AI voiced Skyrim

BrotherMaynard

There has already been some discussion about ways AI could advance game development in areas other than the traditional visual quality and here's one that goes right in the footsteps of the earlier ChatGPT discussion.

Skyrim, being such a rich playground for modders, just got its Dragonborn voice over mod, which adds AI generated voice to the previously silent protagonist. Here is a short sample (for now there is only female voice available):



You can play with this tool at their website, add whatever (English) text you want, select various voice presets and listen to the result. In future there could be all kinds of voice types, accents, tones, moods, etc., it will be interesting to watch how the system develops.

So now we can have quest texts generated by the AI and now also voiced by the AI. A few more steps and we'll probably be able to generate entire games using AI toolsets... But joking aside (or perhaps it's not a joke at all), we seem to be entering a new deep transformation in gaming (and much more, but that's not what this site is about).

Here, have a listen to their AI-narrated excerpt from The Great Gatsby.

Andemnon

Given how deep fakes are increasing in complexity etc. it would not surprise me at all, instead of using a particular actor/actress voice, you could use something 'similar' but different enough to not infringe on copyright maybe, no, the true revolution is when players can choose who they want various characters to sound like in a game, after all we all have favourites, and i wouldn't mind having a couple of npc's in Skyrim etc. to sound like Jessica Alba or Halle Berry 

Vrika

AI has improved TTS tech (Text To Speech) somewhat, but it's actually a tech that's been widely available since 1980s. Nowadays Windows includes its own TTS that produces pretty good results, at least for English language. It's mostly used by people with visual impairment, but there are also some ebook readers and visual novel game engines which have a feature to turn the text into speech and read it aloud.

Games don't usually use TTS because it can't read between the lines what emotion should be added to the speech, and because as long as you don't hire any big names voice actors are cheap. But the tech to turn any text in games into speech is there and has been already for some time, it's just a matter of whether the devs bother to make their game use it.

Sovrath

I've seen that. Great addition though, for me, I prefer a silent protagonist. I absolutely hate in the gothic games that they give your character a voice.

You choose a sentence and then hear that sentence. Gets old after the very first moments for me.

I should add, I found eleven labs voice AI and might play around with it for a mod I'm putting together.

MaxBacon

AI voiced lines should happen soon for large scale NPC voicing in games I'd imagine, as it does run in the client.

Now the thing shown with M&B mod using ChatGPT to having conversations with NPCs that's a whole can of worms as the simulation costs of each query are insane, and have to happen in the cloud, so that is not going to be a thing anytime soon unless we figure out the simulation costs of these things.

Theocritus

Acutally I could do without voice acting and cutscenes....

Quizzical

This is roughly what Vrika said, but speech to text isn't new.  AI will probably make it possible to customize voices quite a bit, with accents, higher or lower pitch, and so forth.  It will also be able to mimic some particular human's voice.  What AI cannot and will not be able to do is to adjust the cadence and express emotion the way that humans commonly do when we talk.  So basically, AI can generate bad voiceovers from text, but not good ones.

It will probably be possible for programmers to manually make that sort of adjustment to make AI-generated voiceovers express emotion properly, but it's going to take a lot of manual tweaking.  If you want good voiceovers, it might well be easier and cheaper to just hire a human voice actor.  Even if a programmer can get a computerized voice to read a line just right after five minutes of tweaking, a human voice actor can do that and move on to the next line a lot faster.

Mendel

AI may have pluses and minuses, but I certainly wouldn't lump text readers in the AI bundle.  So, any thing you might infer from an AI reading text isn't likely to be an accurate indication of the capabilities of AI.

In the context of Skyrim and other RPGs, the voice reader could be just another character customization option for MMORPGs.  Coupled with good writing (and voice overs) for NPCs, the character's voice provided by incorporating a text reader could make a world seem more alive.  It could really help immersion.

As long as there was a 'Disable All Voices' option for when I just want it quiet.

Uwakionna

AI voiceover tools like this do have tag systems usually for you to highlight parts of text and give the AI cues for inflection/expression. That is all has to be done as a manual editing pass over the text for the AI to utilize.

This is handy for some things in my mind, but wouldn't really change much in general. Like Hogwarts Legacy insisted in a player voiceover, but they were limited to a single male and female voice with a pitch shift slider that...kinda sucks and makes the voices break.

If the voiceovers had any big range of expression I might argue against it, but the voiceover in that game already is pretty calm even at the most emotive moments. All in all, you could use a set of AI voices to add more choice to a game like that, and have it arguably come out better than the weird tinny pitch shifted options the game provides.

But it's not really doing that much more for a game in and of itself.

mekhere

I can only imagine what this would lead to lol. A real-life gamer who is group leader, sitting in a discord VC chat and having a real conversation with AI robots who are his teammates. Oh, how fun would that be to listen to.

Iselin

Uwakionna said:

This is handy for some things in my mind, but wouldn't really change much in general. Like Hogwarts Legacy insisted in a player voiceover, but they were limited to a single male and female voice with a pitch shift slider that...kinda sucks and makes the voices break.

Yeah, I noticed that too and then went on Reddit to find out it's a known issue... or is it a feature?

BrotherMaynard

Vrika said:

AI has improved TTS tech (Text To Speech) somewhat, but it's actually a tech that's been widely available since 1980s. Nowadays Windows includes its own TTS that produces pretty good results, at least for English language. It's mostly used by people with visual impairment, but there are also some ebook readers and visual novel game engines which have a feature to turn the text into speech and read it aloud.

Games don't usually use TTS because it can't read between the lines what emotion should be added to the speech, and because as long as you don't hire any big names voice actors are cheap. But the tech to turn any text in games into speech is there and has been already for some time, it's just a matter of whether the devs bother to make their game use it.

Of course it's not new, as you wrote, it's been around for a few decades in some form. But the quality of this new approach and ease of use versus the old tech are not comparable at all.

True, their showcases will favour texts that are especially suitable for this tech demo and will mostly aim for a narrative character, but even with that caveat, you must admit the AI voice is very convincing. Have a listen to their samples or just paste in there random text from this discussion, select voice and listen. The flow sounds very natural, intonation, melody, stress - it's all there.

Now, the added value is also the apparent simplicity of its implementation in games, as we can see with this use of a simple plugin for Skyrim.

As others wrote, once they will develop it into a form that can work with different moods and emotions and adapt the voice accordingly, this will be big.

Voice acting is a huge expense both financially and in terms of time and logistics needed to get it done. Hundreds of hours spent in studios and audio booths for every bigger narrative game is a major expense.

I am not saying it will go away. But the way voice acting is done will most likely change. Small devs and indie studios will have the possibility to quickly add hours of dialogue for dozens of characters just by writing the lines. Big studios will probably keep using big voice talents for the main roles, while AI can populate the game worlds with nearly endless background conversations. Just imagine RPGs like the Witcher with Novigrad full of hundreds of people in the background chatting, conversing, arguing - not to mention MMORPGs, which are ideal for this, given their naturally sprawling and persistent world design...

Scot

I think we need to put some thought into what is happening here, because if we don't the AI's will soon enough.

Throughout human history "tools" have replaced human jobs, the loom put a lot of seamstresses out of work. This tool, AI is often thought of as eventually leading to mass unemployment or the death of human kind. That's the typical negative scenario, so overplayed in SF but I want to raise a different one.

If they can now compose music, draw and write somewhat well, how long before they can do that well consistently? Think of the speed they are capable of, once they are consistently good how long would it take them to put out a library of fiction and reference works equal to everything humanity has written so far? Sure it would only be good not great, but then then most of what is already published is not great and who says one day great works would not be possible?

With the aid of 3d printers the only thing stopping them putting out say a good painting for every one ever created would only be resources and so on for the rest of art.

The unemployment issue has this idea that we will still do more intellectual work, there will always be something for us to do particularly creatives. But that is just another step for the AI to master, no job is safe. No achievement humans might attempt could end up being done just as well as by an AI.

As societies we rarely think about why we are here, what the aim of our societies should be and what the purpose of existence is. Which is just as well, because if we were inclined that way we could be facing an existential crisis, what would be the point of humanity anymore if the AI we created does everything as well as we do?

Being social media stars may be all that is left to us, what a wonderful world.

Quizzical

Scot said:

I think we need to put some thought into what is happening here, because if we don't the AI's will soon enough.

Throughout human history "tools" have replaced human jobs, the loom put a lot of seamstresses out of work. This tool, AI is often thought of as eventually leading to mass unemployment or the death of human kind. That's the typical negative scenario, so overplayed in SF but I want to raise a different one.

If they can now compose music, draw and write somewhat well, how long before they can do that well consistently? Think of the speed they are capable of, once they are consistently good how long would it take them to put out a library of fiction and reference works equal to everything humanity has written so far? Sure it would only be good not great, but then then most of what is already published is not great and who says one day great works would not be possible?

With the aid of 3d printers the only thing stopping them putting out say a good painting for every one ever created would only be resources and so on for the rest of art.

The unemployment issue has this idea that we will still do more intellectual work, there will always be something for us to do particularly creatives. But that is just another step for the AI to master, no job is safe. No achievement humans might attempt could end up being done just as well as by an AI.

As societies we rarely think about why we are here, what the aim of our societies should be and what the purpose of existence is. Which is just as well, because if we were inclined that way we could be facing an existential crisis, what would be the point of humanity anymore if the AI we created does everything as well as we do?

Being social media stars may be all that is left to us, what a wonderful world.

The way that machine learning works is to mimic training data.  Humans manually make a bunch of decisions, then machine learning takes some other situation and scores the possibilities to try to mimic the decision that a human would have made.  But the attempt at guessing will never be better than the original, and will often be much worse.

That sort of mimicry can never surpass or even catch up to what humans can do outside of very simple situations.  What it can sometimes do is to be close enough while being much cheaper.  That, for example, allows for search engines to quickly give you good enough results.  That's critical when high volume is essential.  It's far less useful when humans can readily produce ample or even excessive volume on our own.

If you had your choice between one great novel written by a human or any one of a trillion mediocre novels written by AI, a whole lot of people will be interested in the one great novel.  The only real reason to care about any of the AI novels is if it's highly customized to suit your personal tastes and you value that more than the various stylistic ways that the AI novel is inferior.

Another problem with AI is that it makes a lot of glaring mistakes that a competent human would not.  That's really not a fixable problem, either, as it's intrinsic to how the algorithms work, at least in complicated situations.  That's acceptable for some things, as if a search engine returns six things that you're interested in among its top ten results, having a few that are completely irrelevant isn't a problem.  But it's a crippling problem when being right 99.9% of the time isn't good enough.  That's why, for example, self-driving cars aren't going to be mainstream anytime soon.

The worry about jobs going away has been a perpetual one for centuries.  Many particular jobs did go away, but plenty of new jobs arose in their place.  AI will just be the latest iteration of this old phenomenon.  There will probably be a quite a few jobs of humans working to improve AI by providing better training data.  There will almost certainly be a lot of jobs of AI assisting humans where the human acts as an editor to catch and fix the mistakes that AI makes.

Scot

Quizzical said:

Scot said:

I think we need to put some thought into what is happening here, because if we don't the AI's will soon enough.

Throughout human history "tools" have replaced human jobs, the loom put a lot of seamstresses out of work. This tool, AI is often thought of as eventually leading to mass unemployment or the death of human kind. That's the typical negative scenario, so overplayed in SF but I want to raise a different one.

If they can now compose music, draw and write somewhat well, how long before they can do that well consistently? Think of the speed they are capable of, once they are consistently good how long would it take them to put out a library of fiction and reference works equal to everything humanity has written so far? Sure it would only be good not great, but then then most of what is already published is not great and who says one day great works would not be possible?

With the aid of 3d printers the only thing stopping them putting out say a good painting for every one ever created would only be resources and so on for the rest of art.

The unemployment issue has this idea that we will still do more intellectual work, there will always be something for us to do particularly creatives. But that is just another step for the AI to master, no job is safe. No achievement humans might attempt could end up being done just as well as by an AI.

As societies we rarely think about why we are here, what the aim of our societies should be and what the purpose of existence is. Which is just as well, because if we were inclined that way we could be facing an existential crisis, what would be the point of humanity anymore if the AI we created does everything as well as we do?

Being social media stars may be all that is left to us, what a wonderful world.

The way that machine learning works is to mimic training data.  Humans manually make a bunch of decisions, then machine learning takes some other situation and scores the possibilities to try to mimic the decision that a human would have made.  But the attempt at guessing will never be better than the original, and will often be much worse.

That sort of mimicry can never surpass or even catch up to what humans can do outside of very simple situations.  What it can sometimes do is to be close enough while being much cheaper.  That, for example, allows for search engines to quickly give you good enough results.  That's critical when high volume is essential.  It's far less useful when humans can readily produce ample or even excessive volume on our own.

If you had your choice between one great novel written by a human or any one of a trillion mediocre novels written by AI, a whole lot of people will be interested in the one great novel.  The only real reason to care about any of the AI novels is if it's highly customized to suit your personal tastes and you value that more than the various stylistic ways that the AI novel is inferior.

Another problem with AI is that it makes a lot of glaring mistakes that a competent human would not.  That's really not a fixable problem, either, as it's intrinsic to how the algorithms work, at least in complicated situations.  That's acceptable for some things, as if a search engine returns six things that you're interested in among its top ten results, having a few that are completely irrelevant isn't a problem.  But it's a crippling problem when being right 99.9% of the time isn't good enough.  That's why, for example, self-driving cars aren't going to be mainstream anytime soon.

The worry about jobs going away has been a perpetual one for centuries.  Many particular jobs did go away, but plenty of new jobs arose in their place.  AI will just be the latest iteration of this old phenomenon.  There will probably be a quite a few jobs of humans working to improve AI by providing better training data.  There will almost certainly be a lot of jobs of AI assisting humans where the human acts as an editor to catch and fix the mistakes that AI makes.

I think people are too worried about AI, but I also think the scenario I mentioned could happen. You are focusing too much on what AI is capable of doing now, not where it is going. The recent advances in AI writing should not even be needed to prove it is pushing ahead, that's just the latest in many leaps.

Recently our Chancellor used Chat AI to write the introduction to his speech, as you can expect it got headlines about "who needs politicians"? What they didn't ask was who needs song writers when these chat AI can write a song in the style of a given songwriter. From what I understand it was a good introduction but the artist whose song writing style was copied said it was rubbish, but then he would wouldn't he?

I think you are right about humans always being able to get a supervisor/editing role, the question is for how long?

[Deleted User]

The user and all related content has been deleted.

Quizzical

Scot said:

Quizzical said:

Scot said:

I think we need to put some thought into what is happening here, because if we don't the AI's will soon enough.

Throughout human history "tools" have replaced human jobs, the loom put a lot of seamstresses out of work. This tool, AI is often thought of as eventually leading to mass unemployment or the death of human kind. That's the typical negative scenario, so overplayed in SF but I want to raise a different one.

If they can now compose music, draw and write somewhat well, how long before they can do that well consistently? Think of the speed they are capable of, once they are consistently good how long would it take them to put out a library of fiction and reference works equal to everything humanity has written so far? Sure it would only be good not great, but then then most of what is already published is not great and who says one day great works would not be possible?

With the aid of 3d printers the only thing stopping them putting out say a good painting for every one ever created would only be resources and so on for the rest of art.

The unemployment issue has this idea that we will still do more intellectual work, there will always be something for us to do particularly creatives. But that is just another step for the AI to master, no job is safe. No achievement humans might attempt could end up being done just as well as by an AI.

As societies we rarely think about why we are here, what the aim of our societies should be and what the purpose of existence is. Which is just as well, because if we were inclined that way we could be facing an existential crisis, what would be the point of humanity anymore if the AI we created does everything as well as we do?

Being social media stars may be all that is left to us, what a wonderful world.

The way that machine learning works is to mimic training data.  Humans manually make a bunch of decisions, then machine learning takes some other situation and scores the possibilities to try to mimic the decision that a human would have made.  But the attempt at guessing will never be better than the original, and will often be much worse.

That sort of mimicry can never surpass or even catch up to what humans can do outside of very simple situations.  What it can sometimes do is to be close enough while being much cheaper.  That, for example, allows for search engines to quickly give you good enough results.  That's critical when high volume is essential.  It's far less useful when humans can readily produce ample or even excessive volume on our own.

If you had your choice between one great novel written by a human or any one of a trillion mediocre novels written by AI, a whole lot of people will be interested in the one great novel.  The only real reason to care about any of the AI novels is if it's highly customized to suit your personal tastes and you value that more than the various stylistic ways that the AI novel is inferior.

Another problem with AI is that it makes a lot of glaring mistakes that a competent human would not.  That's really not a fixable problem, either, as it's intrinsic to how the algorithms work, at least in complicated situations.  That's acceptable for some things, as if a search engine returns six things that you're interested in among its top ten results, having a few that are completely irrelevant isn't a problem.  But it's a crippling problem when being right 99.9% of the time isn't good enough.  That's why, for example, self-driving cars aren't going to be mainstream anytime soon.

The worry about jobs going away has been a perpetual one for centuries.  Many particular jobs did go away, but plenty of new jobs arose in their place.  AI will just be the latest iteration of this old phenomenon.  There will probably be a quite a few jobs of humans working to improve AI by providing better training data.  There will almost certainly be a lot of jobs of AI assisting humans where the human acts as an editor to catch and fix the mistakes that AI makes.

I think people are too worried about AI, but I also think the scenario I mentioned could happen. You are focusing too much on what AI is capable of doing now, not where it is going. The recent advances in AI writing should not even be needed to prove it is pushing ahead, that's just the latest in many leaps.

Recently our Chancellor used Chat AI to write the introduction to his speech, as you can expect it got headlines about "who needs politicians"? What they didn't ask was who needs song writers when these chat AI can write a song in the style of a given songwriter. From what I understand it was a good introduction but the artist whose song writing style was copied said it was rubbish, but then he would wouldn't he?

I think you are right about humans always being able to get a supervisor/editing role, the question is for how long?

I'm talking about AI at a level of things that something much like current technology could plausibly do, even if they haven't yet been done.  That's very different from the sort of thing that science fiction writers can dream up.  A dystopia where AI replaces all jobs and humans can't be productive is firmly in the latter.

AI can do mimicry, but not novelty.  If you ask an AI to write something similar to what it already has many examples of, it might well do a respectable job.  For example, it wouldn't surprise me if ChatGPT gave a decent output if you asked it for a patriotic speech that an American mayor might deliver on the fourth of July.  Even if ChatGPT can't do that quite yet, some future AI probably will.

But if you want AI to write the speech making the case for some new initiative, the output is going to be a disaster unless you end up with a "stone soup" style of speech where you have to basically feed the AI all of the points that you want to make in the speech.  At that point, the AI that "writes" the speech barely does any more than a word processor.

Quizzical

Getting back on topic, it might be possible for AI in the near future to turn one voice into another.  That way, you only need one voice actor to give lines the proper cadence and emotion, and then the AI turns that into fifteen different voices for different characters in the game.  Or that might make it possible for the writer to be his own voice actor without needing to hire dedicated voice actors.  That probably won't be quite as good as what professional voice actors could do, but it could easily be good enough.

What's not going to happen in the foreseeable future is to get good voice acting out of a purely text to speech program that doesn't give the AI any special information about the cadence, tone, or emotion involved.

Scot

Quizzical said:

Getting back on topic, it might be possible for AI in the near future to turn one voice into another.  That way, you only need one voice actor to give lines the proper cadence and emotion, and then the AI turns that into fifteen different voices for different characters in the game.  Or that might make it possible for the writer to be his own voice actor without needing to hire dedicated voice actors.  That probably won't be quite as good as what professional voice actors could do, but it could easily be good enough.

What's not going to happen in the foreseeable future is to get good voice acting out of a purely text to speech program that doesn't give the AI any special information about the cadence, tone, or emotion involved.

Where I disagree is that chat AI is already pushing into the area of detecting emotion from voice alone. I would suggest that if you combined that with a video of the voice actors face you would get an even closer mimicry of the actors emotions. I do not regard either of these two methods as perfect, but then when has anything technology started to do been perfect first time round?

Microsoft’s new AI can simulate anyone’s voice with 3 seconds of audio | Ars Technica

AI Emotion and Sentiment Analysis With Computer Vision in 2023 - viso.ai 

Mendel

Quizzical said:

I'm talking about AI at a level of things that something much like current technology could plausibly do, even if they haven't yet been done.  That's very different from the sort of thing that science fiction writers can dream up.  A dystopia where AI replaces all jobs and humans can't be productive is firmly in the latter.

AI can do mimicry, but not novelty.  If you ask an AI to write something similar to what it already has many examples of, it might well do a respectable job.  For example, it wouldn't surprise me if ChatGPT gave a decent output if you asked it for a patriotic speech that an American mayor might deliver on the fourth of July.  Even if ChatGPT can't do that quite yet, some future AI probably will.

But if you want AI to write the speech making the case for some new initiative, the output is going to be a disaster unless you end up with a "stone soup" style of speech where you have to basically feed the AI all of the points that you want to make in the speech.  At that point, the AI that "writes" the speech barely does any more than a word processor.

I've watched AI for almost 3 decades now.  AI tends to produce large, computationally expensive executable code.  The various writing AIs around seem to not follow this model.  Instead of actual AI code, these seem to skim the internet for words, phrases, and usage.  Install one of these writing creators on a computer isolated from the internet and see what it produces.

There's probably an equal amount of nonsense on the Internet as there is factual, accurate information, so any program using skimming is eventually going to be fundamentally flawed.  Quality input is the core requirement for quality output; AI isn't going to change that.

Publishing companies are already being overwhelmed by computer generated novels hoping to get paid.  Schools are working to validate academic work.  It's popular and easy to use chatGPT and the like to produce walls of text for submission.  It would not surprise me if some of the ad bots seen on this site use similar techniques.  It is difficult to distinguish some types of artificial work from human work and is getting harder.

Another case of mankind inventing technology that is both fascinating and dangerous.  And all too easy to misuse.

Quizzical

Scot said:

Quizzical said:

Getting back on topic, it might be possible for AI in the near future to turn one voice into another.  That way, you only need one voice actor to give lines the proper cadence and emotion, and then the AI turns that into fifteen different voices for different characters in the game.  Or that might make it possible for the writer to be his own voice actor without needing to hire dedicated voice actors.  That probably won't be quite as good as what professional voice actors could do, but it could easily be good enough.

What's not going to happen in the foreseeable future is to get good voice acting out of a purely text to speech program that doesn't give the AI any special information about the cadence, tone, or emotion involved.

Where I disagree is that chat AI is already pushing into the area of detecting emotion from voice alone. I would suggest that if you combined that with a video of the voice actors face you would get an even closer mimicry of the actors emotions. I do not regard either of these two methods as perfect, but then when has anything technology started to do been perfect first time round?

Microsoft’s new AI can simulate anyone’s voice with 3 seconds of audio | Ars Technica

AI Emotion and Sentiment Analysis With Computer Vision in 2023 - viso.ai 

AI may not need to understand sarcasm or anger to mimic talking faster or slower, louder or softer, or in a higher or lower pitch.  I don't know how well it will work, but it's at least plausible.

It's common for AI to be able to mimic something decently well without understanding it at all.  But that's also why it often makes ridiculous, howling errors at seemingly random points in the middle of otherwise sensible output.  It doesn't actually understand what it is saying, but only observes that it scores as the closest that it can get to the training data.

Scot

Quizzical said:

Scot said:

Quizzical said:

Getting back on topic, it might be possible for AI in the near future to turn one voice into another.  That way, you only need one voice actor to give lines the proper cadence and emotion, and then the AI turns that into fifteen different voices for different characters in the game.  Or that might make it possible for the writer to be his own voice actor without needing to hire dedicated voice actors.  That probably won't be quite as good as what professional voice actors could do, but it could easily be good enough.

What's not going to happen in the foreseeable future is to get good voice acting out of a purely text to speech program that doesn't give the AI any special information about the cadence, tone, or emotion involved.

Where I disagree is that chat AI is already pushing into the area of detecting emotion from voice alone. I would suggest that if you combined that with a video of the voice actors face you would get an even closer mimicry of the actors emotions. I do not regard either of these two methods as perfect, but then when has anything technology started to do been perfect first time round?

Microsoft’s new AI can simulate anyone’s voice with 3 seconds of audio | Ars Technica

AI Emotion and Sentiment Analysis With Computer Vision in 2023 - viso.ai 

AI may not need to understand sarcasm or anger to mimic talking faster or slower, louder or softer, or in a higher or lower pitch.  I don't know how well it will work, but it's at least plausible.

It's common for AI to be able to mimic something decently well without understanding it at all.  But that's also why it often makes ridiculous, howling errors at seemingly random points in the middle of otherwise sensible output.  It doesn't actually understand what it is saying, but only observes that it scores as the closest that it can get to the training data.

If you are right I think just for the sake of QA on a gaming project you would have an editor as we discussed before. I see these sorts of jobs as "what the humans are doing for years to come, where I am less certain than you is how many years to come? They don't have to be better than us just as good as, I agree as you say they are not 'even as good as' in most areas in which we work or think of as the greatest achievements of civilization like art. But they march on and our evolution measured in geological ages is at a standstill in comparison.

Scot

Mendel said:

Quizzical said:

I'm talking about AI at a level of things that something much like current technology could plausibly do, even if they haven't yet been done.  That's very different from the sort of thing that science fiction writers can dream up.  A dystopia where AI replaces all jobs and humans can't be productive is firmly in the latter.

AI can do mimicry, but not novelty.  If you ask an AI to write something similar to what it already has many examples of, it might well do a respectable job.  For example, it wouldn't surprise me if ChatGPT gave a decent output if you asked it for a patriotic speech that an American mayor might deliver on the fourth of July.  Even if ChatGPT can't do that quite yet, some future AI probably will.

But if you want AI to write the speech making the case for some new initiative, the output is going to be a disaster unless you end up with a "stone soup" style of speech where you have to basically feed the AI all of the points that you want to make in the speech.  At that point, the AI that "writes" the speech barely does any more than a word processor.

I've watched AI for almost 3 decades now.  AI tends to produce large, computationally expensive executable code.  The various writing AIs around seem to not follow this model.  Instead of actual AI code, these seem to skim the internet for words, phrases, and usage.  Install one of these writing creators on a computer isolated from the internet and see what it produces.

There's probably an equal amount of nonsense on the Internet as there is factual, accurate information, so any program using skimming is eventually going to be fundamentally flawed.  Quality input is the core requirement for quality output; AI isn't going to change that.

Publishing companies are already being overwhelmed by computer generated novels hoping to get paid.  Schools are working to validate academic work.  It's popular and easy to use chatGPT and the like to produce walls of text for submission.  It would not surprise me if some of the ad bots seen on this site use similar techniques.  It is difficult to distinguish some types of artificial work from human work and is getting harder.

Another case of mankind inventing technology that is both fascinating and dangerous.  And all too easy to misuse.

Just from keeping an eye on the adbots here I am convinced they are using some sort of chat UI starting about a week ago. Their posts are more intelligible and related to the thread, not all mind you some are still the older ones which are easy to spot. It is more a gut feeling now that this is not what a real person would say.

If working out which posts were adbots was my job this is one human who would be feeling the heat.

[Deleted User]

The user and all related content has been deleted.

Quizzical

Scot said:

Quizzical said:

Scot said:

Quizzical said:

Getting back on topic, it might be possible for AI in the near future to turn one voice into another.  That way, you only need one voice actor to give lines the proper cadence and emotion, and then the AI turns that into fifteen different voices for different characters in the game.  Or that might make it possible for the writer to be his own voice actor without needing to hire dedicated voice actors.  That probably won't be quite as good as what professional voice actors could do, but it could easily be good enough.

What's not going to happen in the foreseeable future is to get good voice acting out of a purely text to speech program that doesn't give the AI any special information about the cadence, tone, or emotion involved.

Where I disagree is that chat AI is already pushing into the area of detecting emotion from voice alone. I would suggest that if you combined that with a video of the voice actors face you would get an even closer mimicry of the actors emotions. I do not regard either of these two methods as perfect, but then when has anything technology started to do been perfect first time round?

Microsoft’s new AI can simulate anyone’s voice with 3 seconds of audio | Ars Technica

AI Emotion and Sentiment Analysis With Computer Vision in 2023 - viso.ai 

AI may not need to understand sarcasm or anger to mimic talking faster or slower, louder or softer, or in a higher or lower pitch.  I don't know how well it will work, but it's at least plausible.

It's common for AI to be able to mimic something decently well without understanding it at all.  But that's also why it often makes ridiculous, howling errors at seemingly random points in the middle of otherwise sensible output.  It doesn't actually understand what it is saying, but only observes that it scores as the closest that it can get to the training data.

If you are right I think just for the sake of QA on a gaming project you would have an editor as we discussed before. I see these sorts of jobs as "what the humans are doing for years to come, where I am less certain than you is how many years to come? They don't have to be better than us just as good as, I agree as you say they are not 'even as good as' in most areas in which we work or think of as the greatest achievements of civilization like art. But they march on and our evolution measured in geological ages is at a standstill in comparison.

Well sure, you have an easy editor.  Someone says a line, the AI converts it to a different voice, and then the person plays it back to make sure that the conversion worked properly.  Maybe you redo the line if it didn't, or if something interfered.  I'd assume that even professional voice actors sometimes have to do multiple takes to get it right.

Quizzical

Scot said:

Mendel said:

Quizzical said:

I'm talking about AI at a level of things that something much like current technology could plausibly do, even if they haven't yet been done.  That's very different from the sort of thing that science fiction writers can dream up.  A dystopia where AI replaces all jobs and humans can't be productive is firmly in the latter.

AI can do mimicry, but not novelty.  If you ask an AI to write something similar to what it already has many examples of, it might well do a respectable job.  For example, it wouldn't surprise me if ChatGPT gave a decent output if you asked it for a patriotic speech that an American mayor might deliver on the fourth of July.  Even if ChatGPT can't do that quite yet, some future AI probably will.

But if you want AI to write the speech making the case for some new initiative, the output is going to be a disaster unless you end up with a "stone soup" style of speech where you have to basically feed the AI all of the points that you want to make in the speech.  At that point, the AI that "writes" the speech barely does any more than a word processor.

I've watched AI for almost 3 decades now.  AI tends to produce large, computationally expensive executable code.  The various writing AIs around seem to not follow this model.  Instead of actual AI code, these seem to skim the internet for words, phrases, and usage.  Install one of these writing creators on a computer isolated from the internet and see what it produces.

There's probably an equal amount of nonsense on the Internet as there is factual, accurate information, so any program using skimming is eventually going to be fundamentally flawed.  Quality input is the core requirement for quality output; AI isn't going to change that.

Publishing companies are already being overwhelmed by computer generated novels hoping to get paid.  Schools are working to validate academic work.  It's popular and easy to use chatGPT and the like to produce walls of text for submission.  It would not surprise me if some of the ad bots seen on this site use similar techniques.  It is difficult to distinguish some types of artificial work from human work and is getting harder.

Another case of mankind inventing technology that is both fascinating and dangerous.  And all too easy to misuse.

Just from keeping an eye on the adbots here I am convinced they are using some sort of chat UI starting about a week ago. Their posts are more intelligible and related to the thread, not all mind you some are still the older ones which are easy to spot. It is more a gut feeling now that this is not what a real person would say.

If working out which posts were adbots was my job this is one human who would be feeling the heat.

There have been spambots for quite a while that would copy a sentence or paragraph or some such from earlier in the thread, then add their dodgy links.  That requires being able to parse the forum syntax, but no real intelligence.  And that also allows them to have a reply that is mostly on topic if they're quoting someone else who was also on topic.