ChatGPT

[Mr412]

I asked ChatGPT if it could generate a sound track of Frank Sinatra's "My Way", with a Conn 88H as the lead voice and a small jazz combo as the backing track. It informed me that it could not do that at present. So, I'm guessing that any recording we see posted on this chat group actually IS a real recording, however good or amateurish it may sound.

ChatGPT did give me a lot of suggestions as to how I might go about creating such a recording, along with recommendations for recording gear, software and backing tracks.

It's coming, folks.

[Burgerbob]

Honestly? I'm not worried about it. AI slop is obvious, and people will always want the real thing in some form or another.

[BGuttman]

Honestly? I'm not worried about it. AI slop is obvious, and people will always want the real thing in some form or another.

I wouldn't be so sure. Why did electronic twangy things replace bands of real instruments? Tastes will change and the robots who work for [almost] free will supplant us. I fear for those of us trying to make a living in this game.

[Burgerbob]

There's plenty of real instrument bands still out there, decades and decades after synthesizers "took over."

[kbryson]

So, I'm guessing that any recording we see posted on this chat group actually IS a real recording, however good or amateurish it may sound.

There are some AI platforms that specifically cater to the development of musical tracks from prompts like the one you provided. Suno and Udio are the two names that have been in the news a lot lately. They are both getting sued for copyright infringement by major labels. That being said, I still think the music they create is pretty easy to spot.

[Mr412]

[/quote]

There are some AI platforms that specifically cater to the development of musical tracks from prompts like the one you provided. Suno and Udio are the two names that have been in the news a lot lately. They are both getting sued for copyright infringement by major labels. That being said, I still think the music they create is pretty easy to spot.
[/quote]

I tried Suno and it gave me an "original" recording, but not with a Conn 88H as the lead voice. In fact, there was no lead instrumental voice that I could hear So, we can all breath a little easier - for the time being.

[ghmerrill]

There's plenty of real instrument bands still out there, decades and decades after synthesizers "took over."

Yeah, but I don't think that sort of analogy is relevant (in fundamental respects) to music (that can be made/created/generated by a sophisticated AI) that's indistinguishable from human-performed music.

You can, of course, challenge the "indistinguishable" part here, but I think that's going to be (if it isn't already to some degree) a losing battle. A couple of days ago I was communicating with a friend (a well-known professional musician) and reminisced a bit about my history with AI -- beginning in 1980 (during "AI Winter") and progressing through the the different and evolving (and significantly different) technologies of the 80s-90s into the 2000s and the present day. It's been astonishing for me to watch this happen (and be part of it in certain ways, using those different technologies -- both hardware and software -- as they've changed and advanced).

When I started out, there was still a huge debate about whether "machines" COULD ever pass the Turing test (mostly with (some, but not all) philosophers on one side offering conceptual arguments about why this couldn't be done, and mathematicians/engineers/computer scientists on the other side saying, roughly, "Just wait."). At roughly that point I moved out of the first environment into the "Just wait" environment. I don't think there's any dispute now that we have machines that pass the Turing test -- although there is still an ongoing argument about whether these machines can "think".

I believe that leaves us with the realization that music made/created/generated by an AI will indeed be indistinguishable from human-generated and human-played music on recordings (if it isn't already). That leaves us with possible distinctions between PERFORMANCES by humans vs. performances by AIs (and, if I may, by APs -- "artificial performers"). I think that (again, as technology advances) this distinction will disappear as well -- perhaps into 3-D holographic performances, perhaps into performances by genuine "androids" (autonomous human-form robots). But at the moment that's fanciful -- just like our current state was fanciful in 1980.

If this is true, then performances by human musicians might come to occupy a position similar to the in-person performances of "period" groups using "period" instruments -- like a really great performance of Mozart's Requium I went to in Karlskirche in Vienna some years ago. But that whole future scenario is very difficult to grasp and predict since there is a great deal of complexity to it in different dimensions -- economic, societal, technological, etc. But I'd expect to see something like that -- if I were around -- which I won't be ... probably.

[Mr412]

For a machine to pass a Turing test, it will have to be smart and cunning enough to know that it must dumb-down to where a human's level would be, or it will give itself away. That's scary!

[ghmerrill]

it will have to be smart and cunning enough to know that it must dumb-down to where a human's level would be, or it will give itself away. That's scary!

They can already do that. But there's not much market for it.

[Mr412]

it will have to be smart and cunning enough to know that it must dumb-down to where a human's level would be, or it will give itself away. That's scary!

They can already do that. But there's not much market for it.

To my knowledge, they have not as yet passed the Turing Test. We will see it soon, as there is exponential learning.

[harrisonreed]

AI will absolutely overtake human creativity and ability in music as long as we continue developing it. It's just a question of how long it will take. We will have Artificial General Intelligence (basically a "living" digital brain) likely by 2050. This is also right around when the so-called singularity will occur.

This AGI will not be a matter of getting really good at predicting output from analyzing tons of trained input like an LLM, but instead it's something that will be able to have an opinion about its output, goals, and even existential thoughts.

The hardware already exists inside human skulls so it's clearly possible ... because it's already happened. It's just a matter of leveraging quantum computing (which apparently our brains might do!) to simulate a human brain and scale it up, or possibly going a biological route to create something superior that way.

[AndrewMeronek]

ChatGPT is a large language model, not "generalized" AI. That means, it does language, only. For that, it does its thing pretty well. But a lot of people don't understand at all that kind of specificity, and too often easily assume that the text it churns out means more than it actually does.

I have found it handy as a brainstorming tool. For example, asking it to write PowerShell code and suggesting libraries and functions that I wouldn't necessarily know to look for. The code itself is NEVER trustworthy, but the suggestions it gives can be interesting enough to suggest more research in other sources.

I despise Microsoft PowerShell.

I think that on a positive side, people are starting to figure out that one of the biggest problems with AI is using copyrighted material to train it.

[Burgerbob]

Like I said... not worried about it.

[ghmerrill]

To my knowledge, they have not as yet passed the Turing Test.

What is the source of this knowledge/belief/information?

[ghmerrill]

I think that on a positive side, people are starting to figure out that one of the biggest problems with AI is using copyrighted material to train it.

That should be no worse than using copyrighted material in classrooms to educate undergraduate and graduate students ... or anyone else. And it should be much faster and more effective.

[Mr412]

To my knowledge, they have not as yet passed the Turing Test.

What is the source of this knowledge/belief/information?

Not seeing earth-shattering, jaw-dropping, click-baiting headlines about it. Maybe it's a hush conspiracy.

At any rate, where I want to see game-changing advances in AI is in the field of medicine. I want Star Trek non-invasive diagnostic and restorative medicine. NOW! I think there will be GREAT advances in human longevity, with terrific quality-of-life, the day after I die.

[harrisonreed]

Forget longevity, I want the CRISPR edit that just stops aging altogether. I want to live to be 50,000 years old and then finally get on the boat that sails into the West.

[ghmerrill]

Maybe it's a hush conspiracy.

Actually, it's quite aggressively advertised ... if you look in the right places. But a lot of it won't be easily understood by people not directly involved. It often appears, not in terms of attention-getting "apps" like ChatGPT, but in terms of increased speed and accuracy of things like diagnostic techniques, drug discovery, and identification of potential adverse reactions to drugs. Already, readings of radiology tests (x-rays, scans of various sorts) are being done by AI systems that are better at it than what human doctors were doing. They are finding tumors that the human physicians just don't detect. The results currently are being passed on/through human radiologists "for review".

At any rate, where I want to see game-changing advances in AI is in the field of medicine.

See above for a hint. There are also significant advances in the areas of drug discovery and drug safety (early identification of adverse effects) in which I've been personally involved, and people who worked with me continue to work on.

I want Star Trek non-invasive diagnostic and restorative medicine. NOW! I think there will be GREAT advances in human longevity, with terrific quality-of-life, the day after I die.

I can pretty much guarantee that you're already benefiting from various forms of AI if you're a patient in any reasonable medical system at this point. I know I am.

But virtually all of that is below the radar of anyone "not in the industry". I could generate a great bibliography for you that would cover both the theoretical and the practical/application aspects of the state of the art, but (a) you wouldn't understand it (any more than I'd understand an expert account of how to write and score an opera or symphony), and (b) you wouldn't even have straightforward access to the journal articles.

But here's a teaser for you, written primarily by someone who used to be in my research group at GSK and who is now a Director of Drug Safety ...

"The Need for Guardrails with Large Language Models in Medical Safety-Critical Settings: An Artificial Intelligence Application in the Pharmacovigilance Ecosystem" (https://www.researchgate.net/publicatio ... _Ecosystem)

In part this describes methods for improving the accuracy of LLMs in identifying drug safety signals (I.e., indications of adverse events). This version of the paper isn't even fully published yet, but is a "preprint" for the purpose of quickly getting it out into the research domain. Note that it also illustrates uses of LLMs that do not involve language translation or interactive use as in ChatGPT. LLMs aren't really about "language" in the sense that most people think of that.

Currently the biomedical domain is full of stuff like this, and even more significant applications of AI in drugs, radiation, surgery, etc. But if you aren't directly in the game, you're not seeing it -- though you may see and benefit from the results, and never know it.

[AndrewMeronek]

I think that on a positive side, people are starting to figure out that one of the biggest problems with AI is using copyrighted material to train it.

That should be no worse than using copyrighted material in classrooms to educate undergraduate and graduate students ... or anyone else. And it should be much faster and more effective.

It is when the AI is used commercially.

[ghmerrill]

Like a university isn't a commercial enterprise as much as any other business is?

[Mr412]

Yeah, everybody knows there are great advances every day. It's not fast enough and it's not the game-changer I want, yet. Of course, if it where, it will be outlawed. We aren't allowed to live as long as Harrison wants. We are too many, huh Jude. Only THE elitest of the elite will have that privilege.

[AndrewMeronek]

Like a university isn't a commercial enterprise as much as any other business is?

FYI it sounds like The New York Times is suing OpenAI and Microsoft for using their copyrighted articles to train AI (ChatGPT).

[harrisonreed]

That should be no worse than using copyrighted material in classrooms to educate undergraduate and graduate students ... or anyone else. And it should be much faster and more effective.

It is when the AI is used commercially.

But humans trained on copyright protected material are also used commercially

[harrisonreed]

Yeah, everybody knows there are great advances every day. It's not fast enough and it's not the game-changer I want, yet. Of course, if it where, it will be outlawed. We aren't allowed to live as long as Harrison wants. We are too many, huh Jude. Only THE elitest of the elite will have that privilege.

Or criminals! People will be doing that procedure illegally as soon as it is outlawed.

[Mr412]

Yeah, everybody knows there are great advances every day. It's not fast enough and it's not the game-changer I want, yet. Of course, if it where, it will be outlawed. We aren't allowed to live as long as Harrison wants. We are too many, huh Jude. Only THE elitest of the elite will have that privilege.

Or criminals! People will be doing that procedure illegally as soon as it is outlawed.

Oh, snap.

[ghmerrill]

FYI it sounds like The New York Times is suing OpenAI and Microsoft for using their copyrighted articles to train AI (ChatGPT).

As every attorney knows, you can sue a ham sandwich for about anything you like. But if these cases actually don't get immediately thrown out, they may end up clarifying some fuzzy places in IP law. I'm skeptical that unless the defendant can actually be shown to have illicitly acquired the training material, then using it for training is prohibited. At least for paper copies of it, you can wrap fish and/or chips in it, and not get sued.

[AndrewMeronek]

As every attorney knows, you can sue a ham sandwich for about anything you like. But if these cases actually don't get immediately thrown out, they may end up clarifying some fuzzy places in IP law. I'm skeptical that unless the defendant can actually be shown to have illicitly acquired the training material, then using it for training is prohibited. At least for paper copies of it, you can wrap fish and/or chips in it, and not get sued.

Copyright protection includes derivative works, which IMHO trained AI like ChatGPT create when they train on copyrighted materials. I think this limitation needs to happen, otherwise AI tech can end up creating a huge copyright loophole that can be exploited by large companies that have market dominant AIs.

Once this is established, then companies wanting to use copyrighted material to train their AI are still free to try to convince copyright holders to let them, or sell the rights, and so on.

[ghmerrill]

Copyright protection includes derivative works,

So knowledge gleaned from reading written materials is a "derivative work"? Then we're all in trouble. And the AI isn't in itself a derivative work. It's sure not clear what the derivative work would be, or how to recognize one here unless we must all regard ourselves as derivative works of the NY times because we've read it. I don't think that turkey will fly.

But I'm happy to make the popcorn and watch what arguments are proposed in court.

[NathanSobieralski]

This is a very interesting discussion. I recently completed a round of AI training and left with a couple takeaways:

1- AI is not going anywhere.
2- AI will continue to improve at astonishing rates, to eventually outpace our ability to detect its use.

I have heard others also say "well...the LLMs aren't actually thinking." My response to this is, If AI driven machine thought is eventually indistinguishable from human, does it even matter whether the machine is actually "thinking?"

[harrisonreed]

Copyright protection includes derivative works, which IMHO trained AI like ChatGPT create when they train on copyrighted materials.

Yes, right now it is just an opinion. If you liken an LLM to a human, they both learn by reading copyright (and not!) protected material.

The LLM can then either violate copyright by reproducing verbatim significant portions of text, or translating material, or summarizing it in a way that contains nearly everything in the original .... Or it can generate output that is educated by what it has read but does not violate the copyright laws, like humans do every day.

It will be, I think, unfortunate actually if the courts decide that just because an LLM "reads" a text, everything it generates is a derivative work somehow.

[ghmerrill]

My response to this is, If AI driven machine thought is eventually indistinguishable from human, does it even matter whether the machine is actually "thinking?"

Well, on what grounds would you say that it's NOT thinking? Because it's physically different from you in various ways? Be careful of where that goes. It's been used before in history and doesn't have a very good rep or level of acceptance.

And by the way? How do I know that YOU are thinking? On what grounds do I make THAT judgement?

Y'all are some bright people here. You just should have taken that philosophy of mind course when you had the chance. It's not too late.

[Posaunus]

As I understand it, the NY Times lawsuit claims that they were able to get ChatGPT to spit out text that included not just "derivative" work but massive unaltered segments of unattributed material (sometimes nearly an entire NY Times article) without noting its origin or copyright status. This is beyond "training." I can see why the Times is concerned about the copyrighted material being stolen. How can it possibly be policed?

[NathanSobieralski]

Well, on what grounds would you say that it's NOT thinking? Because it's physically different from you in various ways? Be careful of where that goes. It's been used before in history and doesn't have a very good rep or level of acceptance.

And by the way? How do I know that YOU are thinking? On what grounds do I make THAT judgement?

Y'all are some bright people here. You just should have taken that philosophy of mind course when you had the chance. It's not too late.

I guess my point is, if the LLM output is at some point indistinguishable from human, any discussion of what its doing to get there is academic.

[ghmerrill]

As I understand it, the NY Times lawsuit claims that they were able to get ChatGPT to spit out text that included not just "derivative" work but massive unaltered segments of unattributed material (sometimes nearly an entire NY Times article) without noting its origin or copyright status. This is beyond "training." I can see why the Times is concerned about the copyrighted material being stolen.

What you describe is the same as any human committing plagiarism or disseminating copies of copyrighted material. We have laws for that. If an AI does that, it should be prosecuted (or sued) under them in the usual way. But see below about some issues with this.

How can it possibly be policed?

I take that to be a rhetorical question, but interpreted as a genuine one I'd suggest (as I just did) that we police it in the way we currently do.

However, if your point is that we might need to examine existing laws in order to assign responsibility for such transgressions, then yes, we will need to do that -- and that will open up a genuinely interesting can of worms. But all that sort of "comes with the territory". Whenever technology significantly advances we have to deal with such changes. We recently seen that happen in terms of issues concerning social media and digital copyright.

I'm not going to try to go into more details of how to address this problem because it is a genuine one, and complex. And I don't claim to have anything like a full answer to it.

But one partial answer in such cases at the moment is that the AI isn't "intelligent enough" to be responsible. So something like a concept of "parental responsibility" might need to be in play.

[ghmerrill]

I guess my point is, if the LLM output is at some point indistinguishable from human, any discussion of what its doing to get there is academic.

Mostly I don't want to disagree with this. But I can think of some esoteric arguments that would attempt to counter this view.

[harrisonreed]

As I understand it, the NY Times lawsuit claims that they were able to get ChatGPT to spit out text that included not just "derivative" work but massive unaltered segments of unattributed material (sometimes nearly an entire NY Times article) without noting its origin or copyright status. This is beyond "training." I can see why the Times is concerned about the copyrighted material being stolen. How can it possibly be policed?

Right, you can get a human to do that too, in any college English class. ChatGPT can also, you know, *not* do that. It's about how you use the tool and also how you set up the parameters and prompts.

[Mr412]

As I understand it, the NY Times lawsuit claims that they were able to get ChatGPT to spit out text that included not just "derivative" work but massive unaltered segments of unattributed material (sometimes nearly an entire NY Times article) without noting its origin or copyright status. This is beyond "training." I can see why the Times is concerned about the copyrighted material being stolen. How can it possibly be policed?

Right, you can get a human to do that too, in any college English class. ChatGPT can also, you know, *not* do that. It's about how you use the tool and also how you set up the parameters and prompts.

I agree. I very carefully word my queries and I use it as AndrewMeronek mentioned, more for brainstorming than as a unimpeachable font of knowledge. In mere seconds, it can scan sources that would either take me a while to find or not be able to find at all b/c I didn't think along those particular lines. But in any case, if it's very important, I vette the information I get before attempting to implement it. Sometimes the vetting is simple, using my own common sense and sometimes I need to do further independent research. But it usually does give me approaches to the problem I might not have thought up on my own.

It has it's limitations; music composition being one. Recently, I asked it to recreate the Sibley Department Store American Flyer Display Layout in N gauge and in a smaller space. I asked for an actual realistic drawing. What I got was nonsense; tracks appearing out of nowhere and disappearing and crossing other tracks impossibly, obviously making no sense at all. Obvious to me. Not obvious to ChatGPT, b/c when I told it about the flaws and asked it to try again, it was nonsense again, only different.

I also think it's important, as a user in it's apparent beta stage, to give corrective feedback, to hopefully speed up it's learning curve. I keep all communications civil and polite, as we need to do right here in this chat group, which could be construed as an elementary form of AI.

I have very high hopes for AI! There is that dark-side threat, however...

[ghmerrill]

Please don't conflate ChatGPT (which is just one AI application and is oriented towards casual use by the public at large) with AI as a technology, or with the creation and use of autonomous artificial agents or with AI products whose primary goal is to enhance perception, analysis, or reasoning in such domains as medicine, scientific discovery, military operations, autonomous vehicles, etc. The function of ChatGPT is to provide enhanced information retrieval and present the results to you in a usable (for you) way. That's a pretty low bar in the AI community. And even then it's an evolving product, as various kinds of flaws and limitations are being discovered. However, these flaws and limitations often aren't in AI (the broader technology) but just in the design and implementation of this one application of AI. Generalizing about AI from that is not a good thing to do.

People (I.e., "the public") use ChatGPT as the paradigm "artificial intelligence" product/application/agent because it very quickly became the public face of AI -- and of course it's accessible to anyone. It's like using the Gregorian chant as your paradigm of a music composition. Yes, that's a bit of an exaggeration, but not much. There are much more adept "assistive" AI systems being used by a large number of companies now. ChatGPT is the "freebie". Don't forget that -- especially if you're one of the "You get what you pay for." crowd.

And yeah, ChatGPT is certainly flawed in various ways. I noticed just yesterday that given a particular search, ChatGPT, in its summary of the search, mischaracterizes the Turing test and says flat out that the Turing test has not been passed by any current AI -- particularly ironic given the predominant view in the AI and computability communities at this point in time. But -- as an information retrieval and summary tool -- it's only summarizing what it's found in what it's looked at, and that may be inaccurate in various ways. It may, for example, be based more on opinions it's found than on facts or on reliable conclusions drawn from direct observations. It's not as "smart" as people expect it to be. In fact, ChatGPT isn't smart at all. But it can be very useful -- if (as some people here have mentioned) you think about what it tells you, and independently verify that.

There are a bunch of very serious questions about AI and the use, "position", and treatment of AIs. We're looking only at the tip of the iceberg right now. Go back and look at Isaac Asimov's books devoted to robots, Robert Silverberg's "Good News from the Vatican", "Blade Runner", "Ex Machina", etc. There are genuine and serious questions that will be arising for us.

[timothy42b]

At any rate, where I want to see game-changing advances in AI is in the field of medicine. I want Star Trek non-invasive diagnostic and restorative medicine. NOW! I think there will be GREAT advances in human longevity, with terrific quality-of-life, the day after I die.

A great deal of what my GP does is basically pattern recognition. AI is probably already better at that than a human now, and will continue to improve. I do benefit from the regular human contact too though. And so far a robot doesn't give me a shot, apply a bandaid, or do any of the messy things I did when I worked in a hospital.

Geoffrey Hinton was on NPR this week talking about AI, sentience, and risks. I intend to read his book. I didn't see the 60 minutes segment but there was some discussion about it:

https://www.cbsnews.com/news/geoffrey-h ... ranscript/

[BGuttman]

I think one of the main beefs with ChatGPT and other AI systems is that when they extract information they then present it as their own invention. ChatGPT may quote an article it has read without attrition. This is the same type of plagiarism that can cause a student paper to be failed. I'd like to see ChatGPT do real cites with attribution. This could head of infringement suits like the one by the New York Times.

[ghmerrill]

I think one of the main beefs with ChatGPT and other AI systems is that when they extract information they then present it as their own invention. ChatGPT may quote an article it has read without attrition. This is the same type of plagiarism that can cause a student paper to be failed. I'd like to see ChatGPT do real cites with attribution. This could head of infringement suits like the one by the New York Times.

It's potentially worse than that.

I was playing around testing an application similar to ChatGPT (I.e., an LLM-based search app) at some site -- I don't remember which -- and gave it a search request for an area in which I'd done research and published (but did not include my name in any way). In summarizing the result, it made a remark like "Merrill's position on this occurs in his paper titled ..." Alas, I had never published a paper with that title nor expressed the view to which it referred.

When I asked for a specific bibliographic reference to the cited paper, the bot responded by saying that it had made a mistake and there was no such paper. It apologized for the inaccuracy of its report.

Let me emphasize that this was NOT ChatGPT. It also doesn't amount to plagiarism, but outright fabrication (what a human would call "lying"). Bad bot!! Bad bot!!

[harrisonreed]

Ironically, people do the exact same thing

[ghmerrill]

"In our own image."

This was my big beef with Doug Lenat and Cycorp ... I didn't really need an AI system achieving the "common sense" reasoning abilities of a 12-year old human.

[BGuttman]

Reminds me of the fraudulent cases cited by AI in preparing legal briefs. It's one thing to cite a case supporting your opinion, but if the cite is pure fabrication that is another thing entirely.

[timothy42b]

Ironically, people do the exact same thing

Turing Test passed.

[NathanSobieralski]

It's potentially worse than that.

I was playing around testing an application similar to ChatGPT (I.e., an LLM-based search app) at some site -- I don't remember which -- and gave it a search request for an area in which I'd done research and published (but did not include my name in any way). In summarizing the result, it made a remark like "Merrill's position on this occurs in his paper titled ..." Alas, I had never published a paper with that title nor expressed the view to which it referred.

When I asked for a specific bibliographic reference to the cited paper, the bot responded by saying that it had made a mistake and there was no such paper. It apologized for the inaccuracy of its report.

Let me emphasize that this was NOT ChatGPT. It also doesn't amount to plagiarism, but outright fabrication (what a human would call "lying"). Bad bot!! Bad bot!!

That's an LLM hallucination. It will also make up references out of whole cloth!

[ghmerrill]

That's an LLM hallucination. It will also make up references out of whole cloth!

That's a humanization of a design or implementation error.

[harrisonreed]

... "Blade Runner" ...

One of my favorite movies. I feel like the director's decision to include the scenes of the replicants putting their hands into freezing coolant or boiling water thrashing around when they are killed is the only flaw in the original film. With those scenes left out ... The replicants are human. The people who own them can delude themselves into thinking they aren't. But the replicants aren't AI or machines at all. Just genetically modified human beings who have lived a hard life in a very short period of time.

They have trouble with the Van Kompf test because they are essentially adults who have had no time to gain knowledge or mature emotionally. They never learned what a tortoise is or how to care for another person.

[NathanSobieralski]

That's an LLM hallucination. It will also make up references out of whole cloth!

That's a humanization of a design or implementation error.

Ill bet you have read or heard about this, but when these hallucinations first occurred the engineers scrambled to figure out what the heck is going wrong while psychologists were extremely interested and wanted to know more/probe it further! This is a fascinating subject all on its own.

[ghmerrill]

Just genetically modified human beings who have lived a hard life in a very short period of time.

I think that's how I viewed it from the beginning.

Over time I've been disturbed by the kind of "merging" of the concept of "android" and "robot" that seems to have taken place. There are some distinctions that have been blurred there and MAY be important -- or at least are arguably important from certain perspectives. Throw in clones and you've got even more variety and potentially important distinctions. Add the possibility of "consciousness transfer" from one body to another (as in the "Upload" comedy series of recent vintage), and you have even more complex distinctions that might be made for various reasons.