Welcome again to Episode 15.
Today, we will talk about a fascinating topic: Machine Learning for non-technical professionals.
This episode started last weekend when I attended a beautiful exhibition at the Houston Museum of Fine Arts called: Virtual Realities: The Art of M. C. Escher from the Michael S. Sachs Collection.
Art and Artificial Intelligence?
Yes, hold on a minute, and it will be clear. I promise this episode will be fun!
Typically, people think that Artificial Intelligence (AI) and Machine Learning (ML) are related to Humanoid Artifacts.
Probably, this is a consequence of expectations created by Science Fiction movies.
In reality, AI and ML are systems made from Hardware, Software, and Business Rules, capable of processing information based on general rules (not mandatory steps) and answering questions with non-anticipated responses.
In this context, building a compelling use case requires some critical steps, among others:
- Define a problem to solve.
- Collect sample data related to the use case.
- Create business rules to build context for the data.
- Define an algorithm to ingest the data in the artifact for training (Machine Learning)
- Connect the business and the AI artifact so the artifact is exposed to real-time data.
- Implement a valid response: we need to be capable of learning by doing
Where does Art connect with AI?
Initially, there was a trend to build massive databases to train the model. Often, IT/AI Architects struggle to create an adequate data sample. Recently, AI pioneers, such as Andrew Ng, are refocusing their agendas on building small smart-sized data-centric solutions.
What can Art teach us about Machine Learning Data Models?
Many times, Visual Art is about symbols and finding connections between them.
Similarly, Sample Data needs to have a proper context to be effective.
Typically, context is referred to as tags, metadata, patterns, cadence, and rhythm, among others: it is about connecting multiple elements into a bigger story (a Masterpiece).
“Order is repetition of units, Chaos is multiplicity without Rhythm” M.C. Escher.
M.C. Escher (1898-1974) is a unique visual artist. His approach to art related to the idea that any form or shape has a meaning, and by understanding the essence, we can transform any symbol into a new one.
Even though M.C. Esher was not math or computer literate, his vision of symbols (data) is a constant inspiration to Machine Learning and Artificial Intelligence experts.
Imagine that you use this masterpiece called “Plane Filled” to train an AI Artifact with the following rules:
- There is no empty space.
- All space is filled with animal forms.
- A symbol may be rotated.
Questions:
- How many symbols do you identify?
- Name the signs (animals) identified.
Simple? Take your time and let us know how many you see and maybe the name of some of them.
Question:
Q: Why do you think the system will learn on a continuous approach?
Tic-tac, tic-tac.
A: The Key is the statement, There is no empty space, so the System will analyze several times until all areas are associated with a symbol (animal). Typically, there are other variables involved (number of iterations).
Moving on! We can have more complex scenarios, such as images changing shapes and transforming into other symbols.
A sample of an AI use-case where guessing is needed
Imagine using this masterpiece called “Sky and Water I” (M. C. Escher) to train an AI artifact to recognize symbols.
Note that some symbols are not entirely defined but suggested (blurred).
How many symbols will the AI artifact recognize?
How many can you?
How would you set a threshold (level of detail) to define either a fish or a bird?
Did you know that pathologists use the same approach to recognize cancer cells?
They look at tissue morphology to identify the cancer-type cell. Later they look for the level of differentiation of the cell elements to determine the level of aggressiveness: A cancel cell with little cell differentiation is considered highly aggressive because it reproduces itself at high speed.
An AI use case of Computer Vision matching Human Intelligence
Even more complex scenarios exist, such as the “Bond of Union” masterpiece.
Imagine we train an AI artifact to do face recognition using sample data such as this one, and we locate the device in public places, such as airports or metro stations. We ask the AI artifact to look at people and identify persons of interest.
See? We can have these devices working 24×7, protecting us.
Since AI is at its initial stages, it can go as far as our imagination and creativity dictate.
To have an effective AI implementation, three guidelines:
- The business rules, not the technology.
- Avoid using AI as a siloed solution. The real benefit is connecting the solution with the business.
- Do incremental steps. Having Big Bang implementation is many times unfeasible. Having a good enough solution many times is a good start. Additionally, having a learning organization approach, including a Continuous Delivery and continuous improvement approach, is the right approach.
OK, Jose, I got you: Testing, testing, testing?
Yeap, 😊
A final comment refers to the masterpiece I included at the beginning of the episode. “Metamorphosis II”
M.C. Escher proposed that any form has a meaning. You can modify any symbol into a new one if you know the meaning, starting with grey lines, going into geometrical lines, color forms, animal forms, living places, and ending with grey lines.
He expressed this concept in this masterpiece, made of one continuous segment (not broken into three as I published) where M.C Escher shows a continuum of Life, humankind, and Society.
- Jose, I got you. What else do you want to show us about the relationship between Art and Artificial Intelligence?
There is a recent trend in Artificial Intelligence that is teaching an Artificial Intelligence System to create visual Art based on concepts, ideas, or randomly.
An excellent example of this is a project called DALL-E-2 AI Project; you can input a phrase, and the algorithm will create an image representing your idea.
Some images are really good!
Good enough?
Please feel free to share your opinion with the audience or privately.
All opinions are welcome and encouraged.
- Do you like the episode? Please click on Like
- Do you want to be informed of new Episodes, please click on Subscribe.