One more I found on my Kindle that I got about a third of the way into - Pastoral Virtues for Artificial Intelligence Care and the Algorithms that Guide Our Lives by Jaco Hamman
From the book --- "AI is ever evolving. While all AI depend on algorithms, statistical models, and the input of data, machine learning can be distinguished from deep learning and reinforcement learning as three subsets of AI. The context in which the word “AI” appears often indicates what type of AI is referred to, as the adjectives “machine” or “deep” “or “reinforcement” are not always used...
Deep learning is based on artificial neural networks (ANNs)—many computers and input sources (called perceptrons) connected to one another—where learning takes place in supervised and unsupervised ways. Neural networks seek to mimic human brain functioning. A perceptron informs classification and makes predictions possible. Examples of deep learning programs include speech recognition, board game software, and medical image analysis (reading x-ray films in radiology, for example). Another example is the motor control of robots, where a robot can navigate obstacles or keep its balance if pushed off-balance. Not only does the algorithm understand all the data gathered, but it also has to intelligently respond to that information. Deep learning has removed much of the need to speak to a live person when we call our banks or a large corporation. Error rates on speech recognition now are in the 5.5 percent range, the same error rate of person-to-person communication.
Deep learning empowers AI to see a dog—one not previously part of the algorithm through data input—and to recognize it as a dog. When one searches for a person named Bert Hathaway, deep learning anticipates you are not searching for Berkshire Hathaway, a process called entity disambiguation. Due to the neural networks, significant training data is used in deep learning. Deep learning AI has the capacity to grow its algorithm to be more efficient at what the algorithm was created to accomplish. This is a strong positive, as inherent biases and faults within the algorithms can be corrected by the program itself. It is no surprise that deep learning awakens hope in some and fear in others. Reinforcement learning builds upon machine and deep learning making experience-driven sequential decision-making possible.
Through reinforcement learning, AI can take actions in the real world without the need to have labeled data where the algorithm is explicitly taught the meaning of a piece of data. The program AlphaGo’s ability to play the game Go better than any human relies on reinforcement learning. The software that can drive an autonomous vehicle is built on reinforcement learning as decisions are made beyond the recognition of traffic signals. The fact that reinforcement learning empowers an AI to take independent actions in the real world raises questions about safety mechanisms and ethical frameworks that will keep the AI from causing harm.
The various types of AI can further be separated into object-specific AI, general-purpose AI (also referred to as artificial general intelligence; AGI) and artificial superintelligence (ASI). Object-specific AI is what we most readily find. It is the AI that has a specific function, such as playing a board game, managing financial systems or robots, or understanding language. Object-specific AI delivers a certain product and can be very good at doing so since computers are highly effective in doing repetitive tasks. General-purpose AI, however, builds on deep learning and reinforcement learning to create an AI that can figure out, by itself, how to do things.
Imagine an AI that creates a model for the most effective ways to build a net zero house. The AI then drafts the plan, codes a 3D printer, and continues to build the house with the help of robots it had programmed. Though general-purpose AI may be some years off it will be reached, adding importance to a conversation on pastoral virtues for AI. Artificial superintelligence is anticipated as an AI that will surpass human intelligence in every form, including scientific creativity, general wisdom, and social skills. ASI is currently only a theoretical possibility as we have yet to achieve general-purpose AI.
Whether it is machine learning or deep learning or reinforcement learning, AI depends on the training data—the input—it receives. This opens AI to algorithmic bias. When Apple entered the credit card business with their Apple Card, their algorithm discriminated against women, as a partner to one a Apple’s executives discovered. This is but one of many examples of AI not protecting equality or justice and is yet unable to navigate matters of intersectionality. When the values and virtues we seek in life have not been included in an AI’s algorithm, the danger related to a biased AI increases."
Joanne sounds like that soccer game was another step along the path to AI figuring out how to do things - playing against a team Not Using AI would not be the equality or justice that we expect to see on any sports playing field - huh maybe these trans men playing on women's sports teams is preparing us for the injustices to come...
