The concept of ‘hype’ is widely used in the business and public sphere and serves as a way to characterize increasing expectations of developments in technological fields. This paper seeks to analyze a ‘hype in the making’ by closing in on a case at the intersection of data science and energy. Following the previous body of literature qualitative as well as quantitative indicators are taken into account in order to assess the promises, hope and hype of the optimization of datacenters through machine learning. The analysis concludes that this techonogy is nearing its peak of expectation but shows favorable signs for activities after disappointment.
1. Abstract
2. Hyping Smart Energy
3. The Gartner Hype Cycle
4. Learning Energy
5. Cycle of Hype
6. Conclusion
7. Bibliography
1. Abstract
The concept of ‘hype’ is widely used in the business and public sphere and serves as a way to characterize increasing expectations of developments in technological fields. This paper seeks to analyze a ‘hype in the making’ by closing in on a case at the intersection of data science and energy. Following the previous body of literature qualitative as well as quantitative indicators are taken into account in order to assess the promises, hope and hype of the optimization of datacenters through machine learning. The analysis concludes that this techonogy is nearing its peak of expectation but shows favorable signs for activities after disappointment.
2. Hyping Smart Energy
What some call the fourth industrial revolution has propelled our society into a knowledge or information society that increasingly relies on computation to make decisions on the ever-expanding sets of data that are available in the age of digitization. At the center of this development is a new wave of artificial intelligence (AI) that allows ICT (Information and Communication Technology) to reshape the way we interact, learn, and make decisions both at a societal and individual level. This information explosion is made possible by exponential growth in computer technologies (Jasanoff, 2016b) combined with advances in software that allow computers to better cooperate with the world around them.
A lot of the promises made in the context of such powerful technological marvels are positive. As Jasanoff (2016b, p. 4) notes “technology and optimism fit together like hand in glove because both play upon open and unwritten futures, promising release from present ills.” Technological utopias are envisioned where robots and software free humanity from work and open up a new society where the individual can follow their true passions, focusing on the arts or other intellectual endeavors. To varying degrees previous technological advances have made similar claims, however few actually fulfill their original promises. In fact, whole industries are dedicated to study the uncertainties associated with technological development. The most prominent attempt at foreseeing the life cycle of technology is the Gartner Hype Cycle, which visualizes expectations of emerging technologies over time.
This essay seeks to entangle the inner workings of these technological cycles by exploring the hopes and promises of machine learning in self-regulating energy systems as an example of a technological hype at the intersection of data science and energy. Therefore, we first dive into hype cycles, followed by a breve discussion of the case and finally apply the analytical framework to the emerging technological niche of machine learning.
3. The Gartner Hype Cycle
Abbildung in dieser Leseprobe nicht enthalten
Figure 1: Hype Level and Engineering or Business Maturity from the Hype Cycle. (Dedehayir and Steinert, 2016)
The Gartner Hype Cycle is a visual representation of the maturity, adoption and social application of specific technologies, developed by the American ICT consultancy Gartner Inc. in 1995. The framework merges the expectations directed to a given technology, which are initially very high but decrease after disappointment, with the engineering or business maturity of the technology which follows the classical technology S-curve (Dedehayir and Steinert, 2016; Figure 1). The initial graph, which develops similar to a bell curve is the expression of the irrational, overconfident and uneducated expectation novel technologies can attract, which after an initial run up get confronted by engineering reality, which cannot follow up on the initial expectations. Fenn and Raskio (2008) find that this is due to the underlying human characteristics of attraction to novelty, social contagion and heuristic attitude in decision making. This phase of hype is crucial as it attracts a vast amount of attention and can lead to decisions that might seem unjustified when the real potential of any given technology becomes apparent later on.
This second phase of growth towards technological maturity follows the path of an S-curve. This is a classical conception of emerging technologies. In the beginning fundamental parameters are only understood by few people and only with limited detail. Through initial prototypes and pilots or early adopters progress is made until a certain threshold of knowledge, interest and dissemination is reached to facilitate a comparably rapid advance is made possible. After this rise in sophistication, the ‘low hanging fruits’ are fleshed out, development becomes harder again and physical or economical limitations kick in, which flattens out the rate of progress thereby completing the S-curve (Dedehayir and Steinert, 2016).
These two elements create the hype cycle shown in Figure 1, which is classically divided into five phases: innovation trigger, peak of inflated expectations, trough of disillusionment, slope of enlightenment, and plateau of productivity. Following Van Lente et al. (2013) this paper aims at combing both a quantitative and qualitative approach in the limits of the scope of this paper. Media attention is a viable indicator for the level of hype for a given technology. However, the quantitative level of coverage does not necessarily correlate with disappointment about the technological progress as highlighted by Ruef and Markard (2010, p. 210): “A phase of high media attention without highly optimistic expectations […] is not considered a hype and a drop of media attention has to be accompanied by clear changes at the level of expectations in order to mark a disappointment”. This is why a combination of both quantitative date of attention as well as qualitative statements about expectations are important to understand how society relates to a novel technology.
4. Learning Energy
Technology from the greek techne (skill) and logos (study of) originally was used to describe the study of skilled craft and only was connected to objects in the last century (Jasanoff, 2016). In contrast, today, technology often gives rise to images of computers, phones and other silicon based electronics. This semantic change is surely connected to the impact that ICT’s have had on our everyday life. Over the last decades computational power increased exponentially, making things possible that were unimaginable half a century earlier. However, these strides were made possible due to rule based software algorithms that could out-calculate the best chess players in the world by many orders of magnitude, but needed to be fed with vast libraries of rules made and coded by humans (Bostrom, 2014). The late nineties, early two-thousands saw a paradigm shift in the way sophisticated software is created, by the practical use of neural networks, a form of machine learning. These networks – a collection of interconnected small units loosely comparable to neurons of the human brain1 – have the crucial difference that they learn by being trained rather than explicitly programmed (Stone et.al., 2016). Such systems crawl through training data sets and learn to recognize patterns which allows for up until now unprecedented capabilities like recognizing images, utilize language and drive cars.
But this form of machine learning cannot just be applied to new products but also innovate existing infrastructure. The lifeblood of modern society as well as the most essential component of digital technologies is electricity. Various innovations and innovators as well as institutions are trying to reshape private and public energy networks into smart grids. Through various technologies (big) data about energy usage is collected in the hopes of finding ‘smarter’ ways of allocating this precious resource. However, in order to find such ‘smart’ ways of managing energy, data itself is not enough. The collected information has to be processed to create these ‘smart’ insights that change the way we interact with energy.
Up until recently either rule-based algorithms, humans or a combination of both were the deciding element in the equation of managing energy. The advances made in machine learning, are promising to change this. With the help of intelligent self-learning software, technologists are hoping to make sense of the panoply of information we collect about our energy networks and utilize this to create a form of ‘smart energy’.
5. Cycle of Hype
Abbildung in dieser Leseprobe nicht enthalten
Figure 2: Efficiency increase due to Machine Learning. (Deepmind, 2016)
After discussing the hype framework and relevance of the case at hand, the following section seeks to apply the conceptual tool to smart energy (machine) learning algorithms. A good example what in the Gartner Hype Cycle is called a first-generation product is Google Deepmind’s datacenter optimization application. The pioneering nature of these datacenter optimizers is highlighted by a statement from Deepmind Co-founder Mustafa Suleyman: “We think there's lots of potential to apply this to large scale energy distribution, so we're giving it some thought and are in early discussions with a number of people on that” (Brugges, 2016). Datacenters are a challenging environment, as the different elements interact in complex, nonlinear ways and traditional formula-based engineering or human intuition often has difficulties capturing these interactions. Each center operates in a unique manner and is subject to rapid and dynamic changes which make it “difficult for humans to see how all of the variables—IT load, outside air temperature, etc.—interact with each other” (Deepmind, 2016). Computers can mitigate this problem by analyzing the enormous amounts of data these centers generate in order to recognize patterns and ‘learn’ from them. Figure 2 shows how this impacted the efficiency of Google’s Datacenters, however, missing labels for example on the axes only let us conjecture ‘this is working’, which might hint at a not fully developed technology or confidence in the technology.
Finding a niche use case for emerging technology is important, but what is particularly interesting is the expectations that come with it. The whitepaper outlining the use case makes the expected importance apparent “At this scale, even relatively modest efficiency improvements yield significant cost savings and avert millions of tons of carbon emissions” (Gao, 2016). In other words, this technology will play a crucial part in saving the planet. The initial statement made by Google Deepmind emphasizes the high ambitions of the project even more directly: “From smartphone assistants to image recognition and translation, machine learning already helps us in our everyday lives. But it can also help us to tackle some of the world’s most challenging physical problems -- such as energy consumption” (Deepmind, 2016).
Abbildung in dieser Leseprobe nicht enthalten
Figure 3: Global Relative Google Search interest, 100 = highest (Google Trends, 2018)
‘Tackling energy’ in this context is not just a way to become more efficient or economical. Rather it is a promise to solve humanities toughest problems: “we can use machine learning to consume less energy and help address one of the biggest challenges of all -- climate change” (Deepmind, 2016).
Promises and expectations are high, if we look closely into the discourse surrounding machine learning and energy. A look at the quantitative side of the hype will give a clearer picture by revealing attention patterns through a keyword search in Google Trends. Indeed, the Google Trends analysis gives a picture of an exponentially increasing search interest in the term ‘Machine Learning’. Figure 3 visualizes that an increasing amount of people around the world want to know more about machine learning. Other search queries like Neural Network or Deep Learning2, which are highly connected to the term Machine Learning, show a similar pattern3.
[...]
1 Each node receives a signal either directly from the input or other nodes, computes a weighted sum and if a certain threshold is reached it sends an output signal. These weights and thresholds are trained by looking at their contribution to the difference between the final network output and the correct answer and changing them accordingly.
2 Simple neural networks have been developed since the 1950s, however the use of multi-layered ‘deep’ neural networks have driven the most significant machine learning advances of the last decade (Bostrom, 2014). Although neural networks are not strictly synonymous deep learning, they closely relate and are often used interchangeably.
3 The global Search Interest for Machine Learning dropped more than 50% in one month in August 2017 (Figure 3). This is a very unusual and significant drop in an extremely short time span. The data from closely related search terms suggests interest in the topic is still at an all time high. Additional material (Figure 4) also suggests that a worldwide drop this significant in only one month does not correlate with changing attention patterns. Furthermore, other additional search queries (e. g. Artificial Intelligence, Reinforcement Learning) of related topics could not reproduce a drop that comes even close throughout the 7-year timeline. As other data points do not support the drop in interest, the likelihood of it being an artefact of the Google Trends API is high and a drop in actual attention patterns very unlikely. However, this could not be confirmed.
- Quote paper
- M.A. Stefan Raß (Author), 2020, Learning Energy. Promises, Hope and Hype in the Context of Machine Learning, Munich, GRIN Verlag, https://www.grin.com/document/1001869
-
-
-
-
Upload your own papers! Earn money and win an iPhone X. -
-
Upload your own papers! Earn money and win an iPhone X. -
-
Upload your own papers! Earn money and win an iPhone X. -
-
Upload your own papers! Earn money and win an iPhone X. -
-
Upload your own papers! Earn money and win an iPhone X.