1 of 60

1

What is Artificial Intelligence?��Principles of AI Technology

​

Georgios A. Pavlopoulos

Research Director, BSRC Alexander Fleming

www.pavlopouloslab.info

​

2 of 60

2

3 of 60

Lab Directions

3

Networks �

​

Visualization

Text-Mining

Data Integration

Genomics

Metagenomics

4 of 60

4

Human Cognition

5 of 60

5

Types of Human Intelligence

6 of 60

6

Artificial vs Human Intelligence?

7 of 60

Human vs Machine

7

8 of 60

The old way: write the rules

IF has_fur AND has_whiskers

AND pointy_ears → cat

  • A human has to think of every rule

The new way: show examples

100,000 photos, each labelled

"cat" or "not cat" → let it work it out

  • Nobody defines "cat" — the machine finds what separates them

How to Teach a Computer

8

9 of 60

9

Biological vs Artificial Neuron?

10 of 60

10

Biological vs Artificial Network?

11 of 60

11

An Artificial Deep Neural Network?

12 of 60

12

Feedback Loop

13 of 60

13

Example

28 x 28 pixels

​

​

784 pixels

14 of 60

14

Example

15 of 60

15

Patterns

16 of 60

16

Hidden Layers

17 of 60

Text

Essays, code, translation, summaries.

Images

A description in; a picture that never existed out.

Video & audio

Clips, voices, music, sound effects.

Code

Writes and debugs alongside programmers.

Molecules

Proposes new drug candidates atom by atom.

Proteins

Designs proteins evolution never got around to.

Is That All? - Different Data Types

17

18 of 60

AI History

18

19 of 60

19

Variety of AI Models

20 of 60

Data

Computing power

A better design

What Has Changed

20

21 of 60

Large Language Models (LLMs)

21

22 of 60

Cosine Similarity

22

23 of 60

Embeddings - Every Word Is A Vector

23

24 of 60

Embeddings - Every Word Is A Point

24

25 of 60

Everything Becomes a Point on a Map

25

26 of 60

Embedding Space

26

27 of 60

Embedding Space 2D vs 3D

Dimensionality Reduction – No Unique Representation�

The Embedding Space is Multidimensional

27

28 of 60

Embeddings, Tokens, Parameters

Every Word is a Vector��Every word has a unique ID (Token)

28

29 of 60

LLM Arithmetics

29

30 of 60

LLM Word Prediction

30

31 of 60

LLM Word Prediction

31

Target to predict

The LLM cannot access words past the target

Input the LLM receives

32 of 60

Looking At The Neighbors

32

The word ‘fly’ can have a dual meaning

33 of 60

Embeddings Beyond Text

33

Embeddings or �Vector Representations

34 of 60

DNA is a Language

34

35 of 60

LLMs and Hallucinations

LLMs are made to sound right� �...not to be right

35

On my birthday, I would like to have Cakes with chocolate

On my birthday, I would like to have

36 of 60

LLMs and Hallucinations

36

37 of 60

AI Present and Future Architectures

37

38 of 60

38

AI – Non Technical Discussion

39 of 60

Present and Future of AI

39

40 of 60

Next Big Jump

Quantum Computing

40

41 of 60

AI Market Size

41

42 of 60

AI: Pros & Cons

42

43 of 60

AI Regulation Emerges

43

Video Demonstration

44 of 60

AI Ethics

44

45 of 60

45

AI in Biotech

46 of 60

AlphaFold

46

47 of 60

AI and Drug Discovery

47

48 of 60

48

Personalized medicine

49 of 60

Image Recognition

49

50 of 60

Vaccine Discovery

50

51 of 60

Cell Mechanisms

51

52 of 60

Biodiversity

52

53 of 60

Robotics

53

54 of 60

Health Assistants & Chat Bots

54

55 of 60

Genomics: Human Diseases and Aging

55

56 of 60

Wearables

56

57 of 60

Health Records

57

58 of 60

Philosophical (or not) Questions - Can We?

58

Can we store our experiences?

Can AI go analog?

59 of 60

Philosophical (or not) Questions – AI’s “Religion”?

59

60 of 60

60