The Data Scientists In Santa’s Workshop

 Editor’s Note: Tom Limongello is vice president of product management at digital marketing firm Crisp Media. 

It seems like everyone has AI on their wish lists this season.

For me, there was Xmas in July. I was nearing the end of my research for a post on office productivity, and I got a strange answer to what I assumed was a simple question: “Why would Google sunset the nice little feature that let me send a calendar invite through Gmail?”

The founder of, Dennis Mortensen, responded that sending a calendar invite doesn’t help schedule a meeting. He said every appointment is a negotiation where each party has to attribute a unique value to the other person and each time block. But then he lost me when he suggested the best way to negotiate this problem was through a robot disguised as an email address.

You see, I believed that we belonged to a generation whose progress hinged upon an ‘Uber for X’ understanding of the world. I believed we no longer wanted assistants like our parents’ generation had. I felt I had all the help I needed as mobile apps automate the dispatching packages and taxis through the luxury of credit card consent. It’s a choice. Should I control every decision in my life through better and better processes, or would I like help that could come through intelligent discourse?

Up to now, we’ve chosen the former, as Brian Christian, author of “The Most Human Human: What Artificial Intelligence Teaches Us About Being Alive,” writes:

“We are replacing people not with machines, nor with computers, so much as with method. And whether it’s humans or computers carrying that method out feels secondary.” 

What’s changed? Certainly from all the coverage we’ve seen, ’tis the season to hype artificial intelligence.

Wired’s Kevin Kelly shared in “The Future of Artificial Intelligence” the magic formula: “the business plans of the next 10,000 startups are easy to forecast: Take X and add AI.”

Columnist David Brooks was on it, too, in “Our Machine Masters” claiming only the largest tech companies like Google can play in any AI games. If the goal were to make a machine better at competing with people or replacing them, Brooks might be right.

But this next stage of building AI can’t be about building a faster Deep Blue that memoizes all the right moves in chess so it can win; it’s about collaboration, which is a slower, more constructive, endeavor.

How Do Computers Think?

Back in the days of Alan Turing, computer programmers had to encode every possible computer action, and reasoning was the only analogy to human thought that computers could achieve. Now, machine learning is helping computers acquire knowledge at a more basic level, but many tech companies conflate the idea of knowledge acquisition with learning.

In a recent podcast Danko Nikolić, who has popularized the theory of practopoesis explained that machine learning only propagates knowledge horizontally, “big data is not enough, you’re never going to have a big enough computer to store all the information…adaptive learning is required.”

Monica Anderson, CTO of considers that artificial intuition and epistemology are the next frontiers after computer reasoning. Reasoning is a linear process that accounts for only one hundred thousandth of what humans do with their brains. Computers will achieve intuition when they can, “perform those human activities that are normally thought to NOT require intelligence;” that can be performed without conscious effort such as walking, listening and talking.

The trick, however, is that computers can’t learn if we program every last function. Instead we need to abstract learning from recognition of digital sensory data like text, audio, images and video.

Can Startups Help Computers Think Differently?

If Kevin Kelly is right, and the next wave of startups will just need to “take X and add AI,” what can they do to deliver practical examples of AI today?

First, the requirements for recognition of objects via machine learning are coming way down with the help of GPUs.

Adam Coates, Director of the Silicon Valley AI lab for Baidu, understands the precise needs for scaling computer power to further his team’s research.

Although he admits that graduate students don’t have the resources to build large neural networks, which is why he’s now at Baidu, his early research at Stanford proved that stringing together a few GPU cores could massively decrease the required number of computers for enabling photo recognition by 80 percent or more. The once $1 million price tag for the Google brain project led by Andrew Ng, who is the chief scientist at Baidu, can now be achieved with $20,000.

Startups also have the ability to make new products without fearing backlash that significant UI changes cause in big companies. Facebook needs to keep most of its users safe from drastic UI changes, and therefore there must be a greater focus on basic research than product development, and that research then gets tucked into smaller corners of the organization or into separately operated acquisitions. Google has access to your searches and email, Apple has immense scale with Siri, and Facebook processes the world’s largest trove of photos.

Yet, these companies can’t break their well-learned UIs without consequences, even when it might facilitate better data collection. The startup elves in the co-working spaces by contrast are trying to stand out based on differentiated UX and are forced to get to practical differences faster and more often. Moreover, when the solution to a problem is uncertain, it is not a safe bet that a few well-scaled companies can match the variation in approach of 10,000, focused, new startups with license to build user experiences from scratch.

Ultimately, there’s no alternative to search. Until now, search had been the fastest tool for rationally ranking results once and for all. The intermediate approach that saves typing on mobile has been to offer predictive results, but prediction without interaction can be unsatisfying, because it comes before you’re ready to believe what the right answer might be. A discussion, in contrast, enables negotiation within a certain context.

Startups need to design programs that can master the art of hanging out and making it easier to converse and play. Of course, startups need to achieve this in a way that people accept, without wigging us out for being creepy, violating our privacy or frustrating us from an uneven value exchange. Some startups are already showing signs of success in getting users to submit that hard-to-reach text, audio, image and video data.