Der ExplAInator

Einfache Erklärungen - auf Englisch -  zur künstlichen Intelligenz

Feel like a deep dive?

  • Our colleagues from Corporate technology created  interactive ExPlAInables for chosen AI nuggets.
  • Thanks to Daniela Oelke for sharing!
  • Try yourself!

 

Defining AI

  1. to correctly interpret external data (e.g. when a camera provides data - know the difference between a person and a shadow)
  2. to learn from such data (e.g. the AI has to learn that it is fine to run over shadows but not over people)
  3. to use those learnings to achieve specific goals and tasks through flexible adaptation. (e.g.  steer a vehicle)

 

  • Key takeaway: AI is not about creating artificial humans but comprises  certain ways HOW computers handle data.  Some AI systems are really powerful some are not at all. 

Industrial knowledge graphs

  • Without being able to interpret, data are just characters with no meaning.
  • Industrial knowledge graphs give AI something like background knowledge. (See a video about the importance of background knowledge)
  • Thus, they enable AI to figure out the context of data and further relevant information.
  • They can be represented as (mind-map-like) graphs that show how things are interconnected. 

Feel like a deep dive?

  • Our colleagues from Corporate technology created  interactive ExPlAInables for chosen AI nuggets.
  • Thanks to Daniela Oelke for sharing!
  • Try yourself!

 

Reinforcement learning

  • A technical term for certain AI-learning algorithms.
  • The AI figures out by itself (in many, many iterations) HOW to solve a problem.
  • In the beginning it starts with random actions.
  • The AI and receives (positive or negative) feedback depending on how successful its actions were.
  • It adapts accordingly and gradually gets better and better
  • Try yourself: Watch an AI learn how to play snakes. Take your time, play around with the parameters and experience the differences yourself!

Generative Design

  • An AI-learning approach that works like the evolution. i.e. the fittest will survive
  • The AI starts learning in a random configuration and generates a couple of mutations of that configuration. All of these mutations have to carry out a certain task.
    The most successful ones “survive “ to the next generation when new mutations are generated
  • Normally It takes at least hundreds of generations to receive satisfying results
  • Try yourself: watch robots learning to walk – it looks very drunken in the beginning, but 1000 generations later …

Neural Network

Why we need AI?

  • AI can handle tons of data and needs tons of data to be trained. Because of that it is so useful for IoT
  • Key takeaway: AI is not about replacing humans but especially useful to handle masses of data.

History

  • Basic concepts were invented in the 1950s
  • AI was almost forgotten during the 1970s and 1980s - there were just not the right use cases (computers too slow, not enough data to train AI).
  • The present AI hype is caused by digitalization, masses of data and IoT.
  • When one needs to evaluate masses of data, AI is a successful technology.

Sometimes you need to know why

  • Normally AI is like a black box –  it gives results but does not tell you why.
  • For important decisions this is just not acceptable. 
  • Explainable AI tells you WHY it made a decision.
  • Also see European guidelines for trustworthy AI

 Create Digital Companions

  • Digital Companions are computer systems that are designed to fulfill the needs of humans – to  help them to do their work efficiently, comfortably and with minimal errors.
  • Video thoughts on Digital Companions
  • Psychology and scientific results on how people learn and work are an highly important corner stone for the development of digital companions.

Recommended long reads

Be aware of Biases

  • Humans are prejudiced against a lot of things – and very often not even aware of it.
  • Watch a video about unconscious bias.
  • AI is just as biased and discriminating as humans are if it was trained by data that – implicitly – contained such biases.
  • Nobody needs badly educated AI becoming an artificial racist!
  • Therefore: AI needs unbiased data for training!

Feel like a deep dive?

  • Our colleagues from Corporate technology created  interactive ExPlAInables for chosen AI nuggets.
  • Thanks to Daniela Oelke for sharing!
  • Try yourself!

 

2020-06

Aenne Barnard, Rebecca Johnson

Subscribe to our Newsletter

Stay up to date at all times: everything you need to know about electrification, automation, and digitalization.