Michael Sherman, whose contributions to spine innovation are, literally, the subject of history books, once told me: “Surgeons are great at articulating problems, engineers at finding solutions.”
Could the explosion of various artificial intelligence (AI) algorithms change that dynamic—towards the surgeon inventor?
It just might.
How Artificial Intelligence Fundamentally Changes Design Engineering
We know AI powered software has the capacity to learn—which is to say, incorporate a feedback loop which causes the software to change its coding on the fly.
Also, that feedback process is powered by probabilities—as calculated to a defined outcome.
But, in order to “learn,” these systems need to be able to “read.”
In order to “read,” AI must convert words to numerical values that also capture nuance, definition and context.
This opens the door to AI design engineering.
Word2Vec
There are three major categories of AI algorithms, about a dozen sub-categories and, as of this article, 14,300 artificial intelligence companies bringing to market various permutations of these type of AI algorithms:
- Supervised learning
- Decision tree
- Random Forest
- Support Vector Machines
- Naïve Bayes
- Linear regression
- Logistic regression
- Unsupervised learning
- Clustering
- K-means clustering
- Gaussian mixture model
- K-nearest neighbor algorithm
- Neural network
- Reinforced learning
- Value-based
- Policy-based
- Model-based
Word2Vec fits into that middle category—it is an unsupervised learning AI program.
It’s also a simple, 2-level neural networking algorithm (2e in the preceding bullet points) which uses logistical regression techniques to vectorize words.
Created by Google
Word2Vec was developed and trained by Google on about 100 billion words. It calculates the vector between words.
Vectors are numbers which give direction and magnitude. Like an arrow. It numerically represents where the arrow is going and how fast or strong it flies.
Word2Vec calculates how words connect with other words—turning words into vectors.
Word2Vec is a natural language processing algorithm. It’s what is known as an advanced Bag of Words model.
Once trained, Word2Vec can detect synonymous words or suggest additional words for partial sentences. Because it represents each word with numerical vectors, it is able to capture the semantic and syntactic qualities of words. It also takes this process to higher levels—capturing the full range of complex geometries of vectors—non-linear, trigonometric vectors, for example.
Word2Vec, by the way, is a comparatively shallow, 2-level neural network that’s trained to reconstruct linguistic contexts of words. Another advanced Bag of Words algorithm, BERT, competes in some ways with Word2Vec.
CPT and diagnostic codes have word structures. They can be turned into vectors. Proteins are being turned into vectors. Chemicals into vectors. Genes into vectors.
Doing AI with vectors gives scientists and engineers the ability to play with novel combinations of anything with a structure.
Consider that connecting amino acids to proteins earned R. Bruce Merrifield, Professor at Rockefeller University a Nobel Prize in 1984!
This is a qualitative leap in engineering productivity. Not just more, but higher quality ideas, faster.
One Example of Engineering With Vector AI
In the near future, say five to eight years from now, an automobile engineer has a rough idea of a new design—a novel kind of four-door, four-wheel drive, SUV. He makes a rough sketch.
Shows it to his computer. The computer “looks” at the sketch. Almost instantaneously, because it’s packed with Nvidia gaming chips, 😊, it generates 200 complete, 3D CAD models for that car concept.
The engineer sifts through these AI generated choices, selects one and asks his computer to create 10 iterations of that specific design. Keep in mind, the AI algorithm incorporates a randomizing function so that it presents that engineer with unexpected ideas.
That repeats until the engineer finds a design he likes. He tweaks the final design himself.
Now substitute a surgeon inventor for that engineer and implants/instruments for the automobile.
Add a 3D printer.
I think surgeon inventors will go insane when they have this.
Imagine the explosion of innovation when these preternaturally gifted problem-definers access this level of intelligent design engineering and fabrication.
Productivity Changes at Manufacturers
We are on the cusp of being able to literally take information of one kind and generate it into information of another kind—because we can vector words.
Surgeon inventors send their sketches and problem descriptions in. Design engineers refine them, and the system creates prototypes instantly.
Manufacturer operating profit margins rise as productivity increases.
Companies will have more great ideas than they can afford to fund.
But, a two-year clinical study is still a two-year clinical study.
The clinical study bottle neck will become the next critical problem.
So, the next growth industry will be clinical studies?
Welcome to a New Age.
