(IME 2022/2023 - 2 fase)Text 1XAI-Explainable arti

(IME 2022/2023 - 2ª fase)

 

Text 1

 

XAI-Explainable artificial intelligence

Gunning, D., Stefik, M., Choi, J., Miller, T., Stumpf, S. e Yang, G-Z

 

     Recent successes in machine learning (ML) have led to a new wave of artificial intelligence (AI) applications that offer extensive benefits to a ___(21)___ range of fields. However, many of these systems are not able to explain their ___(22)___ decisions and actions to human users. Explanations may not be essential for certain AI applications, and some AI researchers argue that the emphasis on explanation is misplaced, too difficult to achieve, and perhaps unnecessary. However, for many critical applications in defense, medicine, finance, and law, explanations are essential for users to understand, trust, and effectively manage these new, artificially intelligent partners. Recent AI successes are largely attributed to new ML techniques that construct models in their internal representations. These include support vector machines (SVMs), random forests, probabilistic graphical models, reinforcement learning (RL), and deep learning (DL) neural networks. Although these models exhibit high performance, they are opaque in terms of explainability. There may be inherent conflict between ML performance (e.g., predictive accuracy) and explainability. Often, the highest performing methods (e.g., DL) are the least explainable, and the most explainable (e.g., decision trees) are the least accurate.

     The ___(23)___ of an explainable AI (XAI) system is to make its behavior more intelligible to humans by providing explanations. There are some general principles to help create effective, more human-understandable AI systems: The XAI system should be able to explain its capabilities and understandings; explain what it has done, what it is doing now, and what will happen next; and disclose the salient information that it is acting on.

     However, every explanation is set within a context that depends ___(24)___ the task, abilities, and expectations of the user of the AI system. The definitions of interpretability and explainability are, thus, domain dependent and may not be defined independently from a domain. Explanations can be full or partial.

     Models that are fully interpretable give full and completely ___(25)___ explanations. Models that are partially interpretable reveal important pieces of their ___(26)___ process. Interpretable models obey “interpretability constraints” that are defined according to the domain, whereas black box or unconstrained models do not necessarily obey these constraints. Partial explanations may include variable importance measures, local models that approximate global models at specific points and saliency maps.

     XAI assumes that an explanation is ___(27)___ to an “end user” who depends on the decisions, recommendations, or actions produced by an AI system yet there could be many different kinds of users, often ___(28)___ different time points in the development and use of the system. For example, a type of user might be an intelligence analyst, judge or an operator. However, other users who demand an explanation of the system might be a developer or test operator who needs to understand where there might be areas of improvements. Yet another user might be policy-makers, who are trying to ___(29)___ the fairness of the system. Each user group may have a preferred explanation type that is able to communicate information in the most effective way. An effective explanation will take the target user group of the system into account, who might vary in their background knowledge and needs for what should be explained.

     A number of ways of evaluating and measuring the effectiveness of an explanation have been proposed, however, there is currently no common means of measuring if an XAI system is more intelligible to a user than a non-XAI system. Some of these measures are subjective measures from the user’s point of view, such as user ___(30)___ which can be measured through a subjective rating of the clarity and utility of an explanation. More objective measures for an explanation’s effectiveness might be task performance, i.e., does the explanation improve the user’s decision-making. Reliable and consistent measurement of the effects of explanations is still an open research question. Evaluation and measurement for XAI systems include valuation frameworks, common ground, common sense, and argumentation.

(. . . )

     From a human-centered research perspective, research on competencies and knowledge could take XAI ___(31)___ the role of explaining a particular XAI system and helping its users to determine appropriate trust. In the future, XAIs may eventually have substantial social roles. These roles could include not only learning and explaining to individuals but also coordinating with other agents to connect knowledge, developing cross-disciplinary insights and common ground, partnering in teaching people and other agents, and drawing on previously discovered knowledge to accelerate the further discovery and application of knowledge. From such a social perspective of knowledge understanding and generation, the future ___(32)___ XAI is just beginning.

 

Adapted from: Science Robotics in [Accessed on 15th April 2022].

 

Text 2

 

Overview of current additive manufacturing technologies and selected applications

Horn, T..J. e Harrysson, O.L

 

1	Three-dimensional printing or rapid prototyping are processes by which components are fabricated directly
	from computer models by selectively curing, depositing or consolidating materials in successive layers. These
	technologies have traditionally been limited to the fabrication of models suitable for product visualization but, over
	the past decade, have quickly developed into a new paradigm called additive manufacturing.
5	It remains to be seen what the long term implications of additive manufacturing will be. In many regards, it is
	a technology that is still in its infancy and it represents a very small segment of manufacturing overall. That small
	segment is growing quickly but the future is by no means certain. Scarcely a quarter century has passed since
	the first stereolithography systems for rapid prototyping appeared on the market.
	In that short time, additive manufacturing has not only become relatively common place in science, academia,
10	and industry, but it has also evolved from a method to quickly produce visual models into a new manufacturing
	paradigm. In the past two decades, revenues associated with products and services show that additive manufac-
	turing has grown into a multi-billion dollar industry
	Additive manufacturing has the potential to radically change the way in which many products are made and
	distributed. Throughout history, key innovations in manufacturing technology have had a profound impact on our
15	society and our culture. An examination of the applications and technologies suggest that additive manufacturing
	may become a truly disruptive technology.
	Prior to the industrial revolution goods were typically produced by skilled artisans and were often tailored
	to satisfy a specific, individual demand. While this approach may have had many inherent advantages to the
	consumer (i.e. high quality, custom parts on demand) it is doubtful that system could have persisted under the
20	growing demands of society.
	The invention of the first machine tools (that is tools capable of precisely controlling the relative motion
	between a tool and a work piece) along with advances in fixturing and metrology facilitated the manufacture
	of interchangeable parts which, in turn, supported the development of the mass production system. The model of
	mass production also has many clear advantages to both the producers and the consumers of products, including;
25	high throughput, high quality and product consistency at a low unit cost. This, of course, comes at the cost of
	reduced product diversity.
	In the last century, the means by which many goods are manufactured has been radically enhanced by
	computer controlled machinery and automation. However, in general, the basic methods and materials are quite
	similar to those used at the turn of the 19th century. Bulk materials must still be either cut, formed, or molded in
30	order to fabricate value-added products. In fact, a large portion of the products that we consume or use at the
	present time are manufactured using processes like forming, injection molding, casting, extrusion, stamping, and
	machining. Each one of these processes requires some form of tooling (mold, die, flask, stamp, fixture, etc.).
	For instance, if we consider casting an exhaust manifold in steel we must first design and fabricate a sand or
	investment mold with the negative shape of the final part. A metal stamped part, as simple as a washer, requires
35	a die and a large stamping press in order to be produced. A simple plastic cover for a smart phone requires an
	injection mold that may cost thousands of dollars and an injection molding machine that may costs hundreds of
	thousands to millions of dollars. The cost and time dedicated to the design and fabrication of tooling that supports
	mass production represents a significant percentage of the total cost of a product.
	The natural result of high tooling costs is that within a given mass production system there is an inverse
40	relationship between the quantity of a product that is produced and the variety of product designs available.
	It is necessary that we recognize that production tooling is not only expensive, but it also constrains the
	design of products based on innate limitations imposed by the various mass production processes. This is a
	widely studied area of manufacturing known as design for manufacture (DFM).
	As a brief example, consider a plastic injection molded part. One of the key limitations is that the mold must
45	provide for the easy removal of the part. This means that the part must have slightly outward sloping surfaces
	(called positive draft), as inward sloping surfaces would essentially lock the part to the mold like a dovetail making
	it impossible to remove. Further, the injection mold itself must be precisely machined, ground, and polished from
	a block of metal, and the processes that are used to do that, like milling with a cutting tool, also have similar
	limitations (i.e. the cutting tool must be able to access the feature that will be cut).
50	Increasing the complexity of the part to better serve a given function can drive up the cost of the tooling
	required for producing it and, in many cases, the optimal design for a given purpose is impossible to produce
	using traditional mass production methods
	Additive manufacturing represents a fundamentally new method of part fabrication. It is the process of
	fabricating components directly from 3D computer models by selectively depositing, curing, or consolidating
55	materials one layer upon the next. Each layer represents the cross-sectional geometry of the part at a given
	height. This is a stark contrast to traditional manufacturing processes like forming, casting, and machining
	because tooling is not required to produce a part. The freeform nature of additive manufacturing is therefore
	changing the way we look at traditional DFM constraints. In many cases the traditional constraints no longer
	apply.
60	By building parts additively, in layers, components can be manufactured with extremely complex geometries,
	such as internal channels, undercut features, or engineered lattice structures with controlled and/or variable
	porosity. These are features that are extremely difficult or impossible to produce with traditional methods.
	The implication of this is quite simple to recognize but at the same time has a profound result. Removing the
	need for tooling facilitates the economical production of small lot sizes of parts (as low as one) without sacrificing
65	interchangeability, thereby reducing the lead time for production (because the tools do not need to be produced),
	allowing flexibility in the supply chain and the production location (parts can be made where and when they
	are demanded), and raising the possibility of transitioning from a system of mass production to one off mass
	customization. It also means that design changes incur much less cost in production so products can potentially
	be customized to conform to the needs of the individual consumer. In many ways this concept goes far beyond
70	the definition of most existing mass customization models in which mass produced components are fabricated
	and then assembled on demand to specific customer orders.

 

Adapted from: Sage Journals. Available at: <https://journals.sagepub.com/doi/abs/10.3184/003685012X134209844630>[Accessed on 10th March 2022].

 

The main idea presented in Text 1 and Text 2 is related to: