Currently, we are seeing three waves of artificial intelligence: expert systems, machine learning, and goal-based AI. The first wave consists of expert systems which are rule-based, very narrow and rigid, have zero learning capabilities, and are poor in the real world. The second wave consists of machine learning which utilizes probabilistic and statistical techniques and is good at classifying and predicting. However, it has limited ability to understand context and needs a lot of data to continuously learn and improve. The third wave is goal-based AI which is a futuristic and contextual adaptation. During this wave, AI has the ability to understand context and reason, and it requires less training data.
Data is essential
Just like oil is necessary to drive a car, data is necessary to reach the third wave of artificial intelligence: goal-based AI. With that said, not all data is created equal.
“Labeled data is the new oil. Unlabeled data is the new dirt.”
Unlabeled data can be obtained easily from the world. There’s no explanation, just data. An example is a photo you’ve sourced off of the internet.
On the other side, labeled data contains sets of unlabeled data equipped with meaningful tags, labels, and classifications. An example is a photo you’ve sourced off of the internet, but it has a description (i.e. recently engaged couple on a beach in California). There’s much more insight in labeled data.
Once you have data, you can build an AI product
If you have clean, labeled data, how do you actually begin to build an AI product? You have to establish a data infrastructure. That data infrastructure involves the following steps:
Understand the model: You have to take the time to build an AI model and truly understand what you are trying to achieve. This is where you take your ideas, align your ideas to the problem or problems your trying to solve, and decide on the approach to get there. You also have to figure out what data inputs you’ll need at this phase of building your infrastructure.
Here are some initial questions that need answers to include:
- What problem are we solving?
- Why is that problem valuable?
- Do we have relevant data that may solve this problem?
- How do we measure success?
- Are we building a system that learns?
Get labeled data: I’ve already stressed the importance of data: the more data, the better. You have to collect the right data in the right format so that your model can easily digest and learn from it throughout the process.
Construct a feedback loop: Feedback loops are extremely important if you want your AI product to continuously learn. A feedback loop is just like any other set of steps for improving processes based on feedback (i.e. think customer service). The goal is to collect as much user feedback (yes, more data) from using the product. From there, you can use this feedback to improve your model over time.
Measure relative performance: You must determine the metrics you want to measure, and these will vary for different products. These are the categories of metrics you should consider: quality of data collected that can be used for training; quality of the modeling in order to generate the right output; AI flywheel growth measurement (for some companies); and, customer success metrics (for your particular business, including quality of output to your users).
In order for artificial intelligence to truly reach goal-based AI in the future, we need entrepreneurs, innovators, and disruptors to continue to build AI-driven products. Once we get to this third wave of AI, it will finally have the ability to understand context and reason and will require less and less data.
About the Author