Notes from the Hype Machine: Making Sense of McKinsey’s $3.5 Trillion AI Report

mckinsey_report_cover

 

"We estimate that the AI techniques we cite in this briefing together have
the potential to create between $3.5 trillion and $5.8 trillion in value..."
 

Notes from the AI Frontier, McKinsey Global Institute, April 2018

 

These estimates are annual. To put the low end — $3.5 trillion — in perspective, nominal annual global GDP is around $80 trillion.

So, according to McKinsey, AI’s impact will be big. When will it happen? They have no idea.

But there’s a lot more to the story...
 

The Origin Story of Consultant Content Strategy

Consultants with some knowledge of artificial intelligence get asked these kinds of questions all the time:

  • How big will the impact of AI be?

  • How will AI affect my business?

  • Where are the biggest opportunities for AI investment?

In other words, “Can you take a wild guess at predicting the future of artificial intelligence BUT make it look legitimate enough for me to put it in my board presentation?”

Now, step into your consultant time machine, transport yourself back 22 years, and you see the same questions being asked — but replace the words “Artificial Intelligence” with “Internet.” We were there. This is exactly how it happened. Lots of people made a lot of money. But there were also a lot of stupid projections thrown about that never came true.

 

McKinsey Steps into Gartner Territory

Generally speaking, McKinsey gives good advice — better than Gartner, who have unhelpfully predicted that AI will create $1.2 trillion of “economic value” in 2018. In the case of McKinsey’s “Notes from the AI Frontier”, however, they have combined some useful thinking and case studies with an irresponsible presentation of market projections.

Why irresponsible? They’ve combined a very reasonable taxonomy of neural networks, their history, and their applications with scintillating market projections no one could reasonably use. To make their numbers look more legitimate, they’ve even packaged the whole AI impact model in a nifty little interface.

We recommend playing with the market sizing model. It’s strangely fun, provided you are not a UX designer looking for inspiration or strategist trying to make actual business plans.

 

 Sample from McKinsey's " Visualizing the uses and potential impact of AI and other analytics "

Sample from McKinsey's "Visualizing the uses and potential impact of AI and other analytics"

 

Alternative Hyperbole-Free Headlines

As is often true in journalism and content strategy in general, the impressive top line numbers, which we should not forget are totally made up, obscure the most important messages. In consulting, we call these caveats “Dependencies” that may interfere with the proposed future. So let’s take a look at some of the other headlines from the paper:
 

The Potential of Artificial Intelligence Faces Substantial Threats

McKinsey cites a litany of obstacles to AI adoption that we could jeopardize their rosy projections. The top 10 include:

1. Availability of training data
2. Opaque algorithms that inhibit understanding “why” something happens
3. Bias in data and algorithms
4. Non-transferability of algorithms across problem sets
5. Continuous model updating to match shifting reality
6. Disconnect between algorithm authors and domain expertise
7. Privacy and information security threats
8. Regulation
9. Corporate bureaucracy
10. Only 10,000 people have the skills necessary to tackle AI problems
 

AI Gilds the Lily of Analytics

Only 16% of AI use cases are classified as “greenfield”, meaning cases where traditional analytics techniques were inapplicable.
 

Traditional Analytics Techniques Will Have a Bigger Impact than Artificial Intelligence

The impact of traditional analytics techniques (60%) will outpace the impact of artificial intelligence applications (40%) in terms of total value creation.
 

The Impact of AI Could Be Big, but the Timeline is Uncertain

McKinsey’s projections have no timeline associated with them. This should tell us something.
 

What’s a neural network and why should I care?

While the data may be attention grabbing, and some of the caveats buried, McKinsey’s report does a good job of explaining neural networks and their applications. These definitions are both helpful for making sense of their projections and providing the executive-level briefing on the technologies at play.

Despite our reservations on the market sizing, we recommend spending some time with McKinsey’s report to come to your own conclusions.