Scribd Logo
Upload

Welcome to Scribd!

  • 24 views

    AI - Introduction - Laura Perea

    Uploaded by

    Netflix Team
    The document discusses artificial intelligence and machine learning. It provides a brief history of AI from its origins in the 1840s to modern developments in deep learning. It then defines AI and machine learning, describing techniques like supervised learning, unsupervised learning, and reinforcement learning. The document outlines some of the capabilities of modern AI systems like image classification, chatbots, and recommender systems. It also discusses risks, limitations, and the importance of responsible AI development and use.

    Copyright:

    © All Rights Reserved
    Download as PDF, TXT or read online from Scribd

    AI - Introduction - Laura Perea

    Uploaded by

    Netflix Team
    24 views23 pages
    The document discusses artificial intelligence and machine learning. It provides a brief history of AI from its origins in the 1840s to modern developments in deep learning. It then defines AI and machine learning, describing techniques like supervised learning, unsupervised learning, and reinforcement learning. The document outlines some of the capabilities of modern AI systems like image classification, chatbots, and recommender systems. It also discusses risks, limitations, and the importance of responsible AI development and use.

    Original Title

    AI_Introduction_Laura Perea

    Copyright

    © © All Rights Reserved

    Available Formats

    PDF, TXT or read online from Scribd

    Did you find this document useful?

    Is this content inappropriate?

    The document discusses artificial intelligence and machine learning. It provides a brief history of AI from its origins in the 1840s to modern developments in deep learning. It then defines AI and machine learning, describing techniques like supervised learning, unsupervised learning, and reinforcement learning. The document outlines some of the capabilities of modern AI systems like image classification, chatbots, and recommender systems. It also discusses risks, limitations, and the importance of responsible AI development and use.

    Copyright:

    © All Rights Reserved
    Download as PDF, TXT or read online from Scribd
    Download as pdf or txt
    24 views23 pages

    AI - Introduction - Laura Perea

    Uploaded by

    Netflix Team
    The document discusses artificial intelligence and machine learning. It provides a brief history of AI from its origins in the 1840s to modern developments in deep learning. It then defines AI and machine learning, describing techniques like supervised learning, unsupervised learning, and reinforcement learning. The document outlines some of the capabilities of modern AI systems like image classification, chatbots, and recommender systems. It also discusses risks, limitations, and the importance of responsible AI development and use.

    Copyright:

    © All Rights Reserved
    Download as PDF, TXT or read online from Scribd
    Download as pdf or txt
    You are on page 1of 23

    We make

    it happen

    Artificial Intelligence
    Dr Laura Perea Virgili

    Accelerator Digital Program

    eae.es
    Today´s Objective

    Main Concepts Risks

    Limitations Opportunities
    Brief History & General culture
    Machine Learning
    AI is first First General AI is born AI beats human & Deep Learning
    envisaged Purpose Computer

    ~1840 1945 1956 1997 2011


    The potential of the Allan Touring Artificial Intelligence IBM’s deep blue IBM’s Watson beats
    computers to create designed the first term is coined by computer beats Gary human players on US
    art and its capacity general purpose John McCarthy Kasparov at chess game show Jeopardy
    for self-awareness computer
    is first envisaged

    2012 2020 2022


    AlexNet beats human AlphaFold 2 amazed ChatGPT
    in an image the world for its revolutionizes AI with
    recognition challenge capacity to predict new LLM models
    protein folding
    What is Artificial Intelligence?
    Artificial intelligence (AI) is intelligence demonstrated
    by machines, as opposed to intelligence of humans and other
    animals.

    Business / Domain
    Artificial intelligence
    expertise

    Machine
    Learning

    Speech recognition
    Data Big Data

    DeepLearning

    Statistics
    Natural Language Next AI
    Processing

    Classical ML
    What can AI (learn to) do?
    Artificial intelligence (AI) is already commonly used to …

    I’ll put you in


    contact with an
    agent

    Image Classification Chat Bots Recommender systems


    LLM Hallucinations …

    Source: ‘Preventing hallucination in GPT 3’


    Types of Machine Learning

    Supervised algorithms

    Unsupervised algorithms

    Reinforcement learning
    Supervised Learning
    The system (agent) learns to map some input into an output based on examples.
    It infers a function from labelled training data consisting of a set of training examples

    Image classification

    DOGS CATS
    ??
    Examples of applications:
    • Quality control on production line
    • Sorting and classificaiton
    • Identification of safety risks
    Supervised Learning
    The system (agent) learns to map some input into an output based on examples.
    It infers a function from labelled training data consisting of a set of training examples

    (Time) series forecast

    Examples of applications:
    • Demand/supply planning
    • Finance
    • Data driven business decisions
    • …
    Supervised Learning
    The system (agent) learns to map some input into an output based on examples.
    It infers a function from labelled training data consisting of a set of training examples

    (data driven) Decision trees

    Examples of applications:
    • Chunk retention
    • Risk forecast
    • Informed business decisions
    • Modelling behaviour

    Source: https://www.datacamp.com/tutorial/decision-tree-classification-python
    UnSupervised Learning
    A type of self-organized learning. The agent finds previously unknown patterns in data
    without pre-existing labels.

    Anomaly detection
    Examples of applications:
    • Fraud detection
    • Safety risk detection
    • Security risk detection
    • Maintenance (early detection of problems)
    • Quality controls
    UnSupervised Learning
    A type of self-organized learning. The agent finds previously unknown patterns in data
    without pre-existing labels.

    Clustering

    Examples of applications:
    • Recomendation systems
    • Market segmentation
    • Social network analysis
    • Pattern recognition
    • …
    Reinforcement Learning
    Agents learns by taking actions in an environment to maximize some cumulative reward.

    Examples:
    • Play chess
    • Self-driving cars
    • Identify attacks on a webpage
    • Logistics optimization
    • Creative works (art, music, literacy, …)
    What can AI
    (learn to) do?
    Generative AI

    Modernist drawing of an AI by NightCafe

    ChatGPT by OpenAI
    AI Today … can generate art

    Artificial intelligence: Midjourney Artificial Intelligence: Dall-E


    AI Today … can express human emotions

    Link to video
    AI Today … can socially interact among them
    “Chatbots like ChatGPT aren't as smart as
    dogs or cats yet”

    Yann LeCunn link

    Selfconscious
    AI
    Risks in AI …

    Confidentiality and Privacy: Input may become available to third parties.

    Intellectual Property: Data input as well as output may raise concerns around intellectual
    property and other proprietary rights.

    Performance & Misuse: Risk of hallucinations, lack of references, toxic content, …


    … and limitations

    Explainability Zero knowledge refresh


    Responsible AI
    Like deep learning in general, it impossible to After the model training. Most of the models,
    explain the reasons of the results provided or like Large Language Models (LLM’s) are based on The generated output may be biased, wrong
    trace on sources used to provide its answers. deep learning models extremely expensive to our outdated. Very high impacct of gas
    train. Once trained on a corpus, knowledge is emissions, specially during training.
    frozen in time and recent events are unknown
    by it.
    Responsible AI
    AI offers plenty of new opportunities but can
    bring significant risk with legal consequences
    and reputation damage.

    Applies to AI developers & AI adopters.


    We make
    it happen

    Gracias por vuestra atención!

    eae.es

    You might also like