Global site tag (gtag.js) - Google Analytics

The initial AI successes of social media companies have masked an important dynamic that will constrain the widespread adoption of AI. Each initial success harnessed hundreds of millions of data samples, tens of millions of dollars, and hundreds of engineers. These are fundamentally brute force exercises. What’s more, social media giants like Facebook, Amazon, Microsoft, Google, and Apple have all been aggressively acquiring AI startups for the last decade1, with a total of 112 acquisitions thru June 2021 (Figure 1). That’s all well and good when you’re operating a hyper-scale platform that can leverage a solution to generate $1+ Billion of value. The rest of the world doesn’t enjoy this luxury.

Figure 1

Making AI work in government, and even traditional industries, will require a different approach. As I’ve pointed out before, all of the action is in narrow-purpose AI. Each opportunity is likely to be materially different from the next. What’s more, government and traditional industries face a “long tail” of AI use cases that have lower value individually – say $0.5-$5.0 million each – and significant value only in aggregate. These use cases are typically characterized by:

small data challenges
limited budgets
a shortage of Engineering talent
• low impact

Given these dynamics, the techniques and methods of the hyper-scalers simply aren’t practical for the rest of us. We, data practitioners, need to develop new methods, processes, techniques, and tools to make AI economically viable in the long tail. After all, any organization might have 10,000 opportunities to apply AI. But what organization can afford to hire 10,000 machine learning engineers in order to build, deploy and maintain 10,000 custom models? And let’s not overlook the fact that data science is a team sport; every machine learning engineer is accompanied by data engineers, software engineers and domain subject matter experts.

The talent shortage may be the most critical factor. As a recent Brooking report discussed, while the US currently has a strong market for skilled technologists, we are:

“…about to drop off a cliff without a strong incoming set of STEM-skilled students. For the US (80th percentile in current market and 18th percentile in future market), this problem is likely to get worse: the majority of our current STEM students are from other countries and are likely to return to their home countries upon graduation. The US position (Figure 2) may be artificially optimistic, since the data does not distinguish the citizenship of current STEM students within the country.”

As my Math Professor used to say, this view is correct but not complete. Inspiring more students to embrace STEM isn’t the complete answer. We also need to democratize data science by making the knowledge, skills and practices consumable by people in all walks of like. The American people are the strongest, most important asset the country has. In order to maintain our competitiveness on the world stage, we need to give every citizen the opportunity to be a citizen data-scientist. What steps is your organization taking to improve America’s future competitiveness on the world stage?

Figure 2

Michael Conlin

Michael Conlin​

Chief Technology Officer​
Phone: (703) 216-5856​​

Suggested content for you