Rex BriggsProjects & field notes

The AI Conundrum / Self-assessment

How well do you understand AI?

Twelve questions about AI’s foundations, limits and risks. Rate how confidently you could answer each one. This measures your own assessment of your knowledge, not demonstrated proficiency.

All 12 ratings are required. Your score updates automatically, from 12 to 120 points. Name, company, email and training interest are optional.

Your answers and any optional contact details are sent to Rex Briggs’s existing Google Form only when you select Submit. Google will display the submission confirmation. Privacy details.

1. Can you describe, specifically, how AI uses weights, bias and functions to iteratively fit patterns?
2. Can you describe how AI differs from human intelligence in terms of using context in visual, language and math?
3. Can you describe how Gradient Descent differs from systematic search of a solution space?
4. Can you describe how the dimensionality of information is calculated, the difference in solution space between two dimensions, three dimensions, and 32 dimensions for example and why it matters to AI?
5. Can you describe under what conditions gradient descent is likely to produce a spurious output?
6. Can you describe how compression bottlenecks dimensions, and the risk in losing precision in the process?
7. Can you describe how false accept and false reject are used to measure the precision of AI output? Can you define when to use training data vs. hold out vs. in-market/clinical path data for model development and validation?
8. Can you describe the difference for AI in playing a game like Chess or Go versus operating an autonomous vehicle in terms of the breadth of AI skills and inputs and where the risk of adversarial attack is highest?
9. Can you describe what it is about AI that makes providing a specific rationale for its output difficult, and some scenarios where a rationale for AI’s output would be important and under what scenarios a rationale for AI output is not necessary?
10. Can you explain the difference in the power law and normal distributions in AI observations and which part of the power law poses a problem for AI? Can you explain how bias can arise in normally distributed AI data, and different strategies for measuring bias and offsetting it?
11. Can you explain under what circumstances AI will likely replace previous non-AI approaches, even if it is less precise?
12. Can you share an example where AI has generated over a billion dollars in losses based on AI applied in a risky scenario, and what could have been done differently to avoid the outcome? Can you describe a scenario where AI generated over a billion dollars in value, and what made the scenario less risky for the application of AI?

Optional details

Your score

0 of 12 answered

— / 120

Keep a note of your score before submitting. A higher score reflects greater self-reported familiarity; it is not a certification.

Nothing is saved until you submit and Google confirms receipt. If Google displays an error, return here or use the original Google Form.

Return to the MMA exercises →