Manufacturing is one of the highest risk industrial sectors to be working in with more than 3,000 major injuries and nine fatalities occurring each year. The involvement of robots in high-risk jobs can help manufacturers reduce unwanted accidents
Safety is the most important in manufactory. Since AI is used to operate machines, then workers are greatly protected when facing accidents.
Quality assurance is the maintenance of a desired level of quality in a service or product. Assembly lines are data-driven, interconnected, and autonomous networks. These assembly lines work based on a set of parameters and algorithms that provide guidelines to produce the best possible end-products. AI systems can detect the differences from the usual outputs by using machine vision technology since most defects are visible. When an end-product is lower quality than expected, AI systems trigger an alert to users so that they can react to make adjustments.
Quality assurance is very important in manufactory. AI is specifically suitable for this job because AI could always perform precisely as long as AI learns how to measure the product quality.
I can program a robot to pick up a three-ounce, travel-sized bottle of shampoo,” Vause went on, “but if I asked that same robot to pick up a different-sized bottle of shampoo, it’s a fundamentally different object.” Robots also struggle with variable outdoor lighting. In the real world, Vause added, “variation is the enemy of the robot.”
In the real world variation is more often than stability. Robots still could not handle those situations.
If the future of fruit-and-vegetable farming is automation, farmers will not only need the machines, and the funds to afford them, they will also require a new class of skilled farm workers who can debug the harvesters when something goes wrong.
It is interesting. Though robots could pick ripe strawberries, farmers still need skilled workers to ensure the performance of those harvesters.
The company has hitched its future to artificial intelligence — whether with its voice-enabled digital assistant or its automated placement of advertising for marketers — as the breakthrough technology to make the next generation of services and devices smarter and more capable
Though AI helps a lot in the society, companies could not depend it in every aspect since during the learning process AI may learn some bias in the dataset and will result in some problems in the society.
Sundar Pichai, chief executive of Alphabet, Google’s parent company, has compared the advent of artificial intelligence to that of electricity or fire, and has said that it is essential to the future of the company and computing. Earlier this year, Mr. Pichai called for greater regulation and responsible handling of artificial intelligence, arguing that society needs to balance potential harms with new opportunities.
AI is a great invention in the 21 century. High tech companies and AI do need great supervision and regulation in order to reduce the potential harms to the society.
By concentrating on racial and ethniccategories, this study aims also to reunite two historiographical fields that in theUnited States are seen as separate: studies on race and studies on immigrationand ethnicity
This is very interesting since in many cultures race and ethnicity are considered similar or even the same concepts. But the two terms are considered separately in US. That's may due to the fact that ethnicity is more related to immigration and hence researchers separate the two terms.
Thus, since the 2000 census, USresidents can choose to identify themselves with more than one race at a time,something that was not possible before.
If one individual could chose multiple races, then why not cancel the racial classification since not only individuals may not be aware of what races they belong to, also such a classification may discourage the communication between different individuals.
The disappearance of the “mulatto” category, followed by the introductionand forcible elimination of the “Mexican” category, shows that over the firstdecades of the twentieth century, the system of racial classification was modi-fied, primarily to take into account the cooperation, or lack thereof, of the pop-ulation.
It's great to see that the racial classification is being modified. As more individuals will tend to marry with different background people, the gap between different races is fading away, and hence a racial classification will become less applicable since it will discourage people from communicating.
that is, in the language of the census at that time, all those races that were neither black nor white.
Mexicians should be considered separately because their history does not coincide with other existing races.
While some of this energy comes from renewable sources, orcloud compute companies’ use of carbon credit-offset sources, theauthors note that the majority of cloud compute providers’ energy isnot sourced from renewable sources and many energy sources in theworld are not carbon neutral.
This is a very interesting point that training a large language model will consume a lot of energy. This may raise questions about how to reduce the energy consumption when training such models? Could researchers modify the model structure and improve the computation capability in order to reduce the consumption?
The larger question, of course, is whether algorithms should be used to determine addiction risk at all.
Maybe algorithms could be used but not be determinative. But they must be under supervised and controlled.
The two canines had been prescribed opioids, benzodiazepines, and even barbiturates by their veterinarians. Prescriptions for animals are put under their owner's name. So to NarxCare, it apparently looked like Kathryn was seeing many doctors for different drugs, some at extremely high dosages. (Dogs can require large amounts of benzodiazepines due to metabolic factors.)
This sounds ridiculous, and people will feel insecure since the pets' prescriptions information is shared and mistakenly viewed as humans' prescriptions without letting people know.
Its newest language app, which just reached the demo stage, is intended to help current speakers refine pronunciation and remove some of the influence of English.
A good news to see that Maori language is returning to its origin with less influence of English.
Selling or giving away the data invites western corporations to mine their language – and the thousands of years of traditional knowledge therein – for commercial opportunity, Jones says. It would mean entrusting data scientists with no connection to the language to develop the very tools that will shape the future of the language
If a language tool designed for a specific language but is not developed by native speakers (or worse by people who do not speak) will harm the language a lot. Also, it may result in the future that this language will be modified as languages developers are familiar with.
Selbst recommends proceeding with caution: “Whenever you turn philosophical notions of fairness into mathematical expressions, they lose their nuance, their flexibility, their malleability,” he says. “That’s not to say that some of the efficiencies of doing so won’t eventually be worthwhile. I just have my doubts.
Maybe a transformation from abstract concepts into mathematical expressions will lose some flexibility, but mathematical rigorousness will also be introduced into practice. From this aspect, proceeding with caution is necessary.
First, it would require companies to audit their machine-learning systems for bias and discrimination in an “impact assessment.” Second, it doesn’t specify a definition of fairness.
Maybe not a so called "clear" definition for fairness will sometimes act well for the machine-learning system. Companies may design such unsupervised/semi-supervised learning to let machine know the algorithms like the examples in this article.
For example, Amazon warehouses are automated to a significant degree, but they are not fully automated. Humans and machines work together and many crucial tasks, such as delivery, are still completed entirely by humans.
Maybe in the future a potential solution towards this situation is that we let machines to do what shows the most professions and we humans to act as their supervisors. Then though machines take some jobs but new job positions are available for humans.
The idea that artificial neural network architecture (and with it, “deep learning”) is the breakthrough technology for creating conscious, or even sentient, machines fuels the looming fear of robots taking our jobs.
May be now neural networks still focus on some specific areas, but in future we still need to consider how to deal with the situation that AI may take our jobs.
he government wants to know about mypurchases, my web visits, my physician visits, and my days off fromwork for a hypothetical possibility that is unlikely to ever occur. It ismore likely that the public health department will use the information tomake me exercise more, eat less sugar, and stop smoking
Though it may help people become healthier, some information like web visits is not necessary and may harm people's privacy.
“Sharing my personal information without my consent is as much aviolation of my dignity as a physical invasion of my body. It is irrelevantwhether the physical invasion caused me physical pain,” a medicalethicist responds.
If transparency could not be ensured during sharing personal information, then the insecurity will surpass the potential benefits brought from the data collection.
we need socialscience and humanities scholars who are able toactively engage in data and computer sciencepractice
Those scholars with social science background will learn more about how inequality in technology will rise when engaging in data and computer science practice and could give their suggestions more precisely.
We need a clearer picture of the terms thatare at stake and currently do important politicalwork, because the unclarity about key terms(such as ‘algorithm’, ‘digitization’, ‘machinelearning’ and so on, but also ‘fairness’, ‘bias’,‘standardization’, ‘accountability’) impacts ourability to have more productive conversationsabout the abilities and limits of new technolo-gies, and explore regulatory possibilities
A clear clarity of those terms will not only help promote equality in technology practice but also a big benefit for the researchers side.
Furthermore, it is undeni-able that the black box nature of data collection in theworkplace serves to stymie any attempts at worker con-trol or agency over data gathered in the workplace.Thus, workers are left at the mercy of shadowy databrokers, since their workplace data may be sold with-out their knowledge or consent
Transparency is required during the data collection process for workers. And also, private data could be sold unconsciously will let workers feel insecure and discourage them form working.
A consumer score may, without any public notice, rely on an underlying factor or attribute that has discriminatory implications (e.g., race or gender) or that most consumers consider sensitive (e.g., health or financial).
Without being noticed, consumers will feel insecure when facing some situations that may be measured with those sensitive information.
Anyone using a consumer score in a way that adversely affects an individual’s employment, credit, insurance, or any significant marketplace opportunity must affirmatively inform the individual about the score, how it is used, how to learn more about the score, and how to exercise any rights that the individual has.
This is a good point. But how to measure the adversely affects in a reasonable way should be considered carefully.
Another category of problem in studying movement through remote or low-income areasis locational accuracy: given that the locations of infrequent callers are updated less often thanthose of frequent users (Bengtsson et al., 2011), those with less resources to buy calling credit,or to charge their phone’s battery, are less visible. This is a particular problem in the case offorced movement where people may become unable to recharge their phones as they move. Arelated problem can be identified where a phone user may run out of credit or battery powerwhile moving, and effectively be pinned to the map at the last place their phone made contactwith an antenna. Thus, people may be only fuzzily visible or may be first visible and theninvisible – demanding that the researchers come up with a way of dealing with missingquantitative data in a context that calls for purely qualitative information.
This issue may also result in high bias in the dataset. Maybe researchers could do a survey first to know the frequency of using mobile phones. Then based on the survey result they could sample the data from different groups.
Another dimension of numerical accuracy is the problem of multiple identities within thedataset: it is risky to assume that one signal represents one individual, since in many places alack of coverage can result in people having multiple SIM cards from different providers toget the best chance of a signal in remote locations (Bengtsson et al., 2011). Equally, one SIMcard can have multiple users. This can mean that if using data from multiple providers,several different profiles may represent a single user, or the reverse.
This issue may also lead to bias in the dataset. If the multiple identites problem is severe, it may result in no availability to use the dataset.
It is well established by now that large LMs exhibit various kinds ofbias, including stereotypical associations [11, 12, 69, 119, 156, 157],or negative sentiment towards specific groups [61]
The bias is inevitable since even if the language sample is equally distributed the bias will also appears in analyzing the grammar or POS tagging (e.g. doing could be considered as verb or adj) or some words have different meanings but still could considered reasonable in one sentence.
This behavioral data is fed to machine learning systems that provide predictions about what people will do in the future. She documents how surveillance capitalists have gained immense wealth through the trading of “prediction products,” as companies profit from laying accurate bets on people’s future behaviors. These systems tend to reward the privileged while entrapping the underprivileged, whose choices are particularly constrained.
Indeed. The machine learning system will tend to learn the most from people's initial wealth. I think we may combine other facts as input (like education background, occupation etc. ) to the machine learning systems to weaken the effect of initial wealth.
And considering that under-resourced languages have never been and are unlikely to become a major focus of large tech corporations, there is a lot of work for volunteers and passion-driven advocates and technologists to do. But what are the risks and drawbacks, if any, of bringing a language, particularly minority and Indigenous languages, into the digital sphere?
Those under-resourced languages should be respected and protected. Though there are potential risks of bring a language into digital sphere, I believe that building a database for those languages is a practical method to project the minority languages.
As Burawoy (1983), noted in his seminal ethnography,the most visible control technology in a factory was theassembly line. But that form of control was primarilyabout the standardization of productivity and, second-arily, about worker surveillance. The new technologiesof work afford both in equal measure.
Indeed. The new technologies force workers not only to behave well in work time, but also somehow focus on work in their spare time.
If the New Jim Code seeks to penetrate all areas of life, extracting data, producing hierarchies, andpredicting futures, thin description exercises a much needed discretion, pushing back against theall-knowing, extractive, monopolizing practices of coded inequity.
In this way, thinness analysis seems to provide us with a flexible but not aggressive analytic method to study the real connection. However, unlike a thick analysis, thinness will tend to seek less information and hence be encouraged to enhance coded equity and individual privacy.
Most important, then, is the fact that, once something or someone is coded, this can be hard to change.Think of all of the time and effort it takes for a person to change her name legally. Or, going back toCalifornia’s gang database: “Although federal regulations require that people be removed from thedatabase after five years, some records were not scheduled to be removed for more than 100 years.”
This is indeed that we human beings will tend to persist our own judgement on the first impression. For a machine, we could first assign a prior distribution (for example the coding in our article) to it. Then we will let machines combine the prior with the coded person/item's present behaviors to recompute the posterior distribution and then recode the person/item. However, it's hard for our human beings to reconsider the posterior distribution for them since we tend to be less convinced by a posterior impression.
The first fundamental flaw of individualism is that it justifies inequality, as illustrated in the “10 Dollar Problem”.62 By incorrectly assuming a pre-existing state of equality, individualism creates inequality by rewarding differences that can be leveraged competitively. This leads to a “winner takes all” attitude, a race to the bottom. The Silicon Valley culture of disruption is a “winner takes all” culture. Speed is treated as an indispensable competitive advantage and precautions in terms of diversity of startups or the societal consequences are treated as secondary. Often private interests win and the public, especially the marginalized, loses. This leads to the belief that those who are poor are poor because they deserve to be poor. As autonomous beings, those with less must have failed in one aspect or another and thus are justly rewarded.
This statement is interesting. In our society we do use the pre-existing state to measure "winner takes all" which will result in inequality. It indicates that though using prior (distribution) to obtain some pre-information is practical, it will sometimes result in huge incorrectness. We still need to combine the pre-state with the posterior distribution to make a conclusion.
The significance of Gödel’s quote is that a human mind is not necessarily superior to a machine, since human mathematical intuition, formal logic, which corresponds to a consistent formal system, is just as incomplete as a computer that cannot always give reason as to why it reaches a certain result.
This opinion is quite interesting. But I am still confused about the statement that human mind is not superior to a machine. Since machines are invented by human, we could view machines as the "Corollary" of human. So even our system is neither complete nor self-consistent, could we still view human being as superior since "corollaries" could be considered to be proved by its prerequisites: human?
It would also publicly and transparently make specific recommendations, and Google would tell us whether they’d followed them, and why.
Indeed. If technology companies are supervised by the public and tell the public regularly about what they are doing, I believe those companies will focus better on how to figure out AI ethics under public's supervision.
Many of Google’s AI researchers are active in work to make AI fairer and more transparent, and clumsy missteps by management won’t change that. The Google spokesperson I talked to pointed to several documents purportedly reflecting Google’s approach to AI ethics, from a detailed mission statement outlining kinds of research they will not pursue to a look back, at the start of this year, at whether their AI work so far is producing social good to detailed papers on the state of AI governance.
We need to standardize the measurement of the fairness & transparency of AI. In essential AI is invented to help people finish some complicated calculations. If we do not restrict the usage of AI, it may challenge our privacy.
Twitter does not represent ‘all people’, and it is an error to assume ‘people’and ‘Twitter users’ are synonymous: they are a very particular sub-set. Neither isthe population using Twitter representative of the global population. Nor can weassume that accounts and users are equivalent. Some users have multipleaccounts, while some accounts are used by multiple people. Some peoplenever establish an account, and simply access Twitter via the web. Some accountsare ‘bots’ that produce automated content without directly involving a person.Furthermore, the notion of an ‘active’ account is problematic. While some userspost content frequently through Twitter, others participate as ‘listeners’ (Craw-ford 2009, p. 532). Twitter Inc. has revealed that 40 percent of active users signin just to listen (Twitter 2011). The very meanings of ‘user’ and ‘participation’and ‘active’ need to be critically examined
Could we apply some methodology to exclude some invalid accounts such as view those users that do not tweet frequently as less reliable accounts? And then apply bootstrap or other methodology to build a bridge between "all people" and "Twitter users"?
In addition to this question, there is the issue of data errors. Large data sets fromInternet sources are often unreliable, prone to outages and losses, and theseerrors and gaps are magnified when multiple data sets are used together.Social scientists have a long history of asking critical questions about the collec-tion of data and trying to account for any biases in their data (Cain & Finch 1981;Clifford & Marcus 1986). This requires understanding the properties and limitsof a data set, regardless of its size.
Sometimes we collect the real-time data that is varying, and hence we could not ensure the reliability of the collected data since we need to analyze it immediately to obtain some results. How could we deal with such data?