6. Applicationsof A.I.
Akinator
Both 20 questions and Akinator exploit a well-known searching algorithm known as
the ‘divide and conquer’ strategy. It’s a strategy often used to solve large problems by
breaking them down into smaller sub-problems.
Whilst it’s not exactly known how Akinator and 20 questions work (both are closely
guarded secrets), it is most likely that they both employ a rules-based “fixed
intelligence” approach which follows a definite series of rules and instructions. A fixed
intelligence approach can be best described using a decision tree (see opposite).
What makes Akinator so special? Akinator differs slightly from other rule-based
approaches by using an A.I. algorithm that learns the best questions to ask through its
experience with players.
Are you human?
Yes No
Are you male?
Yes No
Are you real?
Yes No
7. Applications of A.I.
This was an academic question posed by Cambridge
mathematician, computer pioneer, and A.I. theorist,
Alan Turing.
Turing proposed that, given time, a computer with
sufficient computational power would acquire the
abilities to rival human intelligence.
In order to test his theory, Turing devised a test.
Cancomputersthink?
9. Chatbots
Chatbots
A chatbot is a computer program that processes and
simulates human conversation (either written or
spoken), allowing the user to interact with digital
devices as if they were communicating with a real
person.
Have you ever wondered why chatbots are created?
They are meant to make it feel as if we are talking to
a real human as this is an environment in which we
feel most comfortable.
I am not a
robot!
10. Chatbots
Chatbots
Experiment with some of the chatbots opposite.
Spend some time interacting with the chatbot, and
note down any unusual or unexpected answers
Things to consider:
• What is the purpose of the chatbot?
• What was the interaction with the chatbot
like?
• Do you feel like you are talking to a machine or
a human, and why do you think that?
• Do you feel the chatbot has a certain
personality?
Example chatbots
• Clever Bot - http://www.cleverbot.com
• Eliza - https://www.masswerk.at/elizabot/
Try it:
11. Chatbots
Chatbot Design
In general, there are two
approaches taken by researchers
when buildingAI models.They
either take a rule-based (fixed
intelligence) or learning approach
Whilst the majority of chatbots
incorporate a rule-based
approach some, more advanced,
chatbots have started to use
machine learning
Example of a rule-based (fixed intelligence) chatbot.
12. Chatbots
Chatbots and theTuringTest
The 'Holy Grail' of chatbot design is to create a program that can pass
the famousTuring test. However, despite many advancements in
A.I., we are still a long way from creating a chatbot that can truly
think for itself!
Why do you think it’s so hard to make a machine appear as
if it is able to think like a human?
13. Player 1:
My name is Jason Foster.
I am currently employed as a
Senior Wildlife Rescue Worker
and I am also a member of the
Coastal Wildlife Rescue Team.
My hobbies include boxing, a
specific racing game, magic and
music. I enjoy all kinds of living
things. I love animals and plants
too.
Player 2:
My name is Eric Levenson.
I'm an award-winning game
designer and long-time developer
of electronic games. I've written
two books that are in stores.
My hobbies are watching, reading,
and socializing. I'm a terrible
dancer but love playing the guitar.
I love music. I love programming.
I'm a Star Wars geek, a comic book
fan, and a Halo nerd.
Player 3:
My name is Michael Greenberg.
I have worked with Foxconn, for
almost ten years, developing
high end custom components
for enterprise and consumer
clients.
My hobbies include gardening,
painting, cooking, listening to
music, reading, and drawing. I
would like to spend more time
painting and might try to learn
a new language soon.
Question 1: What is your name?
Question 2: What is your job role?
Question 3: What are your hobbies or interests?
Can you tell the difference between
a human and a machine?
Can you guess which (if any) of the above
statements were written by a human?
Answer: They were all created by a robot!
14. It’s not just conversations that machines try to mimic!
Can you guess which image was NOT created using A.I.?
Answer: The image on the left is real.The image on the right was created using artificial intelligence (AI).
15. Influencers
Meet Lil Miquela?
Miquela Sousa, better known as
Lil Miquela, a 19-year-old Brazilian-
American model, musical artist, and
Instagram influencer with over one
million followers.
What’s so special about Lil
Miquela?
16. AI Influencers
Miquela is not real —
at least not like you
and me. She is an
avatar puppeteered
by Brud, an L.A.-
based start-up, who
specialise in artificial
intelligence and
robotics.
19. Domains of A.I.: Speech APIs
Natural Language Processing
One of the many challenges often faced in the
world of A.I. is understanding human speech.
With the immense diversity of languages,
accents and dialects, the process
of deciphering human speech is quite difficult.
At times, the machine faces difficulties in
learning constructs as there seem to be more
exceptions to the rules (syntax) of a language
than rules themselves.
Try it out: https://aidemos.microsoft.com/luis/demo
20. Domains of A.I.: SpeechAPIs
Speech Authentication
SpeechTranscription
Speech transcription converts the spoken word into
text. It can recognise speech from a microphone or
pre-recorded audio file.
The audio is sent to the server for speech
recognition and conversion into text.
The application can be used to build voice-triggered
smart apps and dictation software such as Cortana
and Siri.
Speech Text
Process of Speech Conversion intoText
21. Domains of A.I.: SpeechAPIs
Text Analytics
Text analytics is the process of converting
unstructured text data into meaningful data
using content analysis, context analysis,
sentimental analysis, and other strategies.
Text analytics applications can use sentiment
analysis to understand the mood of the user and,
in the case of Chatbots or other software, react
accordingly.
For example, it may recognise that someone with
mild Autism is feeling anxious and change the
environment such as the lighting or music
accordingly. Another application could be to warn
a young user in a Chatroom of possible danger
based on the undertones of the conversation.
Try it for yourself:
Try it: https://aidemos.microsoft.com/text-analytics
22. Domains of A.I.: SpeechAPIs
Sentiment Analysis
Sentiment analysis returns a numeric
polarity score between zero (0) and
one (1). Zero indicating 100% negative
sentiment and one indicating 100%
positive sentiment.
It is generated using classification
techniques and supports a variety of
languages.
Negative Neutral Positive
Try it: https://mycomputerbrain.net/php/experiments/ai.experiment19b.php
23. Domains of A.I.: SpeechAPIs
Language Detection
Language detection is designed to analyse
text and assign a numeric score between zero
(0) and one (1). A score close to 1 indicates
that the language identified is correct. A score
closer to 0 is taken as incorrect. This is the
basics of a spell checker.
Remember that some words are common in
more than one language, such as:
• Attaché (French)
• Kindergarten (German) etc.
Both of which are also part of the English
language.
Language detection supports a total of 120
languages.
Example for Language Detection
“Find out a rangeenghar in your
preferred location.”
English language:
Find/ out/ a/ in/ your/ preferred/ location/
Hindi Language:
rangeen/ ghar/
Since most of the words detected are from the English language, the language
detected is English.
24. Domains of A.I.: Vision APIs
ComputerVision
Computer vision is a field of A.I. that
trains machines to interpret and
understand the visual world.
Using still images and videos, machines
can accurately identify and classify
objects.
Try it out: https://aidemos.microsoft.com/computer-vision
25. Domains of A.I.: Vision APIs
There are more than 27 landmarks for each face in the image, including:
1. Age
2. Emotion
Based on these landmarks the application can give a range of answers.
3. Gender
4. Pose
Face recognition applications
are built to detect one, or
more, human faces in a still
image or video and, based
on machine learning
predictions of facial features,
can determine a person’s
age, gender and even their
emotional state.
Face and Emotion Recognition
5. Smile
6. Facial hair
26. Domains of A.I.: Vision APIs
Mapping of key facial landmarks Searching for familiar faces and tagging them accordingly
Face Identification Similar Face Search
27. Domains of A.I.: Vision APIs
Grouping of similar looking faces in numerous different images The basic seven types of emotions
Face Grouping Emotion Recognition
FaceandEmotionRecognition
28. Domains of A.I.: Speech& Vision APIs
Video Indexing in Progress
VideoIndexer
Video Indexer is a cloud application built
on a raft of tools, including Media
Analytics, Search, Cognitive Services –
such as the Face API, MicrosoftTranslator,
ComputerVisionAPI,Custom Speech
Service, and more.
It is designed to extract ‘Insights’ or things
that are happening such as keywords
spoken, recognised people, facial
emotions, tone of speech, etc.
Try it out: https://aidemos.microsoft.com/video-indexer
33. DigitalAssistants:Cortana
• Send reminders based on time, places, or people.
• Find facts, files, places, and information from the Internet.
• Play music, podcasts, and radio stations.
• Track sports teams, interests, and flights.
• Send emails and texts.
• Manage your calendar and keep you up-to-date.
• Create and manage lists.
• Chat with you and play games.
• Open any application on your device
• Interact with smart home devices.
Example tasks performed by a digital assistant
34. Applications of A.I.
Virtual Assistants
Oneofthebest-knownexamplesof'when
chatbotsgetitwrong’isTay,theMicrosoft
chatbotthathadtoberemovedafter
spewingracisttweets.
When chatbots
and virtual assistants go wrong!
35. Applications of A.I.
Virtual Assistants
Okay, from now
on, I’ll call you “an
ambulance.”
Siri, call me an
ambulance!
Another well-known example of 'when
chatbots get it wrong' is the response to "Call
me an ambulance" by Siri during its initial
release.
While the error has since been resolved, it
demonstrates some of the problems faced
when programming a computer to understand
human instructions.
When chatbots
and virtual assistants go wrong!
37. Other Applications
Bing spell check
The Microsoft Spell Check API
enables contextual grammar and
spell checking on the text it is
supplied as input.
While most spell-checkers, such as
those only offline, rely on
dictionary-based rule sets. The
Bing spell-checker leverages
machine learning and statistical
machine translation to provide
accurate and contextual
corrections.
38. Other Applications
LanguageTranslation
The Microsoft Translator API
translates text from one written
language to another. It can be sent
a file containing the text, or
translate in real-time via the web.
When linked to other cognitive
services, an application can
recognise an image, extract the
name of the image to any
language, translate it into the user's
native language and speak it back
to them. Try the Microsoft
Translator App available for all
smart phones to see this in action.
Try it: Bing MicrosoftTranslator
The ‘divide and conquer’ approach is a popular search algorithm used to solve large problems, by breaking them into smaller sub-problems. With the ‘divide and conquer’ approach, the key to success is in the questions. For example, in the case of 20 questions, if your first question was ‘Are you Donald Trump?’ you only rule out one possibility (great if you’re right!), but if your first question is ‘Are you male? or ‘Are you human?’, you can improve your search algorithm massively by ruling out a large proportion of the possibilities. In the case of Akinator, an artificial intelligence algorithm is used to learn the best questions to ask through its experience with players, hence improving its search algorithm each time the game is played!
If a computer can learn from experience, can it actually think for itself? This was also a question posed by computer pioneer and artificial intelligence (AI) theorist, Alan Turing. Turing proposed that, given time, a computer with sufficient computational power would acquire the abilities to rival human intelligence. In order to test his theory, Turing devised the ‘Turing test’.
The Turing Test was based on a Victorian parlour game in which a judge (or interrogator) asks a series of questions to a man and a woman in a separate room. By reading a series of typed answers, the judge must determine which replies were from the man and which were from the woman. Turing adapted the test by replacing the woman with a computer - the aim being to decide whether the answers were from a man or computer thus determining if a computer was able to think for itself.
For years, science fiction writers and filmmakers have dreamed about robots that can think for themselves however, despite huge advancements in technology, this dream is still far from reality. Let’s explore some example chatbots.
In general, there are two approaches taken by researchers when building AI models. They either take a rule-based (fixed intelligence) or learning approach.
Whilst the majority of chatbots incorporate a rule-based approach some, more advanced, chatbots have started to use machine learning (see the next slide).
As of 2020, no machines have truly passed the Turing Test, but there have been claims that a chatbot by the name of Eugene Goostman passed the test in 2014 by fooling 33% of the judges into thinking it was a human.
Some or all of the example statements on this slide were produced by an ‘intelligent’ machine. Can you guess which (if any) were written by a human?
Answer: They were all created by a robot!
The image on the right was created using artificial intelligence (AI). Thispersondoesnotexist.com generates a new lifelike image each time the page is refreshed, using technology developed by chipmaker Nvidia.
Lil Miquela is not quite like any other influencer; can you guess why?
Miquela isn’t real — at least not like you and me. She is an avatar puppeteered by Brud, an L.A.-based start-up, who specialise in artificial intelligence and robotics.
Miquela announces to the world that she is a robot: https://www.youtube.com/watch?v=S6wnHsEoTmc
In the previous unit, we looked at a variety of different artificial intelligent APIs (Application Program Interfaces). The two most common APIs used in AI are Speech & Language and Vision.
The face recognition application is built to detect one or more human faces and return face rectangles in addition to various attributes of the face. These contain machine learning predictions of facial features. The API is designed to detect faces in the images and identify these features. The API can also be used to analyze videos in real-time by looking at individual frames.
The Face API allows searching, identifying and matching faces based on training faces saved in a repository of up to 1 million people. The API searches the face of the human and matches certain points on the face marked as identifiers. It recognises the face before giving the output.
The Face API can also recognise the emotions displayed by the human face. The API distinguishes between emotions such as anger, happiness, contempt, fear, disgust, sadness, and even surprise. The API is smart enough to understand human emotions and expressions across cultures. It also understands the way people universally communicate using particular facial expressions.
A Video Indexer gives you access to video statistics such as how long people view your video ad for. It is recommended especially if you have linked your video to a particular webpage.
An app which exemplifies all these technologies is the Seeing AI app.
Try it out: https://www.microsoft.com/en-us/ai/seeing-ai
Video with captions available here: https://youtu.be/rVF2duPVUTY
The use of digital assistants has become more pronounced in business applications as human assistants can suffer from fatigue and time management issues. A digital assistant by contrast is a machine and can be at our disposal 24 hours and 365 days a year. It does not suffer from fatigue and is consistent and unbiased.
Unfortunately, in the first 24 hours of coming online, a coordinated attack by a subset of people exploited a vulnerability in Tay. And, whilst nobody who uses social media could be too surprised to see that the chatbot encountered hateful comments and trolls, the artificial intelligence system didn't have the judgment to avoid incorporating such views into its own tweets. It also highlights the problem with machine learning and the quality of the data fed into the machine.
While the error has since been resolved, it demonstrates some of the problems faced when computers try to understand human instructions.
The Microsoft Spell Check API helps people ensure that their written document is free of spelling and grammatical errors. It has been a major element of Microsoft Word for many years but now covers all applications as well as online versions under Office365
Translator enables the translation of a document from one written language to another.
What if neither distance nor language mattered? What if technology could help you be anywhere you need to be and speak any language? Using AI technology and holographic experiences this is possible, and it is revolutionary.