For anyone who’s getting their first introduction to Einstein AI features, now rapidly being deployed in the Salesforce platform, two questions may arise. Why is this different from past cycles of AI excitement and disappointment? Why is an integrated layer of AI capability so crucial to a modern CRM?
Six decades of research, development, and sci-fi hype have followed the first famous “Dartmouth memo” – which naïvely proposed “a 2 month, 10 man study” as being enough to make “a significant advance” in “how to make machines use language, form abstractions and concepts, solve kinds of problems now reserved for humans, and improve themselves.” The challenge was clearly much greater than first realized: most observers agree that people had vastly underestimated the portion of our intelligence that comes from what we know, rather than how cleverly we think.
In laboratory settings and research universities, the focus was on making software smart. The challenge of making an overall system knowledgeable was harder to address, in a world of limited connection – with international bandwidth availability at a slow drip of less than one Terabit per second as recently as 2010. At that time, we began an extraordinary surge of new connectivity, already moving into tens of Tbits/second on major routes such as US/Europe and US/Asia (with global capacity growing at order of 40 per cent per year).
In such a massively connected world, it becomes attractive to flip the AI process: to move from a model of “knowledge engineers,” slowly structuring human expertise into unwieldy collections of brittle and soon-outdated rules, to a far more scalable and sustainable model of dense and rich networks of representation built quickly by machine-learning algorithms.
Machine learning would be only a laboratory curiosity, in the absence of readily accessible sources of relevant knowledge to feed the beast – but the raw bandwidth of connection, already discussed, becomes a cornucopia of insights when it is fed by the stream of situational and personal data that arises in a world of smartphones.
Accurately described as “ubiquitous, addictive and transformative” in a special report by The Economist, the smartphone generates tuples of interest and action and location: moments of opportunity to recognize and meet a need, perhaps even before a person would have been able to describe what would serve or delight.
Adding value beyond that raw stream of events is the process-based insight of a full-featured CRM – breaking down past silos of marketing and sales and service, to build from those combined perspectives an organic understanding of customer journeys and customer communities. The deployment of machine learning capability, into an environment where the world’s most vigorous and capable customer connection is already happening, creates the extraordinary phenomenon of Salesforce Einstein: not an ignorant genius waiting to be taught, in the manner of AI past, but an attentive and insightful assistant who shows up and watches you work; who learns, and thinks, and soon begins to offer useful suggestions.
The final crucial link is the accomplishment of making Einstein’s suggestions conveniently accessible, credible, and valuable – because without adoption, there is no feedback loop from which to refine and focus the learning. Again, the smartphone era provides a useful context, in that people have already learned to notice and filter the sources of information that our connected world clamors for the chance to share. In the same way, Salesforce Einstein says “this is why I am making this recommendation” and “this is what you told me I should notice and consider: do you want to change that?”
It’s intriguing, and useful, when machine learning dramatically streamlines the process of translating between different human languages, or when it startles champion players with brilliant new moves in the oldest game known to human history. Don’t let those gems distract you from the far greater number of semiprecious stones to be found in AI’s rock garden.
When a smartphone takes a store-shelf inventory by using machine vision to analyze a photograph of a retail display; when a chatbot, taking a medical history, notices a rare situation that requires asking unusual questions; in these and any number of other everyday situations, we can become more productive in environments where data emerges and where a tireless and consistent machine intelligence can add value.