Scribd
Upload
12 views17 pages

AI_Week12

The document discusses the significance of Explainable AI (XAI) in ensuring transparency and compliance with the EU Artificial Intelligence Act, emphasizing the need for AI systems to be understandable and accountable. It outlines the importance of explainability in high-risk sectors, the arguments for and against requiring AI to explain its decisions, and the regulatory frameworks that mandate a 'right to an explanation.' Additionally, it categorizes AI systems based on risk levels and introduces the concept of General Purpose AI (GPAI) and its associated risks.

Uploaded by

abdulnafeysyed
Copyright
© © All Rights Reserved
We take content rights seriously. If you suspect this is your content, claim it here.
Available Formats
Download as PDF, TXT or read online on Scribd
12 views17 pages

AI_Week12

The document discusses the significance of Explainable AI (XAI) in ensuring transparency and compliance with the EU Artificial Intelligence Act, emphasizing the need for AI systems to be understandable and accountable. It outlines the importance of explainability in high-risk sectors, the arguments for and against requiring AI to explain its decisions, and the regulatory frameworks that mandate a 'right to an explanation.' Additionally, it categorizes AI systems based on risk levels and introduces the concept of General Purpose AI (GPAI) and its associated risks.

Uploaded by

abdulnafeysyed
Copyright
© © All Rights Reserved
We take content rights seriously. If you suspect this is your content, claim it here.
Available Formats
Download as PDF, TXT or read online on Scribd
You are on page 1/ 17

Artificial

Intelligence (AI)
Dr. Yıldız Karadayı
02.December.2024
Explainable AI (XAI)
Explainable AI (XAI)

Importance of Transparency and Explainability


Transparency and explainability are essential for compliance
with The EU Artificial Intelligence Act, ensuring that AI systems
are both comprehensible and reliable:

● Transparency involves openly sharing details about data


sources, algorithms, and decision-making processes to foster
accountability.

● Explainability offers clear rationales for AI decisions,


enhancing trust and facilitating adherence to regulatory
standards.
Explainable AI (XAI)

Explainability :
● The EU AI Act highlights the critical role of
explainability in AI systems, particularly in high-risk
sectors such as healthcare and finance.
● Explainable AI techniques (XAI) provide insights that
humans can understand and address issues of bias
and accountability.
● This promotes trust, ensures compliance with
regulations, and aligns AI decisions with ethical and
societal standards.
Explainable AI (XAI)

• An explainable AI (XAI) is an AI over which it is


possible for humans to retain intellectual oversight.

• Rational = supported by reasons. XAI is rational if it


can support its decisions by reasons.
Explainable AI (XAI)

Rationale: Why should an AI explain itself?

Arguments for explainable AI :


● Weapons of Math Destruction: if an AI decision can have
severe negative consequences for somebody’s life, then they
should be permitted to know the reason for its decision.
● Bias & unfairness are harder to detect if the AI cannot explain
its decisions.
● Debugging: If a decision looks unreasonable to a human, is it
caused by a software bug, or a quirk of training data, or is it
correct despite being unreasonable?
Explainable AI (XAI)

Arguments against explainable AI :


● A decision may be correct even when it cannot be explained in
detail.
● Requiring AI to produce only explainable decisions may reduce
its accuracy.
Explainable AI (XAI)

● Many of the popular AI systems


operate in ways that are opaque to
both those providing AI systems
(‘providers’), those deploying AI
systems (‘deployers’), and those
affected by the use of AI systems.

● This phenomenon is commonly


referred to as the “black box” effect.
Explainable AI (XAI)

Interpretability vs. Explainability

● Interpretability: "level of understanding how the underlying


(AI) technology works"
● Explainability: "level of understanding how the AI-based
system came up with a given result"

Source: https://en.wikipedia.org/wiki/Explainable_artificial_intelligence
Explainable AI (XAI)

Regulatory frameworks:
Regulations about explainability seek to avoid the harms of
unexplained decisions by granting individuals a ”right to an
explanation.”

● Europa: The EU Artificial Intelligence Act


https://ec.europa.eu/commission/presscorner/detail/en/qanda_21_1683
https://artificialintelligenceact.eu/high-level-summary/
Explainable AI (XAI)

● Compliance with the EU AI Act requires a strong


emphasis on transparency and explainability to
ensure that AI systems are both trustworthy and
comprehensible.

https://artificialintelligenceact.eu/high-level-summary/
Explainable AI (XAI)

Four-point summary
The EU AI Act categorizes AI into four levels of risk: unacceptable,
high risk, limited risk, and minimal risk.

● Unacceptable AI, such as social scoring and manipulative


systems, is prohibited.

● High-risk AI systems are subject to strict regulations that impose


significant compliance requirements on providers (developers) and
some obligations on users (applicators).
Explainable AI (XAI)

Four-point summary (cont.)


● Limited-risk AI, including chatbots and deepfakes, is subject to
lighter transparency requirements, such as notifying users when
they are interacting with AI.

● Minimal-risk AI, which includes many common applications such


as video games and spam filters, is largely unregulated, although
this is changing with the advent of generative AI.
Explainable AI (XAI)

General purpose AI (GPAI):


GPAI model means an AI model, including when trained with a
large amount of data using self-supervision at scale, that displays
significant generality and is capable to competently perform a
wide range of distinct tasks regardless of the way the model is
placed on the market and that can be integrated into a variety of
downstream systems or applications.
Explainable AI (XAI)

General purpose AI (GPAI):


GPAI models present systemic risks when the cumulative
amount of compute used for its training is greater than
10^25 floating point operations (FLOPs).
Explainable AI (XAI)

What are FLOPs (the size of floating point operations numbers)?:

FLOPs reflect how many addition and/or multiplication


operations are done in the process of training an AI model.
Quite arguably, one addition and one multiplication can be
totaled as one FLOP. A large number of FLOPs is a direct
indication that the model is very large, and the capabilities
and risks are usually large as well. Legislators across
countries are concerned primarily with the risk of large
models, so FLOPs are used as a regulatory threshold.
Explainable AI (XAI)

End of Lesson 12

THANK YOU !

You might also like