Final Assignment (Risa Takemura)
use of AI for a healthy social society
71745020 Risa Takemura
Interactive Design (Spring 2020)
My contributions:
- simple rock-paper-scissors
glitch project: https://glitch.com/~test-my-rock-pa-scissors
website: https://test-my-rock-pa-scissors.glitch.me/
2. (group proj)facial expression detector & magnification video
glitch project: https://glitch.com/~emoji-proj-interactive-design
website: https://emoji-proj-interactive-design.glitch.me/
3. (group proj) this manga may exist by GAN
glitch project: https://glitch.com/~mangamayexist
website: https://mangamayexist.glitch.me/
Our Well-beings and AI
As we have discussed biases and creativity being very closely linked with AI, I have realized how important it is to contemplate where it is utilized and how humans perceive the generated outputs. I found that similar debates occur in AI technology as with our real human society. In our first reading, the field of artificial intelligence is defined as “the effort to automate intellectual tasks normally performed by humans.” Growing beyond simple automation, AI technology today has been performing the tasks performed unconsciously by humans, such as gender/race/object identification and classification. The common problem as I have discussed in the gender bias assignment is the unconscious tendency to classify people into traditional boundaries, which transfers into AI products, without the intent of discrimination. I believe that as important it is to enhance AI machines for the general population, we must look into areas of AI applicability to the minority groups in equal advantage for the technology.
Inspired by Google’s Project Euphonia video, I believe that the most beneficial field of AI application is for the disabled, who cannot do the “unconscious tasks” the general healthy population, such as clear speech, hearing, walking, etc. As seen with my final group project using the RunwayML GAN training, the generated output greatly varies upon the training dataset. For example, the use of the bird illustration dataset as a pre-trained model restricted the color tone and object density in the generated images, while the use of human faces as a pre-trained model allowed more color variation, the addition of title space, and human-like or character-like figure in the image. This implies how concrete classification of images, such as “illustration,” cannot define to be the best fit option for machine learning. Moreover, we can confidently say, that trial-and-error is a necessity in the perfection and adequacy of such models. I believe that the same theory applies to other image/voice/motion recognition, especially utilized throughout realtime continuous human activities. The Project Euphonia video introduces a man who diligently feeds the speech recognition machine his speech recording, in hopes to expand its ability to pick up difficult speech patterns. Just because humans cannot easily recognize signals of communication, we do not have to eliminate that procedure for machine learning to achieve.
Counteracting bias in AI has the potential to enhance the quality of lives which are generally perceived as “difficult.” Personalizing AI systems may take time and great training data to be sufficient for use, but the combination of trained datasets may enhance the usability of such assistance. For example, focusing on patients who are unable to walk or talk, the combination of body sensors, brain waves, facial expression detectors, and the use of AI text generators may be effective for supported verbal communication. The population in which have speech deficit adds to more than 2 million people, in the United States. They require “adaptive alternative communication,” also known as AAC, in order to have sufficient communication with others. Additionally, many patients with the same diagnosis have similar speech struggles (stuttering, articulation, vocalization, etc). By integrating the datasets for similar conditions, we have the ability to build a condition-corresponding speech assistant. Narrowing the focus population of training data will optimize and widen the range in one’s freedom of expression.
In today’s AAC applications, AI is only viewed as a tool to bring ease in communication for the patients and family members. Although in such particular usage, AI stands as a “tool” of communication, it also is a creator of expression. As we have discussed in class the creativity in AI, humans all experience personal/psychological creativity and we can program machines to do the same. Within the 3 types of creativity, we can set the rules of boundaries to bring forth exploratory creativity to help users express their state of mind. In today’s digital platforms, most modes of assistance only provide a limited number of options. For example, the iPhone’s Siri voice has a limited variation based upon 2 genders and 21 languages. Although people may not feel the necessity of further vocal diversity, when we examine AAC applications which takes place of a person’s voice, the matter of choice influences individualism. If we fill a room with patients of speech deficits using AAC, we will only hear one or two types of voices. This is why output styles need not to be preset; AI can be the creativity leading diversity into the minority groups.
Where digital assistance is highly valued, individualism is a rising topic. Without limitation to just AI, accessible technology has allowed us to connect and communicate without much effort. On such platforms, technology makes it very easy for us to be both very different people and the same person. The traditionally set rules of the programs narrow the perspective into: “everyone fits the general population,” causing numerous debates around the notion of bias. On the other hand, the users who take advantage of the anonymity maximize the freedom of expression to the extremes of popularizing the catfishing trend. Identity within technology is a very controversial issue. Technology has the tendency to generalize the classification of identity, while humans have the tendency to desire multiple identities on the platform. The creative process of AI generation has the power to expand our choices of expression. As AI starts to satisfy more of our human “wants,” the key factor in creating an optimal system is the awareness of the limitations of the rules and training data we allow the machine. This understanding will greatly benefit the coming innovation in AI use for the well-being of minority groups.
