Don’t Lie to Me: Avoiding Malicious Explanations With <i>STEALTH</i>
نویسندگان
چکیده
STEALTH is a method for using some artificial intelligence-generated models without suffering from malicious attacks or associated unfairness issues. asks so few queries (one per data cluster) that algorithms cannot detect its operation know when to lie.
منابع مشابه
Lie to Me : Virtual Agents that Lie ( Extended
In order to deceive, agents need Theory of Mind capabilities (ToM), that is, the capability to model the others, and reason about the consequences of their actions and their implications in them. In this paper we provide a model for deceptive agents that use a theory of mind with N levels. We then present a case study that was used to compare deceptive agents with one level and with two levels ...
متن کاملLie to Me: Lying Virtual Agents
The objective of this thesis was to endow agents with the ability to model the mind of other agents and to use this information on their reasoning process. We developed an architecture which enables N levels of Theory of Mind. Additionally we wanted to assess if a reasoning ability capable of more then one level would be perceived as more socially intelligent than one with only one level. For t...
متن کاملWhy Explanations Lie: Idealization in Explanation
On the causal approach to explanation, explaining a phenomenon is telling the actual causal story as to why it occurs. Science is full of idealizing explanations that deliberately falsify the relevant causal story. Therefore, either (a) the causal approach to explanation is mistaken, or (b) idealizing explanations, however convenient, are deeply flawed. Correct? I don’t think so. I show that on...
متن کاملStop talking to me - a communication-avoiding ADER-DG realisation
We present a communicationand data-sensitive formulation of ADER-DG for hyperbolic differential equation systems. Sensitive here has multiple flavours: First, the formulation reduces the persistent memory footprint. This reduces pressure on the memory subsystem. Second, the formulation realises the underlying predictor-corrector scheme with single-touch semantics, i.e. each degree of freedom is...
متن کاملDon't Push Me! Collision-Avoiding Swarms
This paper examines a novel particle swarm algorithm which has been applied to the generation of interactive, improvised music. An important feature of this algorithm is a balance between particle attraction to the centre of mass and repulsive, collision avoiding forces. These forces are not present in the classic particle swarm optimisation algorithms. A number of experiments illuminate the na...
متن کاملذخیره در منابع من
با ذخیره ی این منبع در منابع من، دسترسی به آن را برای استفاده های بعدی آسان تر کنید
ژورنال
عنوان ژورنال: IEEE Software
سال: 2023
ISSN: ['0740-7459', '1937-4194']
DOI: https://doi.org/10.1109/ms.2023.3244713