What do you think of machine learning?

IF you A have the data.. and B the ability to process the data.. oh man oh man.

The key to AI or machine learning is working things out backwards.. understand what the outcome you want is, and then identify the variables that indicate that outcome.

A personal example I want to sell more Satellite TV and internet installations. What are my variables? Again for my business personally I know that there are "areas" that have cable tv and internet.. there are areas that do not. there are also areas that have phone based DSL, and those that don't. I can look at a utility pole and determine if either or none are available. Over the years I have created a map - literally 10,000 square miles of map with the services available or not are presented. This alone is gold. - BUT this is not machine learning.

So I want to know what areas are most likely to buy TV service, which are most likely to buy Internet service and which are most likely to buy both. Enter USPS's EDDM site. I can insert each and every zipcode+4 into a database. I know specifically how many address' are in each. I then insert my years worth of client address'. What I end up with is a percentage of population per zipcode+4 that has one or the other or both services.

Obviously the areas that have Cable TV and Internet service are less inclined to buy satellite services, and those that do not have such services are more likely. It becomes apparently clear where I need to spend more time and resources.

Overlay the zip+4 location onto my physical data ( the service end point map ) and I now can identify not only where to send direct mail, but where the best location options for billboards and yard signs, what newspapers to target, and what radio stations / tv stations might best suit my targeting needs.

In the realm of AI / Machine learning that is a pretty small data set - but a very effective one.

Then there is BIG data - or for me at least. I have access to some 4000 websites data. a pretty sizable number. about 1000 of them are local market sites and the other 3000 are commerce sites. The data that can be pulled here is mind blowing.

By percentage I know what hours of the day produce the most traffic ( the most conversions ). With this I can control ad spend. 3 to 5 hours of ad clicks vs 24 hours. Optimizing for the best 3 to 5 hours of the day, and minimizing expense on the remaining 19 to 21 hours that pretty much suck.

Reading articles like this: https://www.shopify.com/blog/5213662...h-social-media is a bit misleading... this would NOT be commerce driven data. Lunch time is optimal ( 11am to 2pm ), followed by a window at 10pm and then 2am.

within the 3 time frames; lunchtime 10pm and 2am, there can be some correlations made as to what can be displayed when for best performance. The easiest way to explain it.. the later it gets the deeper into the seven deadly sins you can get.

The BIGGEST piece of information you can pull from that article: "48% of the US population runs of Eastern Standard Time (EST), 33% on Central Time (CST) and 14% on Pacific Time (PST)." This in itself is BIG DATA. Think for a moment when we were all watching the US Presidential election.. ALL EYES are on the west coast.. I mean California is the big dog right? Looking at the numbers above... Yawn... California who?

aside from understanding buying habits based on time, I also look at buying habits based on location. and you are asking how do you do that? well IP address of course. ( here is a pretty cool set of data you can buy https://db-ip.com/db/ ** there are also free methods to obtaining this, just not in a complete list as this is ) Now I can take a piece of known data, a users IP address, and can inject things like bought or looking.. I can insert time of visit, specifically what they were looking at, and know specifically where the end user is located.

Can you imagine what social targeting looks like with this data? "retargeting" is one thing, pushing a geo-demographic with a like campaign, and following up with a sales campaign is something totally different.

So what does this look like? lets say you find a geo region that has 50 people that have purchased recently. you can specifically target that area with say facebook with content that is designed to create followers and likes. Over the course of a few days you transition to selling. When you click on a facebook page, it tells you that there are "friends" that like/follow that page. You have just created "Social Influence". Social influence is the gateway to making a sale. If Martha is following them, then I should, and buy something...

Using the Target example.. they have far greater data pools than I will ever imagine... but even for the small guy, data analysis and data learning can be accomplished.

You may recall how the supercomputer, Watson, outperformed humans on TV quizzes?

What most people don't know is that most of the health records are being uploaded into Watson.

At the same time, a handheld device with an api into Watson's medical data will mean the device will be able to diagnose health problems more accurately and cheaper than doctors.

The device is now available.

We as entrepreneurs can tap into Watson for whatever we want,
for free to start while we get a prototype app up and running.

Best,
Ewen

The best chess players have been beaten by ai machines

The game GO has a lot more possible moves...
ai machines have beaten top players.

Top poker players have been now beaten by ai.

That's now and in the past capabilities.

The machines performance don't stop growing.

Yet human brains haven't had an upgrade for millions of years...
now that's scary!

Best,
Ewen

Baxter robots don't need programming.

They take about 20 minutes of training by moving its arms and it will adapt and self-learn.

Under $20,000 at the moment.

Robotics is another technology where the gap between performance and price is continually widening at a rapid rate.

Best,
Ewen

Machine learning is control points (+/-).