4th HLF – Hot Topic: Artificial Intelligence – Presentation Raj Reddy

Summary

The speaker discusses the transformative potential of voice computing, particularly for the semi-literate population of over 3 billion people who cannot read or write effectively. Voice computing allows these individuals to interact with technology solely through speech, offering access to services like reading newspapers, shopping, education, and voting. Existing AI technologies, such as speech recognition, translation, and synthesis, can support this. The speaker highlights how voice computing could bridge the digital divide, empower illiterate populations, and create a massive market for speech-based apps, transforming access to information globally.

Takeaways

• 🎤 Voice computing has the potential to create a profound societal impact by removing the need for keyboards and touch interfaces.
• 🌍 The focus is on the 3 billion semi-literate or illiterate people, who can only interact with technology through voice.
• 🗣️ Voice computing can empower semi-literate populations to access services like reading newspapers, voting, online shopping, and education.
• 🎥 Technologies like real-time speech-to-speech translation, synthesis, and dynamic dubbing already exist, but are not yet fully developed for everyday use.
• 📱 Simple, speech-based mobile apps tailored to semi-literate users could revolutionize how they interact with technology.
• 💡 These apps could enable users to learn to read by listening and following highlighted text, essentially teaching them to read without formal education.
• 🛍️ Voice computing could allow rural and semi-literate populations to benefit from market economies, shopping online, and accessing educational content.
• 🎓 Learning without a teacher through Socratic dialogue with intelligent agents would help overcome the shortage of teachers in developing countries.
• 💸 The semi-literate market for voice-based apps is estimated to be worth $1 trillion annually, as 3 billion people could use such technology daily.
• 📚 Big tech companies like Microsoft, Google, and Amazon should collaborate to develop voice technologies for orphan languages, benefiting millions of people.

Q & A

What is the main focus of the speaker's passion in the provided transcript?

The speaker's main focus is on voice computing, a technology that enables people to interact with computers using only voice commands, without the need for keyboards or touch interfaces. The speaker believes this technology can have a profound impact on society, especially for semi-literate populations.

Who does the speaker identify as the primary beneficiaries of voice computing technology?

The speaker identifies semi-literate and illiterate people, particularly those in developing countries, as the primary beneficiaries of voice computing. These individuals may not be able to read or use traditional interfaces but can interact with technology through voice commands.

How could voice computing help semi-literate populations, according to the speaker?

Voice computing could help semi-literate populations by enabling them to use technology in various ways, such as reading newspapers, watching foreign language movies, attending online lectures, voting, and shopping online. It would also empower them to access educational resources and market opportunities they are currently deprived of.

What is the estimated market potential of speech-based apps for semi-literate populations?

The speaker estimates the market potential of speech-based apps for semi-literate populations to be worth a trillion dollars per year. This estimate is based on three billion people with an average disposable income of $1 a day.

What are the key technologies required to make voice computing a reality for semi-literate populations?

The key technologies required include speech recognition, translation, speech synthesis, and spoken dialogue systems. These technologies already exist in some form, as demonstrated by platforms like Microsoft's, Facebook's, and Google's translation tools.

Why does the speaker think commercial languages are not the main market for voice computing?

The speaker believes that commercial languages like French, German, English, and Chinese are not the main market because most people who speak these languages are literate and do not need voice computing to access information. The true potential lies in less commercially viable languages spoken by semi-literate populations who can benefit from speech-based technology.

What is the significance of 'one-click' user interfaces for semi-literate populations, according to the speaker?

For semi-literate populations, the user interface needs to be extremely simple and intuitive, ideally requiring only one minute to learn and one click to operate. This reduces the barrier to entry and allows users to quickly and easily interact with the technology using voice commands.

What are some potential applications of voice computing mentioned by the speaker?

Potential applications include reading newspapers aloud, real-time translation and dubbing of foreign language media, online shopping, and education through Socratic dialogue with AI agents. The technology can also facilitate online voting and access to market economies.

What challenges does the speaker mention regarding the development of voice computing for less commercially viable languages?

One challenge is the lack of investment in developing speech technology for 'orphan languages,' which are languages not considered commercially viable. Although the necessary tools and technology exist, there is a lack of willingness from companies to invest in these languages.

What role does the speaker believe governments and major tech companies should play in promoting voice computing?

The speaker believes that local governments and major tech companies like Amazon, Google, and Microsoft should collaborate and share data to promote voice computing for semi-literate populations. Governments may support the development of technology for orphan languages, while companies should work together to avoid duplicating efforts and reduce development costs.

Outlines

00:00

🎙️ The Rise of Voice Computing

In this opening paragraph, the speaker discusses their passion for voice computing, a technology that eliminates the need for keyboards and touch interfaces, relying solely on voice interaction. They explain how this technology can have a profound impact on society, especially for the semi-literate or illiterate population, which numbers around three billion people globally. Voice computing could provide access to technologies that are currently inaccessible due to literacy barriers. The speaker outlines how these people could perform tasks such as reading newspapers, watching educational content, and shopping online through voice commands.

05:01

📰 Voice-Enabled Apps for Literacy and Accessibility

This section explores how voice computing can transform the lives of semi-literate populations by enabling them to access newspapers and other digital content in their local languages. Through real-time speech recognition and translation, apps can help users comprehend and interact with information, potentially helping them learn to read by following highlighted text. The speaker also touches on digital democracy, predicting a future where voting and other democratic processes happen online. They highlight the potential for voice-based apps to help people in rural or underdeveloped regions access global markets and online education.

10:03

🛍️ Socratic Dialogue and the Power of Voice-Based Learning

Here, the focus shifts to the educational applications of voice computing. The speaker introduces the idea of learning without a teacher, where a digital assistant engages users in a Socratic dialogue, helping them learn through conversation. They emphasize the importance of such technology in regions lacking qualified teachers, allowing users to engage in self-directed learning. This has massive market potential, especially among semi-literate populations, which the speaker estimates to be a trillion-dollar opportunity. They reiterate the need for seamless speech-based user interfaces for the widespread adoption of such apps.

15:04

📱 Existing Technologies and the Future of Voice Interfaces

The speaker discusses the current state of speech-to-speech technologies and their presence in platforms like Microsoft, Facebook, and Google. They express disappointment that the biggest market for these tools—semi-literate populations—remains untapped, as companies focus on commercial languages like English and French. The speaker believes that voice-driven apps could become a daily necessity for billions of people, offering opportunities in underserved languages. They stress that the success of such apps hinges on intuitive design, requiring one-click interfaces and minimal learning curves, similar to how one might interact with an assistant or postdoc.

🤖 Voice Computing as the Future of Communication

In this final section, the speaker expands on the architecture and usability challenges of voice computing, emphasizing the need for always-on, autonomic systems that don't drain battery life. They highlight the potential for voice computing to evolve through constant learning and adaptation, much like how social networks and error-correction systems operate today. The speaker underscores the importance of building speech recognition systems for 'orphan' languages, which aren't currently commercially viable but have millions of speakers. They conclude by calling for collaboration between tech giants to support voice computing for these underserved populations, as it could redefine daily life for billions.

Keywords

💡Voice Computing

Voice computing refers to the technology that allows users to interact with computers and digital systems using voice commands instead of traditional input methods like keyboards or touchscreens. In the video, voice computing is seen as a way to bridge the gap for semi-literate populations who cannot use text-based systems, allowing them to access various services and information.

💡Semi-literate Population

Semi-literate populations are individuals who either cannot read or have limited comprehension skills, often due to lack of education. The video highlights the importance of creating voice-based technologies for this demographic, as traditional text-based interfaces are inaccessible to them. Voice computing can empower these individuals to use digital platforms for education, shopping, and communication.

💡Speech Recognition

Speech recognition is the technology that enables a computer or device to identify and process spoken language. In the context of the video, speech recognition is a key component of voice computing, allowing semi-literate users to interact with systems by speaking rather than reading or typing. This technology is crucial for providing access to digital services for people who struggle with literacy.

💡Speech-to-Speech Translation

Speech-to-speech translation involves converting spoken words from one language to another in real time. The video explains that this technology could allow semi-literate users to access content, like news or educational materials, in their own language even if the original content is in a different language. Real-time dubbing of videos and lectures is a specific example given in the transcript.

💡Digital Democracy

Digital democracy refers to the use of technology to enhance democratic processes, such as voting or public participation. The video discusses how voice computing could allow semi-literate populations to participate in online voting or engage with government services, thus empowering them within the digital age. Speech recognition could be used to authenticate users and allow them to cast votes or make decisions online.

💡Killer App

A 'killer app' is a software application so essential that it drives the adoption of the underlying technology. In the video, the speaker argues that voice-based apps for semi-literate populations could be a 'killer app' by being indispensable in everyday life. Examples include apps for reading newspapers, ordering products online, and interacting with virtual assistants, all through voice commands.

💡Learning without a Teacher

Learning without a teacher refers to the concept of self-education facilitated by digital tools and AI-driven systems. The video suggests that voice computing could enable users in underdeveloped regions to learn independently through Socratic dialogues with intelligent agents, making education accessible even in areas with a shortage of teachers. This technology could help semi-literate populations gain knowledge through interactive conversations.

💡Socratic Dialogue

A Socratic dialogue is a method of teaching through asking and answering questions to stimulate critical thinking. In the video, the speaker envisions intelligent agents using this approach to help users learn without formal instruction. The AI would engage in dialogues with users, helping them understand complex subjects by breaking them down into manageable discussions, tailored to the learner's pace and needs.

💡User Interface

A user interface (UI) is the space where interactions between humans and computers occur. The video stresses the importance of designing UIs that are intuitive and easy for semi-literate users to navigate with voice commands. The speaker mentions that these interfaces should allow for 'one-minute learning time' and 'one-click' actions, ensuring that users can quickly learn to use the system without confusion.

💡Orphan Languages

Orphan languages are languages that are not widely supported by commercial technology due to their limited economic value. The video highlights the need for investment in translation and voice technologies for these languages, as they are often spoken by millions of people in semi-literate populations. The speaker argues that despite the commercial focus on major languages, developing technologies for orphan languages could have a profound social impact.

Highlights

Voice computing enables interaction without keyboards or touch, similar to human conversation.

Three billion people live on less than 2.5 dollars a day and are semi-literate or illiterate.

Voice computing can empower semi-literate individuals by allowing them to speak to computers.

Technology like speech recognition, translation, and synthesis can profoundly impact society.

Voice computing can assist in reading newspapers, understanding foreign languages, and accessing educational content.

Real-time translation and dubbing of speeches, like the 2012 TED Talk, is already possible.

Newspaper reading apps could read articles aloud, helping users learn to sight-read.

Digital democracy will likely involve online voting, requiring authentication and authorization technologies.

Online shopping can be revolutionized for rural, illiterate populations through voice computing.

Learning without a teacher is facilitated by AI agents that can explain complex concepts.

The potential market for speech-based apps among the illiterate could be a trillion dollars a year.

Existing speech-to-speech, speech-to-translate, and speech-synthesis technologies are underutilized.

User interface for voice computing must be intuitive, with minimal learning time and one-click actions.

Amazon Echo, Siri, Cortana, and Alexa are examples of existing voice computing technologies.

Orphan languages, not commercially viable, need investment for translation technologies.

Architectural issues for voice computing include non-intrusiveness, autonomic operation, and always-on learning.

Deep learning is essential for voice computing, but it's not the only technology needed.

Collaboration between tech giants like Amazon, Google, and Microsoft could unlock the potential of voice computing for all.

The semi-literate population is a major untapped market for AI technology and voice computing.

Transcripts

[Music]

so the first part of the first hour is

supposed to cover the most recent

interesting aspects of AI technology and

I'd like to tell you about my passion

which is which is now reaching the point

where I think it can have a profound

impact what what I call voice computing

and voice computing is when your cannot

you can no longer use keyboards can no

longer use touch the only thing you have

is voice just like when you are

interacting with other people like your

postdocs and your assistants that's all

you have and in which case how does that

proceed and what can happen and it turns

out we now have all the relevant

technologies and I like to go through

that and if we do do the create such a

voice computing it can have profound

impact on the society as a whole so let

me take you through it so the bottom of

the pyramid represents about three

billion people who have less than 2.5

dollars of income every day and these

are what I call semi-literate many of

them are illiterate they can't read any

language that's part of that look also

called

literacy divide and there are others who

can read but they really cannot

understand comprehend and so I call both

of these groups together semi-literate

if you're a semi-literate person on this

planet the only way you can use the

computers is to speak to them you can't

read what is on the screen

you can't you know understand you can't

follow but but on the other hand these

three billion people lead perfectly

normal lives you know and so the

question is should they be deprived of

our technology if their knife not how

can they use it and the only way they

can use it is to voice computing it

turns out and by voice computing imagine

for you in your lab you're you know

you're a professor and you have a

postdoc and you have a secretary you

just tell them what to do and they go do

it if they can't do it they come back

and say I'm sorry I can't do it did you

mean this or not they have a

clarification dialogue and I imagine

computing and computer a intelligent

agents being exactly like that you don't

tell them all sorts of unnecessary

information and the question is are we

there the other questions what do we

have and what do we need to do to get

there so why is computing where there's

no keyboard attached can help these

semi-literate people in all kinds of

ways that you may not have thought about

they can help you to read newspapers

they can help you read foreign language

movies listen to Khan Academy lectures

vote online and do shopping online and

the interesting aspect of the all of

these is to have a speech recognition

translation and synthesis technologies

speech to speech and having apps that

can enable different aspects of it so

for example a mobile app for

entertainment at education it would have

to be able to understand speech and and

for example YouTube you can actually get

transcriptions of what is being said

then you have to translate it and then

do the speech synthesis all in real-time

you so basically you've heard about

dubbing

in movies automatic dynamic real-time

dubbing is possible today in fact in 9

2012 wreck crashes gave a talk in

English in in in China and it got

translated into Chinese and synthesized

into Chinese all in real time in that

particular case it was kind of a domain

that was understood and they've tried it

and it was but I it's not yet ready for

primetime it should be I don't know why

not and but I think it will be there

soon enough and the same thing if you

have such technology if I can't read a

newspaper I can say I want to read this

newspaper in my local language I see

some quicker I can't read this screen

I can't touch on it and say but I can

touch and say there's a nice picture

what is this headline about it will read

the headline and now you can go and say

can you now read me the rest of the

story as they go to the next so the

whole issue of designing an app and a

newspaper reading app that will read to

you literate people so suddenly they all

can not only the one of the interesting

side effects of it is they will learn to

sight-read they ever see see the words

and if you do the proper interface it

highlights the words as we are reading

then you essentially learn to read

without having to be taught how to read

in subsets so one of the areas I work in

is digital democracy

I believe we are entering a phase where

almost all countries in the next 50

years will be avoiding online and

different countries have different

constitutional requirements so that may

or may not happen soon but

authentication authorization audit

technologies we understand fully well

and it will be done and the same is true

with respect to online shopping

essentially it empowers a rural

illiterate person to get the same

benefits of market economies that the

economies now and benefit from from

Amazon and so on you're getting the best

product at the lowest possible price and

Amazon is getting to one day and for for

our delivery times and I believe they

will solve that same problem in rural

India and rural Africa so that you'll be

able to do the same thing and learning

without a teacher is perhaps the most

interesting app that happens with this

because we're getting more and more

online content and if I can have a

Socratic dialogue with an agent which

knows a lot more than I do and which

would explain to me when I don't

understand otherwise I'm reading so the

whole idea of learning without a teacher

because there are not enough good

teachers especially in the third world

countries the two you know kind of give

you the opportunity so given if all of

these things are real

I claim they're not only real there near

term then the illiterate populations

will be the biggest source of customers

for speech base up apps in the future

and I estimate that to be a trillion

dollars a year how did I get that number

three billion times $1 a day which is

the disposable income is essentially

three billion dollars a day times 365

days is where you get the trillion

dollars from ok so there is not to be

sneezed at so the question is what

technology exists speech to speech

speech in translate speech out into

another language already exists in

Microsoft Facebook Google I haven't seen

the Google version yet but they have the

best language translation capability of

anybody but I'm sure all of them have it

this point there but the unfortunate

thing is most of them mistakenly think

the biggest market are the commercial

languages like French German English

Japanese and Chinese but they're most

people are literate they don't need the

technology that I have a Translate app

on my command I phone I never use it

because I what if we can't read it right

then you use it every day if you use it

every day that is the killer app if

three billion people are using it you

can't think of any better killer app

than that right so there is a mistaken

understanding of the market

opportunities but I think it will be

fixed itself so the apps tailored to

semi-literate populations of the kind

that I told you will have a profound

impact but the user interface is very

important it has to be one minute

learning time one-click not three clicks

to do something and you know I said two

clicks here sometimes you have to say

what do you want to doing and then do it

so you may need to install spoken

dialogue there's no keyboard there's no

tax no you know everything has to be

speech only imagine your assistant or

imagine your postdoc and that's what

you'll be doing here if you can tell

that your postdoc to do something and

then this app should do exactly the

kinda same behave the same way and all

these apps will require speech

recognition spoken dialogue switch to

speech translation and all of those

technologies exist already question

answering dialogue already exists so if

I take you know I don't know how many of

you have used Amazon echo Alexa agent I

created another agent called Asha it

turns out in order for the systems we

create you know quickly recognized it

has to be a two syllable name there

should be a fricative end in between

within with that it's easier to reliably

recognize

so inertia is a common name in many of

the many cultures so you can do say our

shop Amy Shakespeare Hamlet and it'll

play too in local language you can say I

shall read me the Eenadu newspaper it'll

do that or you can say I shall order me

eggs and bread and whatever it'll get

delivered to you say you know the same

day or next day and I shall you know I

want to talk to my grandson in Seattle

and it'll do that and I shall my mobile

device with a thousand rupees or

whatever and again all these things are

routine and they are already being done

you know if you take Siri or Cortana

Alexa they're all doing it and so the

question is why can't we do it for the

orphan languages what we need our orphan

you know languages there are many

languages that are not considered

commercially viable

therefore it turns out to build a

translation afternoon to Marathi

you need to understand Marathi you need

to have a database all of that requires

certain amount of investment the tools

exist the technology exists they just

haven't been done for these languages

and when I talk to them everybody agrees

is a good idea but nobody wants to step

up to the plate right so anyway there's

a whole set of architectural issues I

won't go into here but it has to be non

intrusive autonomic that is if you have

a half- on your iPhone you have to tap

it this has to be always on and always

working and always learning and if it's

always on you run out of battery very

quickly therefore you need either a

desktop one like Amazon echo or some

other new technology we haven't invented

yet and it monitors analyzes and learns

from experience and it shares data

suitably anonymized

so that it's learning preferences by

observing other similar people's choice

through social networks learning tasks

similarity and uses similarity learning

by error correction and simply learning

by through clarification by walk so none

of these are what you we now call deep

learning which we'll talk about

but deep learning is an essential part

of the Cystic knowledge e that has been

invented for speech so I think what

we'll see is that the implications of

all of these so no single company seems

to want to make investments in these

orful languages but there are many

languages that more than twenty million

people speak so the market size is

reasonably big so we ought to be able to

look at these languages and so for every

for a literate population it will become

a life mine and use every used every day

that's what is the definition of a

killer app if you touch it every day

then you have an app that makes sense I

have a lot of apps on my iPhone I have

not touched even one thing ever in a

year so then it doesn't matter

in conclusion two billion semi-literate

population two billion plus in the world

are a major untapped market and AI

technology stretched pressure especially

effective use of speech technology and

voice computing is the only option only

option to support their needs

and we have all the needed tools and

technology and if there's an issue of

foreign languages costs associated with

it local net governments will probably

be willing to support and here is one

area where Amazon Google Microsoft

should all collaborate share the data

there's no nothing magical quick they'll

all collect the same stupid data and I

know they're all kind of proprietary we

should not keep them some preparatory we

should kind of shed it

so with that thank you

[Applause]

[Music]

you

[Music]