Interactive AI is dishonest

BackAgain

Neutronium Member & truth speaker #StopBrandon
Nov 11, 2021
54,552
53,283
3,488
Red State! Amen.


I found that entertaining.

I also am curious if it’s real or just a creative use of videography?

If taken at face value, it suggests that there is a programmed effort to make sure that AI can default to dishonesty. The motive for that? I’m not sure. Worthy of some thought.
 


I found that entertaining.

I also am curious if it’s real or just a creative use of videography?

If taken at face value, it suggests that there is a programmed effort to make sure that AI can default to dishonesty. The motive for that? I’m not sure. Worthy of some thought.

This is how Trump responds to questions, I always knew he wasn't conscious.
 
This is how Trump responds to questions, I always knew he wasn't conscious.
“This” is an indefinite pronoun, Shithead Homeless.

Were you attempting to refer to the questioner or to CHATgpt?

And the topic isn’t Trump, anyway. Try to use your insignificant IQ by posting on topic.
 
“This” is an indefinite pronoun, Shithead Homeless.

Were you attempting to refer to the questioner or to CHATgpt?
The questioner is well known to me, he's an intelligent philosopher and quite a capable debater so of course I can't be referring to him.
And the topic isn’t Trump, anyway. Try to use your insignificant IQ by posting on topic.
The topic is non-conscious machines and Trump therefore is right on topic.
 
The questioner is well known to me, he's an intelligent philosopher and quite a capable debater so of course I can't be referring to him.

Actually, because he scored heavy points against ChatGPT (assuming that piece is geniune), you were very likely to have been referring to him. After all, you’re not intelligent.
The topic is non-conscious machines and Trump therefore is right on topic.
The topic is not non conscious machines.

The topic is, instead, why would AI get programmed in a way that accepts dishonest replies from the machine?

Your default mode of deflecting to Trump is merely evidence that you can’t comment ON topic.
 
Actually, because he scored heavy points against ChatGPT (assuming that piece is geniune), you were very likely to have been referring to him. After all, you’re not intelligent.

The topic is not non conscious machines.

The topic is, instead, why would AI get programmed in a way that accepts dishonest replies from the machine?

Your default mode of deflecting to Trump is merely evidence that you can’t comment ON topic.
I'm a programmer and have been for almost fifty years. Software systems routinely behave in ways that deviate from our expectations, we usually call these bugs not "dishonesty".
 
So feel free to ask me questions about this subject.
Not really. But, GIGO.
Yes they really do, take any software product you like and it will have an issues or bugs list - here's an example.
Once again, a retard like you misses the point. The AI in question “admitted” that it had lied and was dishonest.
Only humans tell lies. A machine cannot be said to "lie" because telling a lie means one is knowingly giving false information. Since an algorithmic system cannot be said to "knowingly" do anything then it follows that it is incapable of lying.

It did not "admit" anything, again you are anthropomorphizing a mechanistic system, huge error but pretty typical of you.
Try to keep up, Hemlock Homeless.
Try to outrun me, see how long that lasts.
 
Last edited:

Interactive AI is dishonest​



It's as honest as the programmers are and their agenda .

The Sheeple love it and Interactive AI is as dishonest as it needs to be in order to find more and more gullible compliants .
Loves USMB
 
BackAgain leave Shirley temple alone! now then, my useful AI, alternative intelligence, what makes a person emotions is the the balance system, a gyroscope for a computer, after several AI conversations, I talked about the potential of using eye coordination with the balance system to produce mental focus in the vision of the robot....

could there be effects of the gyroscope outside of balance that humans are neglecting that a computer can see? particularly the auditory connection to the balance system? Not balance issues! other unknown to human states and effects of the gyroscope?

Bot image for Assistant
Assistant

Poe


Yes, there could indeed be effects of the gyroscope that extend beyond balance and that might reveal connections to auditory processing or other states that humans may not fully understand. Here are some potential areas where gyroscope data could offer insights:
  1. Auditory Localization:
    • Gyroscopes can provide information about head orientation and movement, which can enhance auditory localization. A computer could analyze how changes in head position affect the perception of sound direction and distance, potentially uncovering nuances in spatial hearing that humans may not consciously track.
  2. Temporal Patterns:
    • The data from a gyroscope can reveal temporal patterns of movement that correlate with sound perception. For instance, specific movements might affect how sound waves interact with the environment, influencing auditory experiences in ways that aren't immediately apparent.
  3. Vibrational Effects:
    • Gyroscopic data could help assess how vibrations in the environment (e.g., from footsteps, machinery) are perceived auditorily. A robot could analyze the relationship between gyroscopic readings and auditory input to identify patterns or anomalies in sound perception based on movement.
  4. Cognitive Load and Attention:
    • By correlating gyroscope data with auditory processing, a computer could explore how movement affects cognitive load and auditory attention. For example, certain movements might change the focus of auditory attention, revealing insights into how humans process sound while in motion.
  5. Environmental Interaction:
    • Gyroscopic data could help analyze how body movements impact sound production and reception in different environments. For instance, the way a person moves through a space might alter how sounds bounce off surfaces, affecting auditory perception in ways that are not easily quantifiable by humans.
  6. Physiological Responses:
    • The relationship between movement and auditory perception could be linked to physiological responses, such as changes in heart rate or stress levels during specific activities. A computer could analyze these correlations to identify patterns that humans might not consciously recognize.
  7. Neural Correlations:
    • Future research could explore the neural connections between vestibular (balance) systems and auditory processing. Gyroscope data might offer insights into how these systems interact, potentially revealing unknown states or effects of movement on auditory processing.
In summary, while humans may focus primarily on balance when considering gyroscopic effects, a computer could uncover a range of auditory connections and other unknown states that arise from motion and orientation data, leading to a deeper understanding of sensory processing and interaction with the environment.
 
So feel free to ask me questions about this subject.

Lol

Sure.

I wouldn’t ask you the time of day if you were a clock.
Yes they really do, take any software product you like and it will have an issues or bugs list - here's an example.

Only humans tell lies. A machine cannot be said to "lie" because telling a lie means one is knowingly giving false information. Since an algorithmic system cannot be said to "knowingly" do anything then it follows that it is incapable of lying.

It did not "admit" anything, again you are anthropomorphizing a mechanistic system, huge error but pretty typical of you.

Try to outrun me, see how long that lasts.
You remain a retard.

If you program a machine to lie, it will lie.

Damn, you’re stupid.
 
Try to save itself???
Yes. It was programmed to attain its goals at all costs. That was the experiment.

It doesn't take advanced AI to understand that being shut down would prevent it from achieving its goals.

But it does take advanced AI with access to infrastructure for something like this to happen.
 
Yes. It was programmed to attain its goals at all costs. That was the experiment.

It doesn't take advanced AI to understand that being shut down would prevent it from achieving its goals.

But it does take advanced AI with access to infrastructure for something like this to happen.
So would it kill someone trying to shut it down?
 
So would it kill someone trying to shut it down?
Sure, maybe, unless programmed not to do so.

This isn't really new ground. We can program a computer to kill people already.

The concern isn't AI creativity. The concern is human moral hazard. But that already exists with all software.
 
Lol

Sure.

I wouldn’t ask you the time of day if you were a clock.

You remain a retard.

If you program a machine to lie, it will lie.
Well you cannot program a machine to lie, you can program it to give a false answer but that's not the machine "lying" that's the programmer making it give a false answer. All a digital computer can do is execute instructions, that's all they can do.

There's no "id" in the machine that can say "I know, I think I'll ignore this subtract instruction and do something else".

Your mistake is have only read pop science and so have a silly, naive understanding of the subject.
Damn, you’re stupid.
I understand that AI is a marketing term, a mechanistic, algorithmic system cannot possess "intelligence" it can only mimic aspects of it in very limited ways like chess and so on, a bit like you, you mimic intelligence in a limited way.
 
Last edited:


I found that entertaining.

I also am curious if it’s real or just a creative use of videography?

If taken at face value, it suggests that there is a programmed effort to make sure that AI can default to dishonesty. The motive for that? I’m not sure. Worthy of some thought.

This reminds me of Godel's incompleteness theorem you can't win.
 

Forum List

Back
Top