Everyone talks about AI this days. What steps do we do to make our FileMaker solutions become a AI
interesting thought - if you look at DevonThink that kind of AI would be great to have in container field data.
"Artificial Intelligence" sounds cool, but as soon as we start talking about actually implementing it, we quickly realize that AI means different things to different people, or that AI isn't well-enough defined to know when you've achieved it or not. Can you clarify what scope you're thinking of?
(The earlier items in the list tend to be pre-requisites for tackling the later items.)
I have seen the work that Clickworks did to get tags out of pictures
ClickWorks | FileMaker 16 and Artificial Intelligenice
Luke Rochester demostrated to me how to create economical readable reports with AI
I know that a AI could be almost anything, but looking back to what we had at DevCon this year in DevCon where each day finished of with some companies showing the audience what you are able to do with FileMaker, I think a session demonstrating some of the things that we could do with a AI from FileMaker is absolutely worth for the FM World to see
So you have a broader, rather than a narrower, interpretation of "AI" in mind. I like that. Luke's demo is doing pretty simple statistics. The ClickWorks demo is doing image classification, way at the opposite end of the spectrum.
I think the particular term "AI" is so vague that it's useless to anyone actually wanting to implement something falling under that umbrella, but I suppose we're stuck with it for the marketing value.
You are correct, AI could mean so many different things. But what I am after is to show that we can do very cool stuff using AI. Then I am always up for ideas on what could be fun to lock at
FM AI would generate the TOG by looking at the data you throw at it. The more data you through at it the more precise your structure becomes (something like DevonThink does // repeat ) ..
For a FileMaker Developer to have any decent control over the machine learning models which are necessary for AI, one must connect to a python environment. Think of AI in two parts, machine learning of data (finding patterns within datasets), and AI as the processes or scripts that respond or take action after reviewing predicted results. In Python, you can load many Machine Learning Algorithms from leading universities around the world; however, your FileMaker Dataset must be more structured to allow for machine learning process to commence such as changing the way in which you present categorical data. Machine Learning doesn't consume the text data type, only numbers. Once you are able to create a reliable learned model, you can pass the results into FileMaker and can create scripts that do certain things in response to the data results. Unfortunately, there are many different ML Models to use and requires the data analyst or developer to try many different approaches to getting the desired result of the model.
You could use IBM Watson's API along with FileMaker's Insert from URL / cURL script to interact with IBM's cloud version of Watson (The infamous jeopardy computer that beat Ken Jennings) However, your control of IBM's data models is limited. You can do simple things like detect a sentiment or predicted the next words while someone is typing and so on.
Here is a link to Watson Sentiments : Natural Language Understanding Demo
You are wrong. There are several FileMaker Developers that already have done amazing AI integrations. For example, have a look at this
ClickWorks | FileMaker 16 and Artificial Intelligenice
I personally have create a BOT that are starting to be very smart and already works like a FAQ for our clients
I guess everyone's interpretation of AI is different. I would call that Demo file Image Recognition, Facial Detection, OCR, and so on.
When I think of AI, I see more of learning and making decisions, not just a classifier that cross references a database of know photos.
It also appears, that you are using an API key to web service that is doing hosting the classifier model for you.
Google Images has had that API available for many years now and with out controlling some of the finer details of the
classifier model your image recognition will be mediocre at best.
My initial claim is you will have to use a web service that uses some other technology to setup your model like AWS, IBM, and others. FileMaker will not be able to establish a model on its own or run one. Technically FileMaker is don't nothing in terms of AI or ML. Just because you can load google.com into a web viewer doesn't mean you can say FileMaker can index and search the internet. Technically google.com and their servers are doing the work. In python, you can actually code your "own" machine learning models and build the model files. Which means you have two options use existing online APIs like you did with AWS (I personally found their image recognition model not that great) or build your own that caters to your business application or business data. Having the ability to create your own model will make the ML model significantly more reliable.
Thank you for the demo on using a API with FileMaker.
Like you say, there are several different meanings for the word AI. That is way I think that it would be such a great subject to talk about at next years DevCon. Showing of what we can do from FileMaker and at the same time give some notice of one of the things that are starting to become a very big thing in many areas for the near future
The main reason there aren't any machine learning libraries for FileMaker is that no one has hired a FileMaker developer to make them. If we chose a programming environment to start from scratch with machine learning, Python is not an obvious natural fit, and has several factors working against it. Python is only good for machine learning because the folks built stuff for it anyway. FileMaker has some important limitations, but there are plenty of machine learning algorithms that could work fine. Stream algorithms are especially promising.
Yes, developers have written lots of "stuff" for python. Most machine learning models and data scientists use python. Trying to get robust development for the Machine Learning world to switch to FileMaker Pro because a C++ developer made a somewhat useful plugin that does 1 or 2 models, probably won't ever happen. The FileMaker Development community is very small compared to other frameworks.
You may want to let Apple know that python has some negatives and that its not the place to start.
The development workflow for the CoreML Framework that Apple has released in iOS 11 requires a model file known as a .pkl file. This file can be created by using a very popular machine learning library called http://scikit-learn.org/stable/ "SciKit"
The scientific community uses this kit to create their models to analyze test results from experiments and so on.
Once you create your machine learning model with SciKit you can copy the file into your apple xcode project to run with CoreML Apple's new Machine Learning framework.
Core ML | Apple Developer Documentation
Lastly, you would not be able to start from scratch a whole machine learning framework and library. The current libraries and models are the product of 100s of university and research institutions working together for years and years to create the mathematics and code libraries that we can use. But, more importantly FileMakers Calculation Engine would never be able to keep up with processing these models. FileMaker struggles trying to do too much simple aggregating of millions of records. We have 20 million records of revenue transactions that we try to look at briefly at times to answer a question really quickly for an exec, but FileMaker either locks up or the summation takes forever. We ultimate have to move the data from this transactional system to a data warehouse (ultimately MSSQL Cubes). Machine learning requires lots of processors/cores to process millions of permutations of data. (with what ever algorithm you want but you need to be very informed on each algorithm to even know if its a good fit for your dataset)
Rattling of random names of different algorithms is only one parameter in the entire process of developing an useable model file. Trying to make a product into something it was never designed to be will lead you to grief over time. FileMaker is not by any means an advanced analytics engine. Its for online transaction processing for smaller datasets. Its target market is small to medium sized businesses, these business may also never even produce enough data to process into a machine learning algorithm to make the model useful. These businesses may not be large enough to warrant hiring a 150-180K per year data scientist or even a great statistician.
In think in the end everyone would concluded that whatever you did do with a FileMaker DB in terms of ML or AI wouldn't be that impactful to a small / medium sized business for the reasons mentioned above.
If you are truly interested in learning more about machine learning, may I suggest that you watch this entry-level video.
If you use MBS FileMaker Plugin, you can use the CoreML functions on iOS 11 and macOS 10.13 to use neural networks.
For example you can classify images by content.
Current libraries are indeed the result of a lot of work from a lot of smart people — work that can inform the design and implementation of new libraries, sparing a lot of the trial-and-error from the last several years. You're right that some machine learning algorithms are very computationally expensive. I don't see a practical FileMaker-native implementation of neural networks coming any time soon. You're wrong that all machine learning algorithms are so expensive.
Machine learning research isn't going out of it's way to make algorithms run slower and require more data to get to useful results—exactly the opposite, especially for applications in edge computing, embedded computing, and distributed sensor networks, for example. There are good results from that effort, and there's no reason FileMaker can't handle those same techniques, e.g., VFDTs. Machine learning aside, there are a lot of decades-old sophisticated statistics techniques that FileMaker could handle fine.
Retrieving data ...