Creating a Winning Approach with A.I.

I attended a talk by Andrew Ng at Stanford last week where he discussed the future of artificial intelligence.

I attended a talk by Andrew Ng at Stanford last week where he discussed the future of artificial intelligence.

He reinforced my thinking around the importance of data to the success of any company working on a specific application of A.I. He also confirmed my hypothesis that the large players like Google and Baidu are open sourcing their algorithms and applications in order to gather even more tagged data for their training sets.

While there are emerging techniques that may allow for training off of smaller data sets or ones that are completely unstructured in the future, today these techniques are still in their infancy. As a result, to build a defensible business, a company must create a large, proprietary data set that will lead to the best trained algorithms in its field.

At the same time, it is very difficult to get access to, or to amass, a significant amount of data, especially that is clean and tagged. Even the large companies are still working hard to grow their datasets. As Ng stated, he often releases products at Baidu with the goal of gathering more data, not of driving more revenue.

So how can a new start-up leveraging A.I. create a winning approach?

First, companies should focus on a specific use case where they can get targeted data and become very accurate at one focused application (e.g. inventory forecasting or image recognition in drug discovery)

Second, since data is the scarce resource, start-ups need to either convince an established corporation to access their data by providing valuable insights the partner would not have the capabilities to extract, or grow their data by creating a product that incents users to provide data.

As an example of the partnering approach, Onera is working with retailers to optimize their inventory, logistics, and costs across online and offline channels, providing significant value and collecting their data in the process. Facebook is a great example of a product that collects data by incenting users to input information. Have you seen Facebook suggest a location to tag your post from? Have you accepted or rejected that information? If so, you’ve helped Facebook collect and tag their geolocation data.

In either scenario, it is important to be hyper focused on the quality of the product and service that you’re providing to partners or end users, because that is what provides superior value, attracting users that will provide the data.

Baidu is a great example of this. Ng mentioned that voice usage is growing 100% a year, such that Baidu is going to train its speech analysis algorithms on 100,000 hours of audio, which would take 10 years to listen to! They are also training their facial recognition algorithms on 200 million images, whereas many leading papers use 15 million images.

These numbers are staggering, and illustrate what level of scale is being used at the highest levels in the industry.

At the same time, there are so many processes that will be transformed by artificial intelligence and machine learning. There is an amazing amount of opportunity for new companies who focus on a specific use case with a best in class product where they can win with their unique ability to access data.

For further musings on artificial intelligence and its implications for products and start-ups, follow me on Linkedin, on Twitter at @mkhandel and on Medium at @medhaa