Evaluate Weigh the pros and cons of technologies, products and projects you are considering.

Assessing the business, societal value of AI capabilities

Artificial intelligence isn't just replacing previous technologies and methodologies; it is creating capabilities that weren't possible in the past. At the recent Platform Strategy Summit hosted by the MIT Initiative on the Digital Economy, SearchCIO staff sat down with Sam Ramji, the vice president of product management for the Google Cloud Platform, to discuss how to assess the value of new AI capabilities and quantify a market that isn't a typical, straightforward technology replacement or workload migration. In this video interview, he gives examples of budding AI capabilities and explains the difficulty of accurately measuring the business and societal value of such capabilities.

Read excerpts of the interview below, or click on the player to hear the interview in its entirety.

AI is creating capabilities that weren't possible in the past. How do you assess and quantify that market?

Sam Ramji: This is a really wonderful question. People are starting to understand that we can hand off cognitive tasks -- not just physical tasks -- that we used to ask experts to do. They're not exactly robotic tasks; they're very difficult tasks.

For example, if you look at the oil and gas industry, a lot of oil and gas discovery is reading seismic responses. These things are monochrome; they look like a bunch of waves on a piece of paper. It's going to take a geoscientist with years of experience to recognize the pattern. What they're really doing is mentally extracting a set of features from the data, making some inferences about it and then trying to interpolate that against other forms of information. That other information includes things like maps, other types of surveys or even just information from local people who say, 'Once upon a time, there was a legend that there were puddles of oil in the ground there.' How do you do all that and how do you assess value?

In a system like that, if we can start to create machine learning and AI capabilities that do the feature extraction, generate other machine learning algorithms that can stitch them together and then create a more complete model for a human expert, you have to measure the result in cycle time reduction. Many industrial processes require so much of this rote work from very educated people. Only after that rote work is complete do they get to the high-order work that they really enjoy doing, but the cycle times can be two months to two years.

Now, you can start to value where in your own business you can routinize some of these cognitive tasks and go from very expensive, long cycle times to maybe a five times or a 10 times cycle-time reduction. I feel fortunate that I'm in technology and that I'm on the product development side rather than on the economics or finance side trying to understand the value for an entire market. It kind of escapes my imagination to look at the entire market and be able to say, 'Oh, that's a $4 trillion opportunity.' That's very hard to measure.

If we did measure it, we'd either be too high or too low. There's no way that that complexity is something we can get right, and there's a huge danger in opening markets to set revenue targets against an imaginary share of market. If you set the target too low and you achieve it, you'll be happy, but you didn't do enough work. If you set the target too high and you don't make traction, then maybe you'll despair, give up and try to find something else.

That is something we have to invest in and unfold. We need to keep patiently chasing customer experience and customer value. The faster you can iterate that, the better. To some degree right now, as technology opens up in this field of AI, we have to just trust it and keep going.

There's a societal value that we need to start measuring ... as well as what we could be doing to improve human life.
Sam Ramjivice president of product management, Google Cloud Platform

There are a couple of other examples [of AI capabilities] that I find very inspiring, but also very difficult to measure the value of. One is cancer research -- particularly, the role of radiologists. What we started to find at Google is that when we train an intelligence [tool] on thousands of data sets -- images and radiological data -- against what was the set of physicians' assessment and how it matches the clinical models, we can train an AI [machine] to do actually a much better job of consistently recognizing signs of cancer and matching it to the right stage than we can get from a person.

The AI can't do the treatment, but it's a tremendous augmentation and, perhaps, peace of mind. It's a new tool for the doctor that implicitly taps them into a broader network of intelligence -- not only the machine intelligence, but also the knowledge and wisdom of many other oncologists.

There's going to be a disruption for the radiology profession -- some professionalization and upskilling -- but how do you measure the aggregate economic impact and determine what we should be measuring as value? That's pretty tricky. You should get better cancer outcomes. How do you start measuring that? Somebody will figure it out. It's not going to be me, but I'm fascinated by the application.

The other really motivating example of AI capabilities I found recently is a company called MemSQL, which does real-time intelligence. They built a system on top of TensorFlow that allows feature extraction of video and image databases of missing children. It's able to extract those features in a canonical way, such that they can then look at lots of different social media photos and flag those to the federal government. They're actually helping save 2,000 or more missing children a year with the FBI.

How do you put a societal value on our ability to go through the ridiculous quantities of data that you need to in order to do that kind of facial recognition and pass on the information? There's a societal value that we need to start measuring as well, and understand if we're undershooting or overshooting as well as what we could be doing to improve human life.

To what extent are customers going beyond the traditional cloud offerings and using AI components?

Ramji: The future of cloud is pretty clearly built on AI to improve business outcomes. The main reason we use technology, in any case, is to improve business outcomes. The change is that AI is becoming the leading reason why people are adopting cloud, particularly in financial services and in retail. As AI comes in, companies realize it requires an enormous amount of data, and that data has to be migrated or copied and combined with new sources.

A big differentiator between today's cloud and yesterday's cloud is that today's clouds have a lot of existing data that can be used by companies to combine with their own. Companies are getting more value more quickly because they have a better data set available to train the models. The other big shift is the immediate expectation of, 'Oh, this is going to be extremely difficult for us to do on premises,' because the quality of the compute infrastructure has such a profound impact on the quality of the AI outcomes. You're looking for an AI-tuned compute infrastructure in the cloud in order to get the results that you want.

View All Videos