Today’s Guest Keynote at the 2017 Sitecore Symposium featured Kenneth Cukier, author of Big Data.
The following are my notes from the session:
Question: Is AI overhyped, underhyped, or don’t know?
Over: about 6%
Under: about 7%
Don’t know: Not enough to add up to 100%
Nascar racing has a couple key questions. One is when to bring your car in. The yellow flag means you need to decide whether to bring the car in for new tires. Most managers, make a gut decision based on understanding of some but not all features.
Introducing Cynthia Rudin from MIT Sloan School of Management. She’s a statistics geek. She looked at all the data around of the race like track dimensions, slopes, average change time, etc. She came up with one conclusion: it’s too complex. You can just state the rule set and have a human make the decision.
But in the meantime, she had to became an expert and understand the nature of tire decay. There is a neighborhood affect. The racers affect each other. The fresh air affect: racers at front go faster. There are two types of tracks: 1. tracks which wear tires out a lot 2. those that don’t.
What is important to this: Technology, data, mindset.
We know the tech and the data. But the third factor of mindset is new. There are things you can do with a large body of data that you can’t do with a small body. A change in scale leads to a change in state.
How do you make decisions with all this information? Board of Directors around a table is the old way. Instead, you can rely on an algorithm. The how is machine learning.
Definition of Machine Learning: Instead of giving you all the rules, you give the machine all the data and allow it to infer what to do based on desired outcomes.
Take the game of Go. It’s infinitely complex. Deep Mind AlphaGo just beat the world champion. An AI just learned to bluff and beat the best in Texas Hold Em Poker. Google uses AI when fixing spelling typos on searches.
Where to use AI or machine learning? The answer is to ask the question, “What does technology lower the cost of?” The computer lowered the cost of doing math or arithmetic. You can apply math to things that never you never thought applied like photography.
AI lowers the cost of prediction.
Machine learning turns things into prediction problems and solves them.
Harvard and Stanford Study: Showed a computer lots of biopsies and told it the patient survival rates and nothing more. They found out that the algorithm found 11 different telling elements. turns out that doctors only knew of 8 of those.
The algorithm is now gunning for any decision makers job.
This is now about every activity. Think of the world as a platform for the collection and analysis of data.
New world: the value of the data isn’t inherent. It’s in the secondary events or results. The value of the data lies in the reuse of it.
Statistic: An orange car is the one least likely to have mechanical problems.
Statistic: When Caesars went bankrupt, they found that the value of the data was $1B.
Statistic: Microsoft values Linked In, a company with a relatively small amount of revenue at $26B. It’s because of their data.
AI can spot problems humans miss and search entire problem space. Take Airbus. It wanted to create a cabin partition that was optimal for weight, etc. They put it through a generative design process. The AI came up with 10,000 options. The end result was something a human couldn’t have created in the frame for the partition. It’s a tinker toy set from Aliens.
What Does This Mean For Business?
Things have sped up too quickly and it’s too complex. First thing to do is measure absolutely everything. Then personalize and individualize. It’s about learning. By learning and iterating, it can reach an outcome. Hence the need for lots of data and measurements.
There is a legacy advantage with machine learning. The advantage flips to the legacy player because they are the ones with the data.
Embrace AI but understand “ground truth.” We can’t lose our morality, for example.
It will be humbling when the machine is better than our experts.