And we’ll start with the latter – Dreamforce – at which Salesforce announced a bunch of stuff, including a cloud partnership with Google
and new tools designed to further democratize access to AI capabilities.
The former item seems like a classic tech-industry quid pro quo, in which Salesforce promises to use Google’s cloud computing infrastructure as Salesforce continues to grow (it has a similar deal in place with Amazon Web Services, at least) and Google promises to use Salesforce as its CRM platform of choice for the Google Cloud business.
If there’s one piece of it that might be more noteworthy, it’s the promised “deep” integration between Salesforce and Google’s suite of office products which, as the author notes, comes a couple years after a rumored Microsoft-Salesforce acquisition and indirectly positions Microsoft as a common enemy of both companies.
Salesforce’s AI efforts are a little more interesting, at least in the sense of promising something new and powerful. The highlight here is a new feature called Einstein Prediction Builder, which the company says will let Salesforce developers build predictive models about who’s ready to buy or bolt – all without needing to consult with a data scientist or AI specialist. As I’ve said before
, I think Salesforce has a great opportunity to improve its platform’s UX and capabilities, if only it can successfully transition the work of its growing AI research team
from lab to product. My only concern with something like Einstein Prediction Builder is how much AI rope is safe to give Salesforce developers at this point without some guidance from the data science team, and how much stock companies can realistically give the models they create.
Here are a couple of other things worth noting:
Uber AI Labs open sources Pyro, a deep probabilistic programming language (Uber Engineering): I’m less intrigued by the language itself, and more intrigued by the question of whether Uber is a company we can trust to build, launch and successfully manage important open source projects. Its engineering team has been opening up – and open sourcing stuff – significantly over the past year or so, and the company is probably the best suited to drive development in the on-demand, mobile-centric web space it helped create. However, it remains to be seen whether Uber’s overall culture and financial situation(?) will permit it to become the next Yahoo, Facebook, Netflix, Google, etc.
Building AI that can build AI (New York Times): This isn’t the first time Google’s AutoML work has received attention, but this is a good writeup of its broader ambitions to commercialize the technology. The goal isn’t entirely dissimilar from Salesforce is trying to do with Einstein Prediction Builder, only Google would need to execute across a much broader set of users and applications. There’s a conventional (and understandable) wisdom among a segment of AI types that DIY tools for building horizontal machine learning models don’t have a very bright future, but finding the right formula for delivering this capability would be a boon for a cloud provider like Google. Automating the model-building process would bring in revenue with none (or relatively little) of the resource expenditure that comes with building bespoke solutions for customers.