Thought Leaders in Artificial Intelligence: Ashutosh Garg, CEO of Eightfold.ai (Part 5)

Sramana Mitra: In 1999, I started one of the first ever online fashion company. A very small subset of merchandise applies to me in a store. I’m a petite, dark-skinned woman with dark hair and dark eyes. I have a certain style.

What I want, which is not possible to do in a physical store, is my personalized store which only focuses on my colors, my style, my body type, my size. This not only draws from data and collaborative filtering; this draws just as well from expert system knowledge, which is what we were doing.

Merging both data-based insights as well as heuristics – there is a certain body type that would like to wear certain kinds of clothes. A certain kind of clothes looks good on a certain body type. If you are a bigger person, the same style of clothes would not look good on you.

Those are expert system insights. They don’t change with data. They are written in stone. The question I have is, why aren’t these kinds of systems more prevalent? Today, I still see all the fashion companies that are online. It’s all collaborative filtering type of stuff. There is no expert system plus machine learning systems.

Ashutosh Garg: Can you do that? The answer is yes. When I was at Google, I was doing news personalization. The dimensions that we are talking about are so large. That is one thing.

Sramana Mitra: Let me underscore what you’re saying. You’re saying expert systems lack scalability.

Ashutosh Garg: That is correct. 

Sramana Mitra: With these kinds of insights, you’re saying, it’s not possible to bring in scalably into a machine learning algorithm.

Ashutosh Garg: I’ll give you one example. What we have to do is not think about expert systems versus statistical systems, but how both work together.

Sramana Mitra: That’s right. 

Ashutosh Garg: What we want to codify is that gender doesn’t matter when you are looking at a candidate. That should not be part of any feature, any algorithm, or any analysis.

Sramana Mitra: That’s not entirely true. There are jobs in which you might want to have a gender bias selection. I’m just playing devil’s advocate a bit. If you’re in women’s fashion, for instance, I think a large percentage are women.

Ashutosh Garg: You can also have a male merchandiser.

Sramana Mitra: Yes, you can have. People who have a feel for a product often better tend to be better merchandisers for that product. You can’t completely eliminate that rationale.

Ashutosh Garg: Is it because of your gender or what you have done and shown? For example, I’m color blind. You may say that more males are color-blind than females.

Sramana Mitra: I don’t think it’s the same thing. I’m talking about product knowledge. If you’re doing baby products, mothers who have handled babies have a better visceral understanding of the product. I have not raised a baby, so I don’t have a lot of understanding of a baby product. And I don’t have the patience.

Ashutosh Garg: I change diapers twice a day right now. 

Sramana Mitra: Maybe not gender, but somebody with that experience.

Ashutosh Garg: Exactly. That is the key thing. 

This segment is part 5 in the series : Thought Leaders in Artificial Intelligence: Ashutosh Garg, CEO of Eightfold.ai
1 2 3 4 5 6