MskProdComputing
Back to the topic

AI · 1.1

Thinking Machines: AI from 1843 to 2011

Section A: 10 multiple choice questions, 10 marks. Section B: 3 written questions, 13 marks. Total 23 marks.

Name

Class

Date

Mark  / 23

Section A

Circle the letter of the correct answer. One mark each.

  1. 1Who wrote in 1843 that the Analytical Engine 'has no pretensions whatever to originate anything'?1
    1. AAlan Turing
    2. BCharles Babbage
    3. CAda Lovelace
    4. DClaude Shannon
  2. 2What did Turing's 1950 imitation game test?1
    1. AWhether a machine could see
    2. BWhether a judge could tell a machine from a human in typed conversation
    3. CWhether a machine could beat a human at chess
    4. DWhether a machine could learn from examples
  3. 3Why is 1956 called the birth of AI?1
    1. AThe Dartmouth workshop gave the field its name and its agenda
    2. BA computer beat the world chess champion for the first time
    3. CThe first electronic computer was switched on that year
    4. DRosenblatt's perceptron first learned to recognise shapes
  4. 4What did Minsky and Papert show in 1969?1
    1. AA single-layer perceptron cannot learn functions such as XOR
    2. BExpert systems are better than people
    3. CComputers can never think
    4. DBackpropagation works
  5. 5What was ELIZA?1
    1. AA chess computer that searched millions of positions
    2. BA factory robot arm that could sort objects by colour
    3. CThe first neural network able to learn from examples
    4. DA 1960s program that held conversations by matching patterns
  6. 6What is an AI winter?1
    1. AA period when governments banned all research into AI
    2. BA yearly contest where AI programs compete on hard problems
    3. CA period when funding and interest collapse after overpromising
    4. DA slowdown caused by a worldwide shortage of computer chips
  7. 7How did an expert system such as MYCIN work?1
    1. AIt copied a human brain
    2. BIt applied hundreds of IF-THEN rules written by experts
    3. CIt learned from millions of examples
    4. DIt searched every possibility
  8. 8Why was 1986 important for neural networks?1
    1. AThe phrase artificial intelligence was used for the first time
    2. BBackpropagation was shown to train networks with hidden layers
    3. CMcCulloch and Pitts published the first model of a neuron
    4. DA computer beat the world chess champion for the first time
  9. 9How did Deep Blue beat Garry Kasparov in 1997?1
    1. ABy guessing
    2. BBy learning from millions of its own games
    3. CBy using a neural network
    4. DBy searching about 200 million positions a second using expert rules
  10. 10Why did ImageNet (2009) matter?1
    1. AIt gave researchers over 14 million labelled images to learn from
    2. BIt was the first search engine that could find photos
    3. CIt introduced the first camera able to recognise faces
    4. DIt proved neural networks could never classify images

Section B

Answer in the space provided. The number of marks tells you how many separate points to make.

  1. 1ExplainExplain why the period from about 1974 to 1980 is called an AI winter.3
  2. 2CompareCompare symbolic AI, such as an expert system, with machine learning.4
  3. 3DescribeDescribe three milestones between 1986 and 2011 that made later deep learning possible.6

Mark scheme

Section A

1C2B3A4A5D6C7B8B9D10A

Section B

1 Explain why the period from about 1974 to 1980 is called an AI winter. 3

One mark for each of the following, to a maximum of 3.

  • Funding and interest in AI fell sharply
  • Because AI had failed to meet its promises
  • A specific cause, such as the Lighthill Report, Perceptrons, or computing limits

Model answer. It is called an AI winter because funding and interest in the field fell sharply after the confident predictions of the 1950s and 1960s were not met. Programs that worked on small toy problems failed on real ones, because the number of possibilities exploded and computers were far too slow. In the UK the 1973 Lighthill Report told the government AI had not delivered, and funding was cut, while Minsky and Papert's 1969 book Perceptrons had already shown the limits of single-layer neural networks.

2 Compare symbolic AI, such as an expert system, with machine learning. 4

One mark for each of the following, to a maximum of 4.

  • Symbolic AI uses rules written by people
  • Machine learning learns patterns or rules from data
  • Rules are brittle or expensive to write and maintain
  • Learning needs lots of data, or is harder to explain

Model answer. In symbolic AI, such as an expert system, people write the knowledge down as explicit IF-THEN rules, so its decisions are easy to trace. In machine learning, the program works the rules out for itself from examples, so nobody has to write them. Rules are expensive to write, break when the world changes and become inconsistent at large scale, which is why the expert systems boom ended in the second AI winter. Machine learning avoids that, but it needs large amounts of good data and its decisions can be much harder to explain.

3 Describe three milestones between 1986 and 2011 that made later deep learning possible. 6

One mark for each of the following, to a maximum of 6.

  • Backpropagation in 1986
  • What it made possible
  • Convolutional networks reading handwriting, around 1989 or 1998
  • What it showed
  • ImageNet in 2009, or LSTM in 1997, or deep belief nets in 2006
  • Why it mattered

Model answer. In 1986 Rumelhart, Hinton and Williams showed that backpropagation could train neural networks with hidden layers, solving the problem Minsky and Papert had exposed and making multi-layer networks practical. In 1989 Yann LeCun trained a convolutional neural network to read handwritten postcodes, showing that networks could solve real vision problems; its descendants were reading cheques by the late 1990s. In 2009 ImageNet labelled over 14 million images, giving deep networks the huge quantity of data they need, which made the 2012 breakthrough possible.