging platforms for AI because it relies on the most significant differentiator that sets humans and machines apart: the ability to understand context and nuance within language and communication.
In other words, there is a huge gap between the general state of AI and AI required for intelligent messaging. Consumer-facing AI has come a long way from the infuriating chatbots commonly used in customer service that often created more problems than solutions. However, implementing AI in messaging requires even more sophisticated and instantaneous technical capabilities that, when coupled with the nuance of language and a small margin of error to work within, present a daunting challenge.
First, AI models built for chat should not be keyword-driven. Rather, messaging requires the use of contextual AI, which is more sophisticated but better mimics the way that we use language. Language is creative and complex, imbued with subtleties, inside jokes, sarcasm, and sincerity. So, while it’s much easier to train AI to generate a transactional response based on a specific word or prompt, messaging requires AI that understands context – picking up on nuance and adapting to conversations as they unfold – to surface information that’s relevant and useful to users on both sides of a chat.
In addition to understanding how language works, contextual AI must also understand how language differs on a global scale – across multiple dialects and demographics. For example, to intelligently deliver the right piece of content to communicate “joy,” AI must be trained on preferences for expressing this emotion in Brazil versus the U.K. versus Japan, then make a recommendation accordingly.
Additionally, AI models developed specifically for messaging environments have to receive, process, and deliver information at an incredibly fast rate. Many other forms of AI, like digital assistants and social feeds, are designed with a built-in buffer, allowing time for information to travel to a server for analysis before returning results to a user. In messaging, though, a conversation needs to flow instantaneously, meaning AI only has approximately 10 milliseconds to collect and interpret data and then use that information to seamlessly make the right decisions, whether that be completing a task or delivering content. This is a tight but crucial turnaround, as even the slightest lag or error can disrupt a conversation. The challenge becomes developing AI that can work under these intense conditions and yield great results for people.
The built-in privacy benefit
The final technological challenge that faces AI specific to messaging is that it must also be implemented securely. Today’s users rightfully have high standards and expectations when it comes to data privacy. AI used within messaging will not be successful if people lack trust in the way data from their personal conversations is managed. The good news is that, by design, messaging is one of the most secure use cases for AI. It requires AI to execute instantaneously and at scale, meaning there isn’t enough time (or power, on some devices) for AI to send information from chat apps, to another server, and back. Instead, AI processes information on the device, so conversation data should never need to be sent elsewhere to deliver results to users. So, though on-device AI is more challenging to build, it ultimately offers major advantages when it comes to data privacy.
How long until these capabilities emerge?
We already see glimpses of this intelligent recommendation technology, from swipe-to-complete to recommended words and emojis within Gmail, iMessage, and other SMS providers. Any major technology company working on its contextual nuance in language and natural language understanding research is testing this technology today. Therefore we can expect a rapid acceleration in the market within the next 12 months, and we should start seeing more tech enabled with this functionality across different applications.
However, we still have about two to three years until mainstream application of this technology is fully integrated within messaging environments and we begin to see AI’s near-instant understanding of context in language. By then, I expect to see many more niche startups cashing in on the opportunities that are unique to messaging. After that, the sky’s the limit to what this type of understanding could mean for the market.
Worth the work
AI in messaging remains difficult to solve because successful application essentially combines the biggest challenges currently facing the AI market: It must be regionally and contextually intelligent, near-instantaneous and, most importantly, secure. However, despite these challenges, messaging is positioned to drive the AI market forward, tackling key tech issues and identifying solutions that will very likely be applied across industry sectors in the future. Once this technology delivers a fully-fledged AI-powered refresh to the messaging market and consumers begin to experience what well-executed smart content recommendations really look like, we will see an explosion of new capabilities in the industry.
