Answers To What We’ve Learnt: Part 5.5
Answers in Bold
More than 1 right answer is possible. Answers in the next blog.
1.Traditional software is developed :
a) using human prescribed rules and logic.
b) to handle previously unseen input.
c) to produce output as predictions.
d) sort of like the do’s and don’ts of major religions and legal frameworks.
e) is already perceived as intelligent by people. [The Media Equation]
2. Artificial intelligence :
a) synthesizes experience into machine intuition.
b) recognizes previously unseen input.
c) handles new situations like HMS secret service agent 007.
d) use analysis to learn.
3. Neural networks :
a) are the latest and greatest religious cult of humanity.
b) should be worshipped.
c) has dethroned all ancient gods.
d) are social media networks for crazy people.
e) reverse engineer the systems they mimic.
f) can approximate the governing function of any system using the inputs and outputs
g) invalidate the Universal Approximation Theorem.
4.We recognize our immediate family members instantly because:
a) we are connected with empathetic resonance.
b) they are after all blood.
c) we developed familiarity through many years of living together.
d) their appearance is part of our intuition, synthesized over time.
e) we recollect past memories and identify which ones match the current view of our family member.
f) we keep a list of mental notes of their physical traits to match.
5. The patterns used by AI to make predictions are inexplicable because:
a) nobody cares how it gets the predictions right, just that it does. [there is some truth to this]
b) of similar reasons to why human intuition is not explainable in words.
c) machine patterns are synthesized, while explanations are analyzed.
d) the technocrats who built it want job security.
e) machines don’t want humans to find out the reasons behind their madness.
6. Neo, the product manager, was unable to address the the handwriting recognition challenge because he:
a) was an impostor.
b) lost confidence after seeing Mr. Smith’s deadpan.
c) was limited by traditional software.
d) did no realize that there were too many exceptions for human designed rules to capture.
e) he didn’t play the internal political game well enough.
f) could have done a better job making engineering the scapegoat.
7.Machine learning applies patterns that are:
a) that are precisely described by engineers.
b) extracted from a training set (experience).
c) that are the ‘essence’ of each desired classification (like the digit ‘7’).
d) analogous to how a child learns to play catch.
e) after training on a large dataset like 60, 000 images of handwritten digits.
8.Product managers should:
a) own the training data which represents their domain of expertise.
b) support data scientists in keeping training data separate from the unseen data that AI accuracy is measured against.
c) may have less responsibility towards writing PRDs that have rules of the domain, because rules and patterns are now the domain of machines.
9.In the equation, y=2x — 5, the gist of infinity:
a) is not a mathematical concept, just a term used for explanation.
b) consists of 2 numbers that can be used to find all other numbers on this line.
c) could be discovered by AI using a training set consisting of (x, y) that form the gray area of real world data.
d) are called weights and form the model.
e) has meaning it cannot explain to humans, like an artificial subconscious.
f) is orange juice.
g) can classify previously unseen data points based on which side of the line they fall.
10.The equation, f(x) = wx + b:
a) is linear.
b) is irrelevant to neural networks, which are non-linear.
c) represents the basic concept behind orders of magnitude more complexity in machine learning.
d) is comprised of (w, b) which the AI learns, and (x, f(x)) which the data scientists provide.
e) expresses the model (machine intuition) in the weights and bias.
f) expresses the experience in the input and output (x, f(x)).
11. Real world models may consist of:
a) millions of weights.
b) a function that represents a N-dimensional hyper-surface that classifies the data.
c) sequence of linear sub-functions whose output becomes the input to non-linear sub-functions, whose output, in turn, becomes input to linear sub-functions, and so on.
d) the power towards greater predictability about the gray area of real world data.
e) a bunch of snowboarders and skiers strutting their stuff on the catwalk of mountain slopes.
12. Non-linearity makes neural networks far more powerful because:
a) a straight line is set in its ways and cannot address the vagaries of real world data.
b) it gives greater flexibility to the neural network to carve a path that more accurately separates real world data.
c) it enables higher dimensional pattern recognition.
13. Neural networks use supervised learning, which means that during training/ learning:
a) they learn from making mistakes.
b) they have the answer key.
c) at each learning step, a data scientist needs to monitor the training.
d) at every step, another AI plays the supervisor role to the learning AI.
e) without a supervisor, general laziness prevails.
f) the input consists of (x,y), where the presence of y makes the input “labeled”.
14. In the magical mystery tour, the scouts represent:
a) neural networks.
b) perceptrons.
c) the input to AI.
d) the people who own and work with the input, i.e. the data scientists and product managers.
e) a bunch of kids of the Oompa Kicchu, asked to collect data about beasts from the forest.
15. Unlike distinguishing features of beasts, snapshots are:
a) structured data.
b) unstructured data.
16.Determining structured data can be more of a:
a) art than a science.
b) science than an art.
17. The Council of Perceptrons was limited in its ability to distinguish threats because of their:
a) linear mindset.
b) non-linear predictiveness.
c) old age.
18. The early Tower of Wisdom was a real neural network, but it could not handle snapshots because:
a) the Oompa Kicchu could not figure out how to get more than a couple of councils to collaborate.
b) snapshots have too much data for just a few floors of councils to wrap their heads around.
c) it turned out that human-engineered structured data led to better predictive accuracy than unstructured data.