If you’re a business executive (rather than a data scientist or machine learning expert), you’ve probably been exposed to the mainstream media coverage of artificial intelligence or AI. You’ve seen articles in The Economist and Vanity Fair, you’ve seen emotional stories about Tesla Autopilot and the threat of AI to mankind by such luminaries as Stephen Hawking, and you might even have seen Dilbert make jokes about Artificial Intelligence and Human Intelligence.
So if you’re an executive who cares about growing your business, all this AI media coverage may prompt two nagging questions.
First, is the business potential of AI real or not?
Second, how does AI apply to my business?
The answer to the first question is that the business potential of AI is real. Today businesses can start to apply AI to change the economics of automating business processes requiring human intelligence. AI could allow you to increase the volume of work through a human intensive business process by 100X while reducing unit economics by 90%.
The answer to the second question will take longer. First, we have to dispel myths about AI that have been propagated by the mainstream media. Once we have dispelled those myths, then we will have a framework for you to understand how to effectively apply AI to your business processes.
Myth 1: AI is Magic
Much of the mainstream media is portraying AI as magical and mystical and we just need to applaud the master magicians at companies such as Google, Facebook, Apple, Amazon and Microsoft. This portrayal is a disservice. If we want businesses to adopt AI, then we need business executives to understand AI. AI is not magic. AI is data and math and patterns and iteration. If we want AI to be adopted by businesses, we have to be more transparent and explain the 3 key interlocking concepts of AI.
Training Data (TD). Training data is the initial data set that the machine will learn from. Training data has inputs and pre-answered outputs so the machine learning model can look for patterns for any given output. For example, the input could be customer support tickets with email threads between a customer and a company support representative (CSR). The outputs could be a categorization label from 1 to 5 based on the company’s specific category definitions.
Machine Learning (ML). Machine learning is the software that learns patterns from the training data so it can apply those patterns to new input data. For example, a new customer support ticket with email threads between a customer and a CSR arrive and the machine learning model would predict a categorization and tell you how confident it was about that particular prediction. The key feature of machine learning is that it learns rather than applies fixed rules. So as it digests new data it adjusts its rules.
Human-in-the-Loop (HITL). Human-in-the-loop is the third core component of AI. We cannot expect machine learning models to be infallible. A good machine learning model may only be 70% accurate. So you need a human-in-the-loop workflow so humans can be used when the confidence of the model is low.
So don’t be fooled by the myth that AI is Magic. You now have the AI formula from which you can build a basis of understanding. AI = TD + ML + HITL
Myth 2: AI is only for the technology elite
The media coverage seems to imply that AI is only the domain of the technology elite – companies such as Amazon Apple, Facebook, Google, IBM, Microsoft, Salesforce, Tesla, Uber – who can afford to assemble large teams of machine learning experts and invest $100M. That perception is wrong.
Today it is possible to start to apply AI to one of your business processes for less than $100,000. So if you’re one of the ~26,000 companies in the US with revenues greater than $50M, you can start to apply AI for an investment of 0.2% of your revenue.
So AI is not just for the technology elite. It’s for every business.
Myth 3: AI is only for billion dollar new problems
The mainstream media stories that tend to get covered are the futuristic ones such as self driving cars or autonomous drones making deliveries. Companies such as Google, Tesla, and Uber are investing hundreds of millions of dollars into being the first to market with a driverless car because of a “winner takes all” mentality. This coverage can give the impression that AI is only for billion dollar new problems. But that’s a mistake.
AI is also for million dollar existing problems. Let me explain. One of the core imperatives of any business is to understand your customers. This was true in the first markets such as the agora in ancient Greece and the forum in ancient Rome when buying and selling was done in person. It’s still true today even with the explosion of buying and selling on the Internet. Many companies are sitting on a treasure trove of unstructured data from their customers either in email threads or comments on Twitter. AI can be applied to those challenges of categorizing support tickets or understanding sentiment in tweets.
So AI is not only for billion dollar “exciting” new problem like driverless cars. AI is also for million dollar existing “boring” problems such as understanding your customers better through support ticket classification or social media sentiment analysis.
Myth 4: Algorithms are more important than data
The mainstream media coverage of AI has biased towards focus on the Machine Learning algorithms as the most important element. They seem to equate the algorithms with the human brain. They imply this is where the magic happens with ever more sophisticated algorithms that ultimately will surpass the human brain. The stories about Machines defeating Humans in Chess and Go are examples of this media treatment. The media is focused on “deep neural networks” and “deep learning” and how the machines made decisions.
This coverage can give the impression that for a company to have a chance of applying AI they need to hire machine learning experts to build the perfect algorithm. But if a business did this without thinking about how it would also get high quality, high volume customized training data from which the machine learning model could learn you would have a mismatch between expectations (“we have a great algorithm”) to outcome (“our model is only 60% accurate”).
Buying one of the commercial Machine Learning as service offerings today from Microsoft, Amazon and Google without having a plan or a budget for training data, is like buying a car with no access to a gas station. You just bought an expensive lump of metal. The car and gas analogy breaks down a little though because machine learning models get better as you refuel it with more and more training data. That would be like a car improving its mpg after each tank of gas.
So, training data is even more important that gasoline. For a more in depth coverage of this myth you can read our previous post More data beats better algorithms
So the key takeaway is that the quality and quantity of your training data is at least as important as the algorithm, so make sure your plan and budget to deploy AI reflects that.
Myth 5: Machines > Humans
For the last 30 years the media has loved to portray AI as meaning Machines are better than Humans whether it’s Schwarzenegger in the Terminator or Alicia Vikander in Ex Machina. It’s understandable why the media does this because it wants to set up the simple narrative of Machines versus Humans and who will win. But this is a misrepresentation of what is actually happening.
For example, the recent the media coverage of Google’s DeepMind/AlphaGo victory over Lee Sedol was simplistically portrayed as Machine defeats Human. This is an inaccurate representation of what is actually happening. The more accurate description would be Machine plus many Humans defeats single Human.
The core reason to dispel this myth is that Machines and Humans have complementary capabilities. Take a look at the picture above. Machines are great at handling structured computation. They would do great at the “Find the Eigenvectors” task and not so great on the “Find the Leopard Print Dress” tasks. Humans are great at discerning meaning and context. They would do great at the “Find the Leopard Print Dress” task and not so great on the “Find the Eigenvectors” task.
So the right framing within a business context is to realize that Machines and Humans are complements. AI is both them working together.
Myth 6: AI is Machines replacing Humans
The mainstream media likes to paint a dystopian future because they think it drives attention. That may be true, but it does a disservice to the true understanding of how machines and humans can work together.
For example, let’s consider again the business process of categorizing support tickets. In most businesses today this is a 100% human process. As a result the process is slow and costs scale linearly, so you limit how much you do. Imagine now after categorizing say 10,000 support tickets you have a model which is 70% accurate. Being wrong 30% of the time is unacceptable, but that’s where Human-in-the-loop comes in. You could set the acceptable confidence threshold as 95% and only accept the model output where the model’s confidence level for any given output is 95% or higher. So initially the Machine Learning model may only have a small percentage of the work, say 5-10%. But as new human labeled data is created that is fed back into the Machine Learning model and it learns and improves. So over time the model can handle an increasing percentage of the customer support ticket classification work and the business can greatly increase the volume of tickets it classifies.
So by combining Machines and Humans you can increase the volume, maintain the quality, and decrease the unit economics for an important business process.
This dispels the myth that AI is about Machines replacing Humans. The truth is that AI is about Machines augmenting Humans.
Myth 7: AI = ML
This brings us to the final Myth about AI being perpetuated by the mainstream media. This myth is that Artificial Intelligence and Machine Learning are the same thing. This myth can lead to unrealistic management expectations that just by buying one of the commercial Machine Learning as service offerings from Microsoft, Amazon or Google a company will magically be able to deploy AI into production.
You need training data and human-in-the-loop workflows in addition to Machine Learning to have a chance of a commercially viable AI solution.
Machine Learning without Training Data is like a car without gasoline. It’s expensive and won’t go anywhere
Machine Learning without Human-in-the-loop will lead to bad outcomes. You need humans to override the low confidence predictions of the Machine Learning model.
So if you’re an executive thinking about applying AI to your business then you now have a framework. You can replace the 7 Myths with the 7 Truths of AI
- Truth 1: AI = TD + ML + HITL
- Truth 2: AI is for every business
- Truth 3: AI is for million dollar existing business problems
- Truth 4: Algorithms are not more important than the quality and quantity of training data
- Truth 5: Machine complement Humans
- Truth 6: AI is Machines augmenting Humans
- Truth 7: AI = TD + ML + HITL