AI (143)
Find narratives by ethical themes or by technologies.
FILTERreset filters-
- 7 min
- MIT Technology Review
- 2020
This article details a new approach emerging in AI science; instead of using 16 bits to represent pieces of data which train an algorithm, a logarithmic scale can be used to reduce this number to four, which is more efficient in terms of time and energy. This may allow machine learning algorithms to be trained on smartphones, enhancing user privacy. Otherwise, this may not change much in the AI landscape, especially in terms of helping machine learning reach new horizons.
- MIT Technology Review
- 2020
-
- 7 min
- MIT Technology Review
- 2020
Tiny four-bit computers are now all you need to train AI
This article details a new approach emerging in AI science; instead of using 16 bits to represent pieces of data which train an algorithm, a logarithmic scale can be used to reduce this number to four, which is more efficient in terms of time and energy. This may allow machine learning algorithms to be trained on smartphones, enhancing user privacy. Otherwise, this may not change much in the AI landscape, especially in terms of helping machine learning reach new horizons.
Does more efficiency mean more data would be wanted or needed? Would that be a good thing, a bad thing, or potentially both?
-
- 7 min
- The New Republic
- 2020
The narrative of Dr. Timnit Gebru’s termination from Google is inextricably bound with Google’s irresponsible practices with training data for its machine learning algorithms. Using large data sets to train Natural Language Processing algorithms is ultimately a harmful practice because for all the harms to the environment and biases against certain languages it causes, machines still cannot fully comprehend human language.
- The New Republic
- 2020
-
- 7 min
- The New Republic
- 2020
Who Gets a Say in Our Dystopian Tech Future?
The narrative of Dr. Timnit Gebru’s termination from Google is inextricably bound with Google’s irresponsible practices with training data for its machine learning algorithms. Using large data sets to train Natural Language Processing algorithms is ultimately a harmful practice because for all the harms to the environment and biases against certain languages it causes, machines still cannot fully comprehend human language.
Should machines be trusted to handle and process the incredibly nuanced meaning of human language? How do different understandings of what languages and words mean and represent become harmful when a minority of people are deciding how to train NLP algorithms? How do tech monopolies prevent more diverse voices from entering this conversation?
-
- 12 min
- Wired
- 2018
This video offers a basic introduction to the use of machine learning in predictive policing, and how this disproportionately affects low income communities and communities of color.
- Wired
- 2018
-
- 12 min
- Wired
- 2018
How Cops Are Using Algorithms to Predict Crimes
This video offers a basic introduction to the use of machine learning in predictive policing, and how this disproportionately affects low income communities and communities of color.
Should algorithms ever be used in a context where human bias is already rampant, such as in police departments? Why is it that the use of digital technologies to accomplish tasks in this age makes a process seem more “efficient” or “objective”? What are the problems with police using algorithms of which they do not fully understand the inner workings? Is the use of predictive policing algorithms ever justifiable?
How Cops Are Using Algorithms to Predict Crimes: How Cops Are Using Algorithms to Predict Crimes-
- 7 min
- Venture Beat
- 2021
As machine learning algorithms become more deeply embedded in all levels of society, including governments, it is critical for developers and users alike to consider how these algorithms may shift or concentrate power, specifically as it relates to biased data. Historical and anthropological lenses are helpful in dissecting AI in terms of how they model the world, and what perspectives might be missing from their construction and operation.
- Venture Beat
- 2021
-
- 7 min
- Venture Beat
- 2021
Center for Applied Data Ethics suggests treating AI like a bureaucracy
As machine learning algorithms become more deeply embedded in all levels of society, including governments, it is critical for developers and users alike to consider how these algorithms may shift or concentrate power, specifically as it relates to biased data. Historical and anthropological lenses are helpful in dissecting AI in terms of how they model the world, and what perspectives might be missing from their construction and operation.
Whose job is it to ameliorate the “privilege hazard”, and how should this be done? How should large data sets be analyzed to avoid bias and ensure fairness? How can large data aggregators such as Google be held accountable to new standards of scrutinizing data and introducing humanities perspectives in applications?
-
- 5 min
- MIT Tech Review
- 2020
With the surge of the coronavirus pandemic, the year 2020 became an important one in terms of new applications for deepfake technology. Although a primary concern of deepfakes is their ability to create convincing misinformation, this article describes other uses of deepfake which center more on entertaining, harmless creations.
- MIT Tech Review
- 2020
-
- 5 min
- MIT Tech Review
- 2020
The Year Deepfakes Went Mainstream
With the surge of the coronavirus pandemic, the year 2020 became an important one in terms of new applications for deepfake technology. Although a primary concern of deepfakes is their ability to create convincing misinformation, this article describes other uses of deepfake which center more on entertaining, harmless creations.
Should deepfake technology be allowed to proliferate enough that users have to question the reality of everything they consume on digital platforms? Should users already approach digital media with such scrutiny? What is defined as a “harmless” use for deepfake technology? What is the danger posed to real people in the acting industry with the rise of convincing synthetic media?
-
- 7 min
- VentureBeat
- 2021
New research and code was released in early 2021 to demonstrate that the training data for Natural Language Processing algorithms is not as robust as it could be. The project, Robustness Gym, allows researchers and computer scientists to approach training data with more scrutiny, organizing this data and testing the results of preliminary runs through the algorithm to see what can be improved upon and how.
- VentureBeat
- 2021
-
- 7 min
- VentureBeat
- 2021
Salesforce researchers release framework to test NLP model robustness
New research and code was released in early 2021 to demonstrate that the training data for Natural Language Processing algorithms is not as robust as it could be. The project, Robustness Gym, allows researchers and computer scientists to approach training data with more scrutiny, organizing this data and testing the results of preliminary runs through the algorithm to see what can be improved upon and how.
What does “robustness” in a natural language processing algorithm mean to you? Should machines always be taught to automatically associate certain words or terms? What are the consequences of large corporations not using the most robust training data for their NLP algorithms?