1 of 42

  1. Teaching Machines to Learn the World

2 of 42

Who are we?

3 of 42

Our Goal

Value

Investment

For the value to be greater than the investment.

Understanding

AI technology.

AI knowledge opens doors!

Once a week

for 1 hr.

4 of 42

Involvement In HUMIC

  • Future of Intelligence Fellowship
    • Goal: Understanding AI technology
  • AI Project Incubator
    • Goal: Apply this knowledge to actual AI coding projects

5 of 42

Involvement In HUMIC

  • Future of Intelligence Fellowship
    • Goal: Understanding AI technology
  • AI Project Incubator
    • Goal: Apply this knowledge to actual AI coding projects

FIF (optional meetings)

  • Short coding labs
  • Smaller sections
  • Paper reading groups

6 of 42

Who are YOU?

7 of 42

Computer Science

  • Supervised learning, gradient descent, clustering algorithms, reinforcement learning, transformer architectures, neural networks, etc….
  • AI is grounded in Computer Science!

17

8 of 42

Math/Statistics/Applied Math

21

9 of 42

Physics/Astrophysics

3

10 of 42

Economics

  • All the big companies have opened AI ventures: Google, Microsoft, Meta, etc.
  • Open AI has raised ≈ $58B total (2025)
  • In 2024
    • Corporate AI investment hit $252.3B U.S. private AI investment was $109.1B
    • 2,049 new AI startups formed
  • WHY????... BC AI IS A BIG DEAL

20

11 of 42

Government

7

EU Laws will be implemented rolling in 2026.

12 of 42

Environmental Science

  • It does help with some things
    • Weather forecasting, Renewable energy optimization, endangered wildlife imaging, coral reef imaging, deforestation detection, etc.

3

13 of 42

Social Studies, Sociology, English, History of Science, Philosophy

8

At Harvard’s halls, bright minds begin to weave,�A fellowship where future thought takes flight.�Through learning’s spark, new visions they conceive,�Machine and human joined to seek the light.

14 of 42

Neuroscience, Psychology, MBB

7

15 of 42

Bio (Bioengineering, HDRB, Chemical and Physical Biology, Integrative Biology, etc.)

17

16 of 42

AFVS, Music, Theater

3

ChatGPT generated: Mona Lisa

Harvard club!

In Harvard square!

17 of 42

18 of 42

You will be introduced to each of these technologies!!

Now LOCK IN for our first bit of learning… :)

19 of 42

Agenda

  1. Welcome to HUMIC
  2. Overview of AI
  3. ML fundamentals
  4. Preview for Deep learning

20 of 42

Agenda

  • Welcome to HUMIC
  • Overview of AI
  • ML fundamentals
  • Preview for Deep learning

21 of 42

AI > ML > DL

Deep Learning

Machine Learning

AI

“Computer systems able to perform tasks that normally require human intelligence”

“Using data and algorithms to enable AI to imitate the way that humans learn, gradually improving its accuracy”

“teaches computers to process data in a way that is inspired by the human brain”

22 of 42

Logical Reasoning

Spatial Reasoning

Language understanding

Creativity

x + 3 =y,

y + 1 = z

=> x +4 = z

23 of 42

Agenda

  • Welcome to HUMIC
  • Overview of AI
  • ML fundamentals
  • Preview for Deep learning

24 of 42

Machine Learning

Supervised learning

Reinforcement Learning

Unsupervised Learning

25 of 42

Supervised Learning

26 of 42

Supervised Learning

  • Given some training data pairs: (x_1, y_1), (x_2, y_2),…,(x_n, y_n)

  • If guess differs from label, adjust the model a little bit to learn from that error
  • After training, we can input some new data x_(n+1)

ML model

?=

x_i

g_i

y_i

27 of 42

Gradient Descent

  • Loss function

  • Improve parameters of model
  • Iteratively get lower and lower losses
  • Is a loss of 0 always good?

28 of 42

Unsupervised Learning

29 of 42

Clustering data points with similar properties

CS

Chemistry

Econ

Chem/bio/physics

CS & Phil

Finance

30 of 42

K-Means Algorithm for Clustering

31 of 42

Reinforcement Learning

32 of 42

Reinforcement Learning

Model

y=h(x)

Input x

Output y

Supervised Learning Setting

Reinforcement Learning Setting

Model

a=h(s)

State s

Action a

New state s’

33 of 42

Reinforcement learning

  • Don’t exactly know how to get correct answer, but can tell if an answer is good / bad
  • Use a reward function — encourage certain intermediate states more than others
  • Many different solutions to the same end result – some might be surprising

34 of 42

Atlas from Boston Dynamics

35 of 42

Agenda

  • Welcome to HUMIC
  • Overview of AI
  • ML fundamentals
  • Preview for Deep learning

36 of 42

What’s coming :)

37 of 42

Who are YOU?

38 of 42

Attendance and Feedback

©2022 HUMIC. All rights reserved.

39 of 42

The Godfathers of AI

40 of 42

Geoffrey Hinton

  • Neural Networks & Backpropagation

“Now that neural nets work, industry and government have started calling neural nets AI. And the people in AI who spent all their life mocking neural nets and saying they'd never do anything are now happy to call them AI and try and get some of the money. “

41 of 42

Yoshua Bengio

  • NLP, AutoEncoders, Sequence Models (RNNs)

42 of 42

Yann LeCun

  • Convolutional Neural Networks, Computer Vision