Vision Search
While everyone is over the moon about voice search, vision search gets no attention. Companies like Google, Pinterest and Snapchat have released products enabling searches with a photo. For example, don’t know what type of cute dog that is, take a photo and Google will tell you

Like a style you see on Pinterest? Take a photo and, viola, you can buy it:


It’s not definitive how many vision searches happen a month, with eMarketer guessing around 1% of total search volume. But, monthly Pinterest visual searches did increase 140%, from 250 million in Feb 2017 to 600 million in Feb 2018.

My prediction is the vision/voice conversation flips in the coming years. Voice shopping hasn't gone anywhere because the interface isn't that good. Vision just feels easier to use, more intuitive and gives you a better interface for buying. 

Data Traceback
Last week my wife got a call from a basketball team asking if she wanted to buy tickets for upcoming games. My wife is the last person a basketball team should be calling about tickets. Whoever is running their CRM system needs to be fired. But that got me thinking. 

If companies reach out to you, you should have the right to request how they got your information. For the overwhelming amount of airline credit card offers I get, I know where these come from; I am in their loyalty program. For the other 90% of mail, calls or emails that I have no idea how I got targeted, I should know why I was. Hopefully, this would discourage companies from misusing data and force them to think who they are targeting, instead of just mailing and praying. 

Elizabeth Warren, you are free to add this to your tech regulations. 

Language nuances
Two weeks ago we mentioned a tool Alibaba created that can produce thousands of lines of advertising copy. While some Chinese based companies probably use the tool to save money on copywriters, foreign companies should be the ideal user.

Let’s say you were a European retailer and wanted to expand distribution across China. Instead of taking existing ads and translating them into another language, you could use the Alibaba tool to create entirely new ad copy. This gives companies an instant "agency" in China and should lead to better results. If Alibaba has global ambitions, they should develop an ad generating tool for all markets so companies can scale across the globe.  

Google changes some bidding formats
An important thing to understand about machine learning; algorithms are very fragile. If you change what you are trying to predict, even if it's very slight, everything falls apart.

Take the recent move by Google to change how some of its display advertising is bid for. Google is moving from a second place to a first place auction, which means whoever bids the most, wins. Since algorithms are optimized to bid in a second place way, they need to be completely reprogrammed to bid in the new style. The bidding process will probably be a mess before the algorithms collect enough data and figure it out.

A great example of this is a computer playing the game "breakout." Researchers trained an algorithm to be a world class breakout player, but then changed the position of the paddle. As shown in this link, the computer algorithm completely fell apart and was unable to adapt to the new environment.

That should be a lesson: if computers ever try to take over the world, just dress like a computer and you’ll be fine.

Like this article? Have insights and analysis about everything machine learning/artificial intelligence delivered to you weekly, by signing up here