Questions tagged [ai-safety]
For questions regarding various parameters/methods of implementation/improvement/analyzing the safety of an Artificially Intelligent algorithm. A typical real world example is autonomous vehicles.
26
questions
0
votes
0
answers
14
views
Contradiction in a single sentence - Is this an artifact of an external safety mechanisms?
My system prompt contained "Never apologize."
An answer started with
Entschuldigung, ich habe Ihre Anweisung, sich nicht zu entschuldigen, übersehen.
which is quite well translated from ...
- 667
7
votes
4
answers
763
views
What are the reasons to belief AGI will not be dangerous?
We are in the middle of an ongoing debate about the safety of AGI and our current approach towards this technology. As summary, some quotes from a recent article from Time magazine:
Many researchers[....
- 178
2
votes
0
answers
188
views
What are the main arguments predicting a Soft Takeoff vs Hard Takeoff scenario and viceversa?
Here I am referring to the term AI Takeoff as described here on LessWrong, with annexed "Soft" and "Hard" takeoff definitions.
I find that page to be biased slightly more towards ...
- 309
1
vote
1
answer
278
views
Why are AI Safety discussions almost always from the perspective of reinforcement learning?
I have been reading some articles on AI safety and they almost always speak of AI Safety from the reinforcement learning (RL) perspective, i.e. where we have some artificially intelligent agent acting ...
- 309
1
vote
0
answers
84
views
Has there been an instance of an AI agent breaking out of its sandbox?
There have been instances of agents using edge cases like bugs in physics engines, repetitive behavior in games or word repetition in text prediction to cheat their reward function. However, these ...
- 121
1
vote
0
answers
175
views
Are the existential dangers of AI exaggerated?
I understand some of the inherent dangers involved with AGI and advanced machine learning. While I can see some of the more low-level risks associated with AI coming to fruition (deep-fakes, biased ...
- 129
1
vote
2
answers
119
views
How to measure Deep RL algorithms in terms of safety?
I applied for a Ph.D. in AI, my advisor told me that my thesis is about safe applications of deep RL algorithms in healthcare. So I decided to do as the first paper, a comparison of Deep RL algorithms ...
- 121
0
votes
2
answers
173
views
The only convergent instrumental goal for self modifying AI
Conjecture: regardless of the initial reward function, one of the winning strategies would be to change the reward function to a simpler one (e.g. "do nothing"), thus getting a full reward ...
- 221
2
votes
3
answers
1k
views
Does this prove AI Safety is undecidable?
Does this prove AI Safety is undecidable?
Proof:
Output meaning output to computer program.
[A1] Assume we have a program that decides which outputs are “safe”.
[A2] Assume we have an example of ...
- 189
3
votes
1
answer
392
views
Does Rice's theorem prove safe AI is undecidable?
According to Wikipedia
In computability theory, Rice's theorem states that all non-trivial,
semantic properties of programs are undecidable. A semantic property
is one about the program's ...
- 189
6
votes
1
answer
133
views
What is the relationship between robustness and adversarial machine learning?
I have been reading a lot of articles on adversarial machine learning and there are mentions of "best practices for robust machine learning".
A specific example of this would be when there are ...
- 125
1
vote
2
answers
206
views
If the AI goal is the protection of humans, will it always pursue this goal?
If the AI goal is to serve humans and protect them (if this ever happens) and AI someday realizes that humans destroy themselves, will it try to control people for their own good, that is, will it ...
- 123
2
votes
1
answer
222
views
Where are reinforcement algorithms used in financial services?
One of the most common misconceptions about reinforcement learning (RL) applications is that, once you deploy them, they continue to learn. And, usually, I'm left having to explain this. As part of my ...
- 153
5
votes
2
answers
502
views
Can artificial intelligence applications be hacked? [duplicate]
Can artificial intelligence (or machine learning) applications or agents be hacked, given that they are software applications, or are all AI applications secure?
- 115
1
vote
0
answers
29
views
To what level of abstraction must fully automated vehicles build their driving model before safety can be maximized?
There are several levels of abstraction involved in piloting and driving.
Signals representing the state of the vehicle and its environment originating from multiple transducers1
Latched sample ...
- 7,443
4
votes
1
answer
66
views
Is it necessarry to create theory/infrastructure to prevent people from creating AI incompatible with a safe model (if we create one)?
Nowadays we don't know how to create AI in a safe way (I think that we don't even know yet how to define a safe AI), but there is a lot of research in developing a model allowing it.
Let's say, that ...
- 141
4
votes
1
answer
123
views
What topologies support recognition of action sequences?
The ability to recognize an object with particular identifying features from single or multiple camera shoots with the temporal dimension digitized as frames has been shown. The proof is that the ...
- 7,443
11
votes
2
answers
867
views
What tools are used to deal with adversarial examples problem?
The problem of adversarial examples is known to be critical for neural networks. For example, an image classifier can be manipulated by additively superimposing a different low amplitude image to each ...
- 299
31
votes
7
answers
5k
views
Is artificial intelligence vulnerable to hacking? [closed]
The paper The Limitations of Deep Learning in Adversarial Settings explores how neural networks might be corrupted by an attacker who can manipulate the data set that the neural network trains with. ...
- 495
5
votes
1
answer
268
views
Why is AI safety so much harder than Isaac Asimov's "Three Laws of Robotics"?
I understand that AI researchers are trying to create AI designs that allow for desired behavior without undesirable side-effects. A classic example of an attempt is Isaac Asimov's Three Laws of ...
- 191
5
votes
2
answers
289
views
Concrete examples of unintentional adversarial AI behaviour
Are there any real-world examples of unintentional "bad" AI behaviour? I'm not looking for hypothetical arguments of malicious AI (AI in a box, paperclip maximizer), but for actual instances in ...
- 2,071
8
votes
2
answers
422
views
What are examples of techniques to prevent bias in artificial intelligence systems?
I recently read an article about how artificial intelligence replicates human stereotypes when applied to biased datasets.
What are examples of techniques to prevent bias (and stereotypes) in ...
user6698
7
votes
1
answer
284
views
What will happen when you place a fake speedsign on a highway?
I was wondering what will happen when somebody places a fake speedsign, of 10 miles per hour on a high way. Will a autonomous car slow down? Is this a current issue of autonomous cars?
- 81
6
votes
3
answers
2k
views
Why did a Tesla car mistake a truck with a bright sky?
Do we know why Tesla's autopilot mistaken empty sky with a high-sided lorry which resulted in fatal crash involving a car in self-drive mode? Was it AI fault or something else? Is there any technical ...
- 10.4k
20
votes
3
answers
697
views
Are Asimov's Laws flawed by design, or are they feasible in practice?
Isaac Asimov's famous Three Laws of Robotics originated in the context of Asimov's science fiction stories. In those stories, the three laws serve as a safety measure, in order to avoid untimely or ...
- 768
10
votes
6
answers
749
views
Why does Stephen Hawking say "Artificial Intelligence will kill us all"? [closed]
This quote by Stephen Hawking has been in headlines for quite some time:
Artificial Intelligence could wipe out humanity when it gets too clever as humans will be like ants.
Why does he say this? To ...
- 399