1What is the primary conceptual difference between 'Weak AI' and 'Strong AI'?
Weak AI focuses on building machines that act intelligently, while Strong AI seeks to create machines with actual minds.
2In the standard Turing Test (TT), how is a computer's success as a 'thinking machine' determined?
The machine must be indistinguishable from a human judge in a text-based conversation at least \$50\%\$ of the time.
3What is the core argument of Ned Block's 'Aunt Bertha' (Blockhead) thought experiment against the Turing Test?
Intelligence depends on the internal functional organization of a system, not just its outward linguistic behavior.
4How does the 'Total Turing Test' (TTT) proposed by Harnad differ from the original Turing Test?
It requires the artifact to possess sensorimotor capabilities to interact with the physical world in a way indistinguishable from humans.
5List the three layers of mechanizing deduction in order of increasing difficulty.
Proof checking, proof discovery, and conjecture generation.
6According to the Computational Theory of Mind (CTM), what is the relationship between mental states and computation?
Mental states are symbolic representations, and thinking is the algorithmic manipulation of these symbols based on their syntax.
7What does the 'Physical Symbol System Hypothesis' (PSSH) propose as the necessary and sufficient means for general intelligent action?
A physical symbol system capable of creating and evolving collections of symbol structures.
8What is the central conclusion of Searle's 'Chinese Room Argument' regarding computer programs?
Formal syntax alone is not sufficient for the possession of semantic mental contents or understanding.
9In the context of GOFAI, what is the 'Frame Problem'?
The challenge of efficiently determining which beliefs should remain unchanged after a specific action is performed.
10How does 'Situated AI' (AI from below) approach the use of internal symbolic representations?
It minimizes or ignores internal representations, focusing instead on continuous sensorimotor interaction with the environment.
11What characteristic of neural networks allows them to continue functioning proportionately after damage to some of their nodes?
Graceful degradation.
12What does the acronym 'GOFAI' stand for in the context of classical AI research?
Good Old-Fashioned Artificial Intelligence.
13What was the primary achievement of the 'Logic Theorist' (LT) program presented at the 1956 Dartmouth conference?
It was capable of proving elementary theorems in the propositional calculus.
14Why is 'conjecture generation' considered the most difficult task in mechanizing human-level reasoning?
It requires the ability to propose 'interesting' conclusions that are likely to follow logically from a set of premises.
15According to Creel and Hellman, what is an 'algorithmic leviathan'?
An automated decision-making system that makes uniform, standardized judgments across broad swathes of a specific sector.
16Under what specific condition do Creel and Hellman argue that isolated arbitrary decisions are of moral concern?
They are problematic only when other rights or reliance interests make nonarbitrariness relevant to the decision.
17How is 'predictive multiplicity' defined in the context of machine learning models?
The existence of multiple, equally optimal models that produce different individual classifications for the same input.
18According to Creel and Hellman, when does the arbitrariness of an algorithm become a significant moral problem?
It becomes problematic when it is used at scale to systematically exclude an individual from a broad range of opportunities.
19What is a 'contaminated' feature in an algorithmic decision-making system?
A feature that combines a genuinely relevant attribute with an arbitrary data artifact or noise.
20Why does training different algorithms on the same data set (e.g., ImageNet) often lead to uniform outcomes?
The models learn the same brittle artifacts and accidental patterns present in the shared data distribution.
21What technical solution do Creel and Hellman propose to mitigate the harms of 'algorithmic leviathans'?
The use of random classifier ensembles to distribute the burden of misclassification.
22How do 'random classifier ensembles' improve fairness for qualified individuals who might be misclassified?
By randomly selecting from a set of near-optimal models, the system ensures that excluded individuals have opportunities across different trials.
23In the 'evidence-relative' view of nonarbitrariness, what justifies the use of a specific screening tool?
The tool is justified if current evidence suggests it is reasonably likely to achieve the agent's specific goals.
24Why might the use of data from only a few elite hospitals lead to standardization in medical AI?
It creates a bottleneck where models from various firms are trained on the same unrepresentative cohort.
25What is 'overfitting' in machine learning, and how does it relate to arbitrariness?
Overfitting occurs when a model memorizes accidental patterns or noise in training data, resulting in arbitrary classification features.
26Why is the 'monopoly of opportunity' created by standardized algorithms morally concerning to Creel and Hellman?
It limits an individual's autonomy and life chances by preventing them from finding alternative paths within a sector.
27What is the primary distinction between the 'right to be informed' and the 'right to explanation' in Wachter et al.'s analysis?
The right to be informed covers general system logic, while a right to explanation would cover the specific reasons for a particular decision.
28How does an 'ex ante' explanation of automated decision-making differ from an 'ex post' explanation?
Ex ante explanations describe system functionality before use, while ex post explanations justify a specific individual's outcome after it occurs.
29Which GDPR Article provides individuals with the right not to be subject to a decision based solely on automated processing?
Article 22.
30What is the main legal limitation of the 'meaningful information about the logic involved' requirement in the GDPR?
It typically refers to the general architecture and goals of the system rather than the weights or features of an individual decision.
31How do Wachter et al. characterize the 'right to explanation' in the current form of the GDPR?
They argue that it does not legally exist as an enforceable right for specific automated decisions.
32In the context of the GDPR, what does 'significance and envisaged consequences' refer to?
The intended high-level impact or the type of outcome the automated system is designed to produce.
33What major hurdle do the authors identify for explaining the 'logic' of complex AI like deep learning?
The inherent opacity or 'black box' nature of these systems makes their decision-making mechanisms difficult for humans to interpret.
34How do Articles 13 and 14 of the GDPR differ regarding notification duties?
Article 13 applies when data is collected directly from the subject, while Article 14 applies when data is obtained from third parties.
35Why do Wachter et al. suggest the GDPR may be 'toothless' against automated decision-making?
It lacks explicit, well-defined rights that allow individuals to effectively challenge or understand specific automated outcomes.
36What is the 'right to access' granted under Article 15 of the GDPR?
The right for a data subject to confirm if their data is being processed and to receive a copy of that data.
37What is the 'systematicity' problem of algorithms identified by Creel and Hellman?
Standardized algorithms ensure that a person misclassified once is likely to be misclassified by every employer or lender using that system.
38In the context of mechanizing deduction, what is the goal of 'proof discovery' given premises \$P_1, \dots, P_n\$ and conclusion \$P\$?
To determine if \$P\$ follows logically from the premises and, if so, produce a formal derivation of it.
39How does 'procedural semantics' (conceptual-role semantics) define the meaning of a mental symbol?
Meaning is fixed by the role the symbol plays in the system's cognitive life, such as its participation in inferences.
40What is the 'symbol grounding problem' as it relates to CTM?
The challenge of explaining how mental symbols acquire meaning and refer to objects in the physical world outside the brain.