quote:

Originally posted by gearhounds:
*** I suppose almost a month qualifies as resurrecting the thread.

No. Four months

Copy that- many thanks.

quote:

Originally posted by gearhounds:...

No matter how good AI gets, it cannot seem to actually think like a human and never will. Simple things like pronunciation, inflection, etc evades “them”. Listening to AI trying to have a natural flow while saying more than a few words bears this out.

...

You should print out your post, dated and put in your Bible (or whatever place you will be able to read it once in a while).

Humans are "feeding" AI at a fantastic rate.


HUMINT(Human Intelligence), COMINT(communications ""), ELINT(electronic ""), IMINT(Imagery ""), SIGINT(Signals ""), MASINT(Measurement and Signature ""), and OSINT(Open-Source "") which are all part of our everyday interactions with people (where recording/transmission) gathering/analyzing of that data is being accomplished is used to "fine tune" the "understanding" and for "AI" to better be able to "learn" and adjust to producing a more believable output.

Ways this occurs:

Telephone, any and all text already written, from books, magazines etc, in digital format, all internet "public source" and available text, including anything sent in plaintext/non-encrypted.

Everything on any public social media.

All voice and video content, including movies, music etc.

Everything we do that has value can be taken freely where data collection can be employed.

Second is "willingly providing", on FB there is trending the "most people cannot identify X" type things. And a ton pf people will "show their intelligence" by "answering a question, solving a puzzle, naming a song/band/actor/movie/TV show.

And I have also seen a great number of "Special Interest" FB groups showing up with obvious things that I am subject matter expert, with many discrepancies of facts, and many people posting to correct or point out the errors.

Some of these are "click bait" type pages, designed to monetize traffic, but I also believe that they can be used in an effort to "gather" deliberate response to specific topic/subject matter to better "learn" how people see and perceive.

In effect, "we" become AI's eyes, ears, nose, mouth, hands, and provide important information where AI is not/cannot actually "sense" certain stimuli and input.

Just look at the AI images being produced. A year ago we had images of people with "screwed up hands and teeth" illegible "words" and other obvious things we laughed at. Now, people are being fooled regularly by such images as the "realism" is rapidly becoming near "perfect". It will be indistinguishable in the future. Maybe one analyzing content for a long period of time can detect "AI produced", but what if one has little time to analyze, like "live stream" where benefit of having the stream recorded and ability to scrutinize the content. (and then take that as many ways as you can in application.

Consider the "mechanical voice" such as used be Stephen Hawking, 7 years ago, compared to the voice you think is so obvious today, and realize that within a year, one will be hard pressed to "catch" the errors in the "speech patterns".
(everyone that responds about such defects, as often is the case in comment sections of videos with AI produced text to speech, provide valuable feedback of what needs to be corrected, and all that is fed back into the code.

I used to read a ton of science fiction, and was a deep adherent of OMNI, various "science" magazines for the style as it "worked" with my "abstract relational" mental processing, and worked well with my having parallel, but different consecutive streams of thought, and pattern recognition made perceiving things, trouble shooting, problem solving a strong suit.

AI will be the closest "artificial us" there is, because we will have created it from the myriad and great numbers of people over our recorded history being the input, nurturing and educating source. It is all "all of us", and the rapid ability for the input to be analyzed and processed into "meaningful" information, by a machine that can access any and all of it in any combination, using all the "various methods of filtering", "information connecting" (Et al. algorithms).

It is not so much that AI may become "self aware/sentient" in the understanding of "alive" as we know life, but that given the ability to "make decisions and operate autonomously, without a body of human layer preventing arbitrary actions by AI, then we screw ourselves, because we have many "automated systems" already that can operate independent of the "human decisions layer in real-time". Example: "self driving cars", where the programming takes into account all the "known" (at the time of programming) variables the car may encounter and the "If, Then, Else, ElseIf..." being limited, and the occasional "never realized, never accounted for, never tested" so the car "does only what it is programmed" and it fails to navigate this "unknown", we have a problem.

Now, put AI in charge of "nukes" (yeah, old saw), without "Human layer" to be the final decision maker, then not only Houston, but the entire world, has a problem.

Pogo was right.



According to Turing, I might be AI...

Do AI programmers constrain AI with Asimov's Three Laws of Robotics?

quote:

Consider the "mechanical voice" such as used be Stephen Hawking, 7 years ago, compared to the voice you think is so obvious today...

Not a good example, as Hawking famously rejected upgrades to his computer voice over the decades. Even though advances in computerized speech technology had long passed that original 1986 version by, he continued using it, because he had decided that this original 1980s style of heavily computerized mechanical speaking was "his voice".

Therefore all that this represents is his deliberate choice to continue to use a 30 year old mechanical voice, rather than representing the best computerized voice there was just 7 years ago (as you seem to be implying).

Not implying, directly pointing out the curve at which synthesized voice is becoming indiscernible from recorded voice sampling.

Hawking used digital voice synthesized "speech", today's AI is using synthesized speech (although, still somewhat hybrid), and will be able to synthesize the entire "voice" and "speech" very soon. The reason I used this was to contrast where "synthesized voice" was and it's "presentation" compared to the more realistic synthesized voice "presentation".

What people are typically accustomed to in the past 20 years, has been "recorded human individual words" used from text to speech.

I worked with a guy developing a PC based system in the late 90s using Dragon Dictate for people with spinal cord injuries that either quadriplegic or upper paralysis. I learned a great deal studying language structure and the process of the speech recognition engine in order to deal with interfaces. I also saw where the algorithms used at that time are nothing like what is being rapid developed in the past few years.

It was voice activated (using dragon to control and a programmed interface/application) to manage and control various electronic devices, cameras, door latches, intercom etc. as well as the PC itself using voice commands controlling various macros, dictation, on screen keyboard and mouse via a IR transceiver and a reflective dot placed on glasses or forehead.

The "voice" for text to speech was not much better at that time than what Hawking's used over a decade, and the differences now from just a few years ago. Granted, the "speech files" used in the very recent past (couple of years) were often simply "recorded" human voice libraries and audio files indexed to the words, and strung together create "speech", but that is being replaced as the "voice samples" are being broken down and analyzed by computer to synthesize "real voice" where the "voice/speech" is created on the fly solely by machine.

(I was contacted for a position "reading" for this process, but it is not something I believe I would be suited for doing, so I declined, but I did have a good discussion on the technology and the goals.)


The point was, where things are today compared to a year or two ago, will be far beyond people's ability to perceive if real vs artificial much sooner than people realize.

People are fooled by videos purported to be recorded from live events, that are from PC games, like aircraft strikes, dogfights and the like.

Recently saw a video where a guy was showing "leaked" footage of the F-47 taxing and flying. Clear to me they are CGI, but quite a few people less familiar with real aircraft, as well as the realism of today's sims, are accepting them on the face of it.

The thread is: Black Mirror AI - we won't be able to trust this stuff

And part of its database is Ebonics.


It must be so confused.