RDL VL 14 AI

On_Computable_Numbers_1936-1.pdf
lntelligent_Machinery_1948-1.pdf

• In many applications we can not predict the future completely.
  Therefore:

Claims

1. Robots need to be very complex.
2. The Robots need to live in an complex unpredictable enviroment.
3. The Robots have to learn.
4. The Robot needs to run over a very long amount of time.
5. The learning will never end.
   • In contray to how we normally build systems.
     • Error/Cost Funktions

In his time robots where not advanced enought to try to slove the Claims:

• Therefore we should start sovling these Problems:
  • PLaying Games (Chess, Go, …)
  • Natrual Language
  • Translation
  • Cryptology
  • Mathematics (Logic, Reasoning, …)

There many Areas where we have not sloved yet:

• Two Arms

They wanted very stiff and acurate systems.

• The Robot did not need to adapt at all.

• Just a machine

• Function tht maps Perception to Actions:
  $[f:S⟶A]$
• Optimizing the long term Reward:
  $:E[r(t)+Y^{1}r(t+1)+Y^{2}r(t+2)+Y^{3}r(t+3)+...]=Q(s,a)$

Robot Drivers

⇒ Touch is not much used!