Written by Joel Engardio for Customer Service CEO Noah Horton
Why We Need Artificial Intelligence 2.0
By Noah Horton
Artificial intelligence and machine learning are not as new or unique as you might think. When I tell audiences that Excel is machine learning, they often look surprised. But Excel has been doing linear regression for decades.
The Excel example is what I call AI 1.0, or “supervised learning.” This takes something humans are already good at and gets a computer to do it. We’ve been very successful in AI 1.0.
What’s new is “unsupervised learning” — where we get machines to do things humans have never been good at. This is AI 2.0 and an exciting direction for the future that I’m actively working on now.
I talked about AI with a group of Nexient client representatives recently and explained how unsupervised learning will change how businesses look at data.
I started my presentation with what I call the “all the money in the world” case study. I used the example of a major bank that was under pressure from U.S. regulators to pass a stress test in reaction to the world financial market meltdown.
The bank had failed the test three years in a row, despite having access to every possible resource to tackle the test. It showed that simply throwing money, manpower and algorithms at a problem rarely leads to solutions — especially when the problem involves massive amounts of data.
Let’s break down the insurmountable obstacle the bank faced. The financial stress test included 125 variables from the price of gold to the price of oil in dollars. Then there were eight scenarios for each of the 125 variables. That adds up to 1,000 pieces of data.
It seems like a manageable number, but factor in the test’s six different regulatory models and the bank was facing 1,000 pieces of data to the sixth power. That’s one quintillion items to account for. Can’t picture one quintillion? It’s a one followed by 18 zeros.
The bank spared no expense hiring 1,000 consultants to tackle this herculean task. But even with that much human power, each person still had to look at four trillion models. Imagine the fatigue and errors.
Humans are just not built for looking at 1,000 columns of data on a spreadsheet.
This is where AI 2.0 — unsupervised learning — saves the day. It can discover patterns in data, which helps humans find the needles in a data haystack.
The bank that had repeatedly failed its stress finally passed after using AI. It no longer needed 1,000 consultants working overtime for many months. The job was finished by only three employees in just three months.
Banks aren’t the only businesses that can benefit from unsupervised learning. Retailers are shifting to be more customer-centric and the attributes of individual customers are exponentially more numerous than what banks deal with on financial models.
While the bank has to contend with 1,000 columns of data on a spreadsheet, a giant online retailer faces hundreds of thousands or even millions of columns. People come in all shapes and sizes with a seemingly infinite number of combinations that account for their behavior and tastes.
Previously, retailers would rely on universal truths like “more people drink soda in the summer when it’s hot.” This “data” helped them determine production schedules and pricing strategy. But today, retailers need to know a lot more to stay competitive.
They need to know who drinks soda when the weather is cold. And when married people drink soda. And married people who like dogs. And homeowners who have college degrees but don’t have kids. The combinations are endless.
The power of AI can process the data of 50 million customers and find patterns that create 45,000 sub-populations for determining strategy.
For centuries, the world was built on the following process:
1. Hypothesis
2. Experiment
3. Result: Eureka!
4. Result: Fail (start over)
But this process is now broken because the number of hypotheses needed to generate eureka moments today have become too large.
That’s why the world is adopting a new process:
1. Use Artificial Intelligence
2. See pattern
3. Confirm pattern
4. Act
We must remind ourselves, however, that AI in itself is not what transforms business and the world. We still need humans to interpret what the AI-produced data says. AI simply gives humans the rapid ability in a scaled way to understand customers and create strategy. AI finds the patterns and humans confirm the result — doing something with those results is what’s transformational.
Noah Horton is the CEO of Customer Service, working on the leading edge of artificial intelligence. As Chief Product Officer for Ayasdi, he helped develop AI services and strategies for clients like HSBC, Citi, the US Federal Reserve and United Healthcare. Prior to Ayasdi, Noah ran product and core engineering for Oracle Public Cloud, driving the company’s multi-billion dollar shift from on-premise to cloud-based offerings.