Here's "Brian" - their English, male, received pronunciation voice - reading John Betjeman's poem "Slough":

The pronunciation of all the words is incredibly lifelike. If you heard it on the radio, it might sound like a half-familiar BBC presenter. It has a calm, even tone which suits the poem splendidly.

The rhythm is also spot on. That's mostly a function of the short lines and helpful punctuation the poem contains. Much like iambic pentameter, or a limerick, the syllables lend themselves to a specific and identifiable cadence.

But the emphasis is all wrong. The poem just... ends. There's no sense of finality in the tone. You'd expect a competent reader to recognise "tinned minds" as being worthy of stressing. Polly does have some capability to mark specific words for emphasis, but it's all very manual.

There's no synthetic emotion. Do you feel the rage, desperation, sadness, hopelessness of the poem? While Polly has some SSML (Speech Synthesis Markup Language) support - the range of emotions it can express are severely limited. And, again, must be applied manually.

"I used to be an adventurer like you, but then i took an arrow in the knee!"

One of the reasons stock phrases pop up so often in video games is that it is expensive to write and record thousands of different lines of dialogue.

We're almost at a stage where a computer can procedurally generate lines for background characters to speak, and then "record" an audio version in an array of styles. No more expensive voice actors, no more memetic references for in-group homophily. Each player of a game will have a completely different dialogue experience.

But the bit that we're still missing is the automation of emphasis and emotion and comic timing and understatement and... all the things which trained actors spend years learning how to do successfully.

In 2011, the film critic Roger Ebert had surgery which eliminated his voice. He proposed the following "Ebert Test" for synthetic voices:

If the computer can successfully tell a joke, and do the timing and delivery, as well as Henny Youngman, then that’s the voice I want.

We're so close, I can taste it. The Turing Test for realistic voices is whether they can move the audience to tears with poetry.

If you've not read Alan Turing's The Imitation Game, I can highly recommend it. The paper is short, well written, and contains a whole world of ideas.

This is where we get the concept of the Turing Test. Can a human be fooled into thinking that the computer they are communicating with is a human? It is often assumed that the communication must be typed, but I don't think that's necessarily the case.

In order that tones of voice may not help the interrogator the answers should be written, or better still, typewritten. The ideal arrangement is to have a teleprinter communicating between the two rooms.

Alan Turing (October 1950), "Computing Machinery and Intelligence", Mind, LIX (236): 433–460, doi:10.1093/mind/LIX.236.433

In this part of the paper, Turing is writing about what the existing imitation game is. He says nothing about how the game should be played with a computer.

Turing was aware of the current limits of technology.

In the early 1950s speech recognition was in its infancy, so it is perhaps understandable that the idea of talking to a computer was infeasible.

Is there any reason why, today, the questions and answers couldn't be spoken aloud?

Am I being heard by a human?

It can't have escaped your attention that listening devices are popping up in all sorts of places. From telephone banking to home assistants, computers are listening to us and (mostly) understanding what we are saying.

I've played with cheap crowd-funded computers which can hear and transcribe text instantly.

We're now safely at a point where a computer can understand clearly spoken text as accurately as a human.

Is this a human voice?

The history of speech synthesis goes back centuries - although the results were little more than the whinges of cacophonous bagpipes.

But by the early 1940s, electronic speech was a reality:

(There is a longer demonstration available.)

Today, the latest artificial intelligence research from Google has produced WaveNet - which claims to produce the most lifelike computer generated speech.

It's not exactly compelling, but it is getting there.

Aside from determining if the voice is generated organically, there is another aspect - timing and intonation. In 2011, the film critic, Roger Ebert proposed his "Ebert Test"

If the computer can successfully tell a joke, and do the timing and delivery, as well as Henny Youngman, then that’s the voice I want

You can listen to Ebert's synthesised voice in his TED talk.

Could you sit in the back of a taxi and make conversation with a computer? At present, it would probably understand you. And you would understand it easily.

Is This Car Being Driven By A Human?

As Turing's paper goes on, he talks about the various rules that a computer may need to be aware of.

By "rules of conduct" I mean precepts such as "Stop if you see red lights," on which one can act, and of which one can be conscious.

An AI might provider a superior ride to a human. Should we build our AIs to imitate our flaws?

Does this smell like a human?

What about our non-obvious senses? Back in the 1950s, Turing wrote:

No engineer or chemist claims to be able to produce a material which is indistinguishable from the human skin. It is possible that at some time this might be done, but even supposing this invention available we should feel there was little point in trying to make a "thinking machine" more human by dressing it up in such artificial flesh.

Nowadays, the quest for artificial flesh is driven by lascivious desires - but again it raises an interesting point. Would you be able to tell that your taxi driver was a robot if they smelled of sweat rather than WD-40? If they got goosebumps during a breeze?

At the moment, our attempts to flesh out robots hit the "uncanny valley" - a visceral reaction to something which isn't quite human enough to be convincing.

A good example of the uncanny valley is this almost-human-yet-really-creepy robot of the actor Scarlett Johansson.

It's is hard to identify what precisely makes the robot look so wrong. The pose is a little off, the skin tone is not quite right, the eyes are... I can't put my finger on it.

To experience the full terror, you can watch the video about the robot's construction. Warning you may find this disturbing. Also it contains (robot) nudity which... I dunno... feels like it ought to be NSFW.

Is this car as ethical as a human

The brakes on the car fail while you are driving down a mountain. The driver can either crash into a bus of school children - killing them but saving you, or can drive over the side of the mountain killing you but saving the children. What should your driver do?

The "trolley problem" is a classic of moral philosophy. How does a human decide who lives or dies? Can we teach a robot to have ethics? That's what MIT are trying to understand.

Welcome to the Moral Machine! A platform for gathering a human perspective on moral decisions made by machine intelligence, such as self-driving cars.
We show you moral dilemmas, where a driverless car must choose the lesser of two evils, such as killing two passengers or five pedestrians. As an outside observer, you judge which outcome you think is more acceptable. You can then see how your responses compare with those of other people.

Moral Machine - MIT

Should an autonomous car prioritise the life of its owner? Would you expect a human taxi driver to do so?

Is "Being Human" The Goal?

For some tasks - humans rule. Conversation, pattern recognition, falling in love, creating art.

But humans suck at a lot of things. We're fragile, imprecise, impatient, we smell, we cut corners, and we carry diseases, our morals are dubious, and we tire easily.

I'd argue that in many uses of artificial intelligence, Turing's test sets an artificially narrow limit. Do I want an autonomous vehicle which is indistinguishable from a crabby, distracted, unintelligible, malodorous, and immoral taxi driver?

Achieving parity with humans seems like a low bar for machines.

It emerged this morning^* that the Guardian newspaper has realised that the way it writes is unsuitable for the web.

^*Source: Guardian Newspaper, 18/11/2011, page http://www.guardian.co.uk/media/mind-your-language/2011/feb/18/mind-your-language-day-date-time

By using non-specific language, I have introduced a degree of ambiguity which makes it hard for reader - both in the present day and the future - to understand the ideas I am trying to convey.

For example - the above text doesn't state which of the many Guardian newspapers is under discussion. The words "this morning" are highly subjective depending on timezone - and the date at which the article is read. Finally, the source link is separate from the text making it hard for an automatic process to understand what words relate to which website.

The above could be rewritten as...

The Guardian has realised that the way it writes is unsuitable for the web.

We can, of course, extend our markup to make it easier for both humans and robots to understand what it is we have written. Using the ideas behind the Semantic Web we can embed information about the relationships, dates, times, locations, etc, in a meaningful way.

Page Numbers

This leads me into the bizarre decision of Amazon to introduce page numbering for its Kindle eBook reader.

I hope this post will convince you that this move is philosophically wrong and - potentially - dangerous.

Why Do Paper Books Have Page Numbers?

Paper books (not "real books" as some people term them) are physically printed on individual sheets of paper. Page numbering is a limitation with non-digital media to allow semi-rapid and predictable access to pseudo-exact locations.

For example, a paper book may say

"For further information on the Infinite Improbability Drive, see Adams 1979, page 42"

It doesn't tell us where on a page unless it also introduces a paragraph number. As a paragraphs aren't usually numbered, this makes locating the exact reference hard for both machines and people.

Furthermore, different versions of the same text may have substantially different formatting - larger print, more images, smaller paper, etc - so every time a book is republished all page references have to be recalculated.

Page numbers are a hack. They are an illusion designed to give humans the impression that an imprecise reference is specific.

Why Don't eBooks Have Page Numbers?

Because they don't need them! An eBook is a digital document which may be marked up in a many ways. From the humble text document as an unbroken stream of characters - to a meta-data rich ePub full of syntactic markup.

An eBook never knows the size of the screen upon which it will be displayed. Even if it did, it wouldn't know how large its text would render, nor what font would be used, nor if the screen orientation were to change.

The idea of a page number for a digital document is an absurdity. Page numbers are only needed for an item that may be printed out in the physical world. Even then, they still suffer from the above problems.

Why Use Page Numbers In Quotes?

The idea is simple. To use the Harvard Style as an example - other styles are equally useless - we see this.

Lawrence (1966, p.124) states "we should expect..."

So, assuming we can find the exact copy of "Lawrence's" paper, printed in exactly the same font and size, on exactly the same shaped paper, and typeset identically, we can simply flip through to page 124 and scan through the page until we find the quote we're looking for.

Rubbish! Utter, unmitigated, total and utter rubbish. A long, boring and error-prone process which does little to help us find the information for which we are looking.

The Role Of Computers

We have computers for one simple reason - they perform boring jobs with speed and accuracy.

If I want to find where Alan Turing declared "It is possible to invent a single machine which can be used to compute any computable sequence" within "On Computable Numbers, with an Application to the Entscheidungsproblem" I can do one of two things.

1. I can click on a link which takes me straight to the sentence or chapter
2. I can hit "search" or "find" in my document viewer

Having documents which are correctly marked-up is the key to successful linking of data. Manually marking up a document is tiresome and problematic for humans. Where we fail, computers make up for our shortcomings with their incredible speed at searching through documents.

My Kindle is a fairly slow as far as modern computers are concerned. Yet it has no difficulty searching through millions of words in hundreds of documents to find the single sentence I'm looking for. And it all takes less time than it would for me to flick through some pages.

A Changing World

Page numbering as a system of referencing relies on such an unlikely series of events as to be worse than useless. If both reader and writer do not have identical copies of the quoted work, there is a real risk of confusion and misunderstanding.

Page numbers are an ugly hack which fail to achieve the precision they so desperately crave. Let's remove the hack and replace it with something which actually works.

People who quote a page number from an electronic book don't understand how the modern world works. Documents have evolved - so must our citation styles.