Question 1. (2 pts.) Match each AI problem description to the class of technique best suited to the problem by entering the letter corresponding to the best technique into the blank. You will not use a technique more than once. You do not need to solve the problems.
(i.) A robot must navigate a collapsed mine looking for survivors. If it falls down a shaft it will be destroyed. Air breezes are detectable next to most shafts, but breezes aren’t always detected even when they are present. The rough terrain means that the robot’s movements are unreliable.
(ii.) The traffic on I-85 is unobservable from my office, but I would like to infer the probability of a traffic jam by observing the amount of honking horns. Of course, people honk their horns for lots of reasons. You also know that traffic is influenced by rain, sporting events, and time of day. You also know that once the traffic is jammed it tends to stay jammed, and if there is no traffic jam it can suddenly become jammed.
(iii.) Infer the probability that a patient has colon cancer without running expensive tests. You are able to make direct observations of symptoms:
diarrhea, fatigue, and weight loss. The probability of this cancer is affected by the city you live in, your diet, and genetic factors.
(iv.) Suppose you need to assemble a piece of Ikea furniture but lost the instructions. Find a sequence of operations that transform the pile of parts into fully assembled furniture. Assume the problem is fully observable and that each operation cannot fail.
(v). Classifying rodents found on campus by species. There are 3 types of rodents (rat, shrew, or,mouse) that can be identified by inspecting size,color, tail length, whisker length, and size of front teeth. You must consider that any of these attributes can take on a range of values (e.g., a baby rat could be the size of a full-grown mouse).
(vi). An agent that can play tic-tac-toe against a 6-year old child. The child can be assumed to play completely randomly, meaning it has equal chance of marking any open cell.
(a) Perceptron
(b) A*
(c) Markov decision process
(d) Partially-observable MDP
(e) Dynamic Bayesian network
(f) Bayesian inference
(g) Simulated annealing
Question 2. Suppose a person named Zeb who is a US citizen is playing a massive-multiplayer fantasy role-playing game made by a US company. The game company has developed a new AI system that uses a neural network trained on a massive amount of text from the internet to create a dialogue system so that players can talk with virtual, AI-controlled players. Included in this data is financial investing discussions from a social media platform called Reddit that allows anyone to talk to anyone else about any topic. The dialogue system is still experimental and players must agree to opt in. Zeb has been offered the terms of service and has opted in. Further,Zeb talks to an AI-controlled character who recommends investing in a particular stock. Zeb ends up losing a large amount of money. Who can Zeb hold liable?
2.a. (1 point) Give one entity (including the possibility of the AI system itself) that might be held liable in the US legal system. Provide a justification.