What makes AI development special?

 💥 Nothing happens on the surface...then boom, problem solved... no human can compete. 

  
I'm all about practical applications of AI and 2022 I got a few "wow" moments and I would like to share one with you now. 
  

👨‍🎓 As a student at Lund University, I remember one of my professors telling a story about the early days of determining the 3D structures of proteins using X-ray crystallography. The output data from the X-Ray machines was printed on several large (square meter) transparent sheets that was hung from the roof, in front of each other in a large room. By looking through all these sheets, it was possible to "see" the 3D structure. As the story goes, there were one guy who had eyes like Marty Feldman which made him exceptionally good at seeing the 3D structure. Complicated stuff. However, it turns out to be worth the effort. 
  
Proteins are the building blocks of all living organisms, and their 3D structure plays a crucial role in how they work. Determining the structure of a protein can provide important insights into how it functions and how it can be used to e.g., develop treatment for diseases. Proteins can also be described with a 1D amino acid sequence, this is quite easy to determine, but not as useful as the 3D structure. Considering that there are more than 100 million proteins known to science, let's build X-ray machines like crazy...or?! 
  
In September 2010 DeepMind was founded, for the wider audience (such as me) the company is most known for beating the world champion in the board game Go. 
  
💥 In the early days they wanted to beat the computer game Pong using their AI algorithms. It took 6 months for the system to score its first point... then it got one point... then it won a game...then it started winning 21-0 and could not be beaten by a human anymore. Boom. 
  
Going back to protein folding (1D to 3D). DeepMind wanted to solve this problem with AlphaFold and so they did. Training their AI on the known 3D protein structures (shout out to experimentalists) resulted in a system that can go from a 1D to 3D in seconds, this with good enough accuracy for practical applications. 
  
Alphafold computed the 3D structures of 20 000 proteins overnight, and in 2022 over a million protein 3D structures was released. What earlier took thousands of man years is now done in less than 24 hours and is accessible for us with normal eyes. Nothing... then boom, the problem is solved, and no human can compete at that specific task. 
  
We will see more and more of these "explosions", I'm curious to see how it will change the (project) leader role, which of my tasks will I be able to hand an AI-assistant with faster and/or better results? I keep my fingers crossed for maintaining the project budget up to date or document management. 
 
What do you think? Which parts of your profession will be disrupted? 
  
*The DeepMind info is from an interview with Demis Hassabis.