Detailed schedule and resources

Class 9

Before class: show handwritten notes on “The AI Deluge.”
Announcements:
- Required event tonight – 7pm ATS
- Quiz 3 on Friday – questions 1-4 will not be tested
- Exam 1: Topics 1-3. All questions are similar to questions in Q1, Q2, Q3.
- Questions on video lecture (class 7A below).
device-free from this point please
Discuss the reading from Elizabeth Ayanna Johnson, which is the interview of Mustafa Suleyman, “The AI Deluge”.
What is sustainability?
- our own ideas
- UN Sustainable Development Goals (SDGs)
- In teams: Each team discuss which SDGs are relevant to AI.
  - Name the top 1-3 SDGs that are degraded by AI
  - Name the top 1-3 SDGs that could be improved with AI.
  - Report back.
Data center sustainability. powerpoint
- optional to watch at home: “Real-World Effects of AI.” Clarke Forum for Contemporary Issues panel session, February 26, 2026; web; video.
Devices may be used for the rest of the session
Go over the sustainability poster (SP) assignment .
- example of presentation style: jmac poster
- any topic linking AI and sustainability
Remaining time: individual/team work on SP assignment

D1 discussion: Thinking AI Ch1-4, Turing (1950).

Quiz 2 (15 minutes)
D1 discussion class will be this Friday 9/25 in Tome 228 (the science library). Same requirements as before – bring handwritten notes etc.
In the remaining time today we will run through as much of Quiz 3 as possible. Anything that isn’t covered will be covered in a video lecture to be released later this week.

Key definitions that are not in the quiz solutions:

softmax function is defined as: \( \text{softmax}(x_i) = \frac{e^{x_i}}{\sum_{j} e^{x_j}} \)
cross-entropy loss is defined as: \( \text{cross-entropy}(y, \hat{y}) = -\sum_{i} y_i \log(\hat{y}_i) \)
- compare with binary cross-entropy loss, which is defined as: \( \text{binary-cross-entropy}(y, \hat{y}) = -[y \log(\hat{y}) + (1-y) \log(1-\hat{y})] \)
Softmax is a generalization of the logistic function in the following sense: \( \text{softmax}(x, 0) = (\sigma(x), \sigma(-x)) \). So when there are only two classes, the logistic function plays the role of softmax.

Visiting speaker: Justin Ross, COO & Chief Engineer at Pennsylvania Data Center Partners.

Announcement:

Note new website content:
- quiz retakes
- discussion grading
- reading for next time: skip Turing sections 4 and 5.
- Friday’s D1 discussion class will be in Tome 228 (the science library).

Lecture:

Matrix Algebra section 2 and 3, especially broadcasting + quiz 2 question 10
Quiz Two questions 6, 7, 8, 9. If we run out of time, please go over the rest for homework.

we skipped:

we did:

Also: are you keeping up with the readings? i.e. Thinking AI Ch1-2.

arrangements for next week’s field trip on Tuesday 9/8:
- Meet in Kaufman parking lot near the entrance to Public Safety no later than 1:30 PM.
- optional: Wear shoes that can get muddy if you want to stroll on the Appalachian Trail for a few minutes.
- our route: Kaufman parking lot to PAX1 site then Appalachian Trail
- We will be back on campus before the end of class at 2:45 PM.
any questions on the syllabus?
- Discussion of AI use policy
review search node with fields state, parent, action, depth, cost
- show Java file SearchNode.java from HW1
Show BFS via quiz1 qu1, skip DFS
Show UCS via quiz1 qu3
Show A* via quiz1 qu4
Reference implementation available: search.zip
define admissible heuristic, consistent heuristic:
- admissible: never overestimates the cost to reach the goal, so \(h(n) \leq h^*(n)\)
- consistent: the estimated cost of reaching the goal from a node is less than or equal to the cost of reaching a neighbor plus the estimated cost of reaching the goal from the neighbor, so \(h(n) \leq c(n, n’) + h(n’)\)
define 8-puzzle heuristics:
- h1: number-of-misplaced-tiles
  - h1(state) = number of tiles in the wrong position
- h2: manhattan-distance
  - h2(state) = sum of horizontal and vertical distances of the tiles from their goal positions
if time, review Matrix Algebra section 1