Uber shows how AI is more about automation than revolution

May 1, 2019 Derrick Harris

This post originally appeared as part of the April 25 Intersect newsletter. Click here to view the whole issue, and sign up below to get it delivered to your inbox every week.

Uber's recent S-1 filing sheds light on some of its artificial intelligence efforts, which, while intimidating on the surface, should probably give traditional enterprises the confidence that they can do AI, too. Uber's entire business might seem intimidating because of just how differently companies can operate when they're built around data, algorithms and mobile apps from the start. But, in the end, Uber is using AI to power automation in ways that any large organization with a solid understanding of the underlying technologies can likely pull off itself.

We'll start with the intimidating-dare we say "revolutionary"-part, which can be summed up in this industry-analyst quote from a Wall Street Journal article about Uber's AI efforts: "Its drivers don't have human bosses. They literally answer to algorithms." Uber hasn't created a super-intelligent system that's going to pass any sort of Turing test or threaten humanity, but the company does trust its data and its algorithms enough to let them run the show when it comes to scheduling rides. That's a level of trust in algorithms that many companies might only ever aspire to.

And getting there took Uber collecting, and then analyzing, quantities and varieties of data in a manner that's a little easier for companies born during the "big data" era-if only because the concepts (e.g., data science)  and systems (e.g., Hadoop) already existed. There was no "What's our big data strategy?" discussion. Big data was the strategy.

When you peel back the layers of the onion and get to the core, though, you see that Uber's AI strategy is relatively simple-automate tasks that can be done faster by machines than by people. This definitely applies to the complex logistics of ride-scheduling and dynamic pricing, but also to other areas where Uber uses AI. For things like building conversational interfaces; scanning drivers' licenses and restaurant menus; and, eventually, operating self-driving cars.

However, Uber still employs more than 20,000 people (not including drivers) to carry out the parts of the business that can't be easily automated and that will always require a human touch. These include things like software development, business strategy, marketing and government relations You don't need to accept Uber's myriad cultural shortcoming to appreciate its approach to AI as a force multiplier for its core business of getting people (and things) from Point A to Point B.

The driving question for any company seriously looking to adopt AI, or even to modernize how it does IT overall, should be: "Where are we expending too many resources on undifferentiated tasks, and are there technologies and/or techniques to automate those tasks?" The answer to the second question is probably yes, and the solutions will probably require a much smaller engineering investment than Uber is willing to eat because it like to build everything itself. 

The important thing is to first identify the problems or opportunities, rather than locking into a solution ("We are going to remake ourselves with AI!") and then looking for places to apply it. Because the ideal solution might look different for everyone and every application depending on business requirements, legacy systems, and what kind of talent you have in place to carry it out.

For more on the topic of automating monotonous tasks and even bridging the AI gap by including humans in the training process, check out this trio of recent posts:

About the Author

Derrick Harris

Derrick Harris is a product marketing manager at VMware.

More Content by Derrick Harris
Previous
How Pivotal Works With Customers Who Use SAFe
How Pivotal Works With Customers Who Use SAFe

Next
How to Distribute Data Across a PCC Cluster: Replicated or Partitioned?
How to Distribute Data Across a PCC Cluster: Replicated or Partitioned?

Pivotal Cloud Cache dramatically reduces latency when users request certain types of data. Read the blog ab...

×

Subscribe to our Newsletter

!
Thank you!
Error - something went wrong!