Imagine having 1000 smart interns working for you for a single day.
Now imagine what you could accomplish with them working for you every day — for free. What could 1000 interns do that would make a measurable difference to your business? How could they empower your people?
This isn’t how we talk about AI today, but it could be.
From 80/20 to super-Paretos
The 80/20 rule, also called the Pareto principle (it was named for Vilfredo Pareto, the Italian economist), states that for most events, roughly 80 percent of the effects come from 20 percent of the causes. For example, 80 percent of a business’s profits are said to come from 20 percent of its customers; 80 percent of a business’s sales are supposed to be made by 20 percent of its salespeople. Businesses use the 80/20 rule as a general-purpose guideline in analytics.
Today’s AI is about identifying points where better classifications and assessments can drive significant business outcomes. As a recent HBR article put it, these are not merely “Pareto 80/20 outcomes,” but rather “super-paretos” — points where there can be a massive discrepancy between input and output.
This really matters: it’s not cheap or risk free to collect massive amounts of data, reformulate a problem, apply deep learning, and put these insights into production — particularly given the requirements of deep learning and other statistical methods for massive amounts of data.
What if AI was more like our 1000 interns?
How would that change the game?
A smart, college-educated intern who understands your domain wouldn’t need tens of thousands (or millions) of examples to learn how to deal with a new problem. Indeed, they’d turn and walk away if you told them they had to think this way. In the real world, you’d provide a few examples, expect that they can pick up enough of the language of your space, test to make sure they have the right ideas, and have them make a few attempts at the problem — then ask them to explain why they made those choices, assessments, or adjustments. Once you were sure they were on the right track, you’d let them work on successively larger and larger sets, each time making sure they remained aligned with your expectations.
How would your 1000 interns spend their time? You could require them to read every document ever created in your field, read the documents in your corporate database and review a library of material for relevance. With this new knowledge, they would be able to transform your business and workforce by super-charging every knowledge gathering task.
Just a couple of examples of these tasks:
- In the field of finance, to understand the differences between individual mortgage documents — what special clauses have been added here and there.
- In the field of finance, to understand the impact of news and annual reports on the value of a given enterprise.
- In the field of science and technology, to immediately apply complex models like “Technology Readiness Level,” a formal model NASA and the US government use to understand a technology’s maturity. (Imagine, as a futurist or investor, being able to have every technology in the world ranked — and seeing how that ranking changes in the literature, year by year.)
Working memory, intelligent chunking, and the third wave
Let’s take it a step further: How about 1000 smart interns with absolute recall, and the flexibility to hold many more items in short-term memory than the average human? The theory of chunking suggests that most of us can hold around seven unique items in our short-term memory at a time. Indeed indications are that higher IQ correlates with the number of “chunks” available to short-term and working memory. Imagine that your one thousand interns have absolute recall and genius-level capabilities around manipulating chunks of information.
A new flavor of AI
The 1000 interns scenario is what we’re looking toward with emerging AI capabilities and is a substantial step beyond today’s deep learning and pattern matching solutions.
The new approach to IT is able to create and reason about abstract knowledge. In a DARPA presentation about the much sought ‘third wave’ of IT, John Launchbury states that “the systems themselves will over time build underlying explanatory models that allow them to characterize real-world phenomena.” We agree.
My next post introduces more details about the waves of AI.