13 Trends in IT

Question 1

What are the seven primary abilities by Louis Thurstone?

Answer 1

• verbal comprehension
• word fluency
• deductive reasoning
• spatial imagination
  retentiveness
• numeracy
• perceptual speed

Question 2

What is Artificial intelligence and what is its goal?

1.
2.
3.

Answer 2

• based on the idea of “man as a machine” - strongly controversial
• goal of strong AI: to create an intelligence that can think and solve problems like humans and that is characterised by a form of consciousness (debatable whether even possible)
• goal of weak AI: imitate intelligent behaviour without claiming consciousness or the like (already in everyday use)

Question 3

What is…

1. Symbolic AI?
2. Sub-symbolic AI?

Answer 3

1. approaches problem from “above”, considers logical reasoning as the basis; symbols and rules are used for this purpose (fitting for clear tasks like playing chess)
   Imitation of decision-making process of human expert (database + processing rules)
2. approaches problem from “below” and simulates groups of neurons - central model is connectionism (structure, model of neurons)

Question 4

When is symbolic AI used?

Answer 4

• favored in situations where the probem can clearly be defined and expressed by symbols and rules
• e.g. special programming languages, like prolog, legal decision making, and finance

Question 5

When in sub-symbolic AI used?

Answer 5

• favored in situations where the problem is more complex, requires large amounts of data, or involves learning from experience
• e.g. image recognition, natural language processing, robotics

Question 6

What are advantages and disadvantages of Symbolic AI?

1.
2.
3.
4.
5.
6.

Answer 6

+ can justify decision
- problems with uncertain knpwledge
- problems with exceptions
- knowledge acquisition is difficult (=Wissensaneignung)
- speech recognition is hard
- translation
- creativity!
- image recognition

Question 7

What are other approaches?

Answer 7

• machine learning: you DON`T set rules but let the AI learn by itself
• just provide data; AI tries to derive value

Question 8

What is the Chrun-Analysis?

Answer 8

• who will churn
• historicla data from telecom
• 21-dimansional dtaa set

Question 9

What is a neuron and what is the corresponding approach?

Answer 9

• nerve cell in biology - connected to each other by synapses
• many interconnected neurons = neural network
• neurons have two states: on and off (simplified) that depends on inputs from other neurons
• compatible with IT (on / off like digital logic, linking different neurons like input - process - output principle)
• neural networks can LEARN

Question 10

What does learning neural network mean?

1.
2.
3.
4.
5.

Answer 10

• you giver a set of input data and classify it
• network calibrates the weights so that it can eventually classify unknown input data itself
• e.g. neural netwrok is trained on cat pictures: each pixel of the picture forms signal, the net itself can recognise cats in oictures
• correct “training” of a net is important
• you dont need an absrtact model (unlike symbolic AI)

Question 11

Examples for neural networks in today medicine

Answer 11

• diagnosis of tissue images (analyizing X-ray images etc. - NN already better than humans and have independently identified further features that indicste diseases)
• medical research -> with help of deep learning a netwrok managed to identify the most promising molecules among thounsands in just two weeks

Question 12

Problems with Machine Learning

1.
2.
3.
4.

Answer 12

• quality of training data
• result is not comprehensible (e.g. AI rejects credit application, user does not know how the decision was made)
• problems with GDPR - right to review algorithmic decisions
• results often not questiones because “faith” in the AI