1/37
Looks like no tags are added yet.
In the Washington Post article “Is this AI?” – There are different interpretations between experts and laypeople about what counts as artificial intelligence. What does this mean about the definition of artificial intelligence?
This means that the definition of artificial intelligence (AI) is socially constructed and subjective. Laypeople tend to determine what AI is based on how creepy it is such as conversational chatbots. However, they tend to overlook normalized tools such as Google translate or social media algorithms. Experts tend to define AI more logically by basing it on machine learning or pattern recognition.
Is there a technical or scientifically accepted definition of the term “Artificial Intelligence”? Why or why not
No there is not a technical or scientifically accepted definition of the term “Artificial Intelligence” because it was developed as a marketing term. Humans have struggled with determining what human intelligence is so determining what artificial intelligence has been more complicated, especially as it continues to evolve.
In what decade does the first notable computer system that is described as “artificial intelligence” appear?
1950s
What did the term “computer” mean in 1891?
The term computer was an occupation, usually taken by a woman, who computed long lists of numbers, including calculations.
What does “the hard things are easy, easy things are hard” mean in the context of the history of artificial intelligence?
“Hard things are easy” summarizes the advanced cognitive tasks that humans consider difficult such as chess or statistics that are easy for AI. “Easy things are hard” is the reverse of the previous, things that humans consider easy such as facial recognition as considered AI systems to replicate.
What is the difference between hallucination, lying and bullshit according to Bergstrom ?
Bergstrom writes about how ChatGPT isn’t ‘hallucinating’, it's bullshitting. What's the difference? Hallucination describes a perceptual mistake such as seeing something that is not there. Lying requires the model to know the truth and try to deceive the user. Bullshitting attempts to persuade without regard to the truth. Since AI has no concept of truth, it cannot be lying and hallucinating would suggest that AI has a perception of reality, it cannot be hallucinating. Rather, AI models are bullshitting.
Pick a task: image generation, reading comprehension, handwriting recognition, speech recognition and describe how it has developed in the last 20 years.
Handwriting recognition is an interesting benchmark task because it is one of those “easy things are hard” tasks. When AI was first tested on handwriting recognition, AI found it easy to recognize neat and printed text rather than messy and cursive. As it practiced, it was able to start reading messier handwriting. Today, AI can handle the messy and cursive handwriting it once struggled with, surpassing human accuracy and reaching near-perfect benchmark scores.
What is one danger of anthropomorphizing AI, according to Mitchell (2024)?
A danger of anthropomorphizing AI is creating a false sense of trust as it makes the model appear to have a “mind.” This also instills the sense that the model can show empathy. This can make people think they are able to discuss emotional topics such as suicidal thoughts. News within the last few years prove that entrusting AI with this kind of information leads to more harmful effects such as encouraging harmful behaviors.
In the brief history of AI article from Our World in Data, it describes current AI systems that are used in air travel, the legal system, translation, recommender systems, self-driving cars and warfare. Pick one of these examples and briefly describe the benefits and disadvantages of using artificial intelligence for that purpose. What would the experts in the Ghost in the Machine movie think about the usage you picked?
Self-driving cars are interesting to me. On one hand, they have the potential to reduce traffic accidents that would typically be caused by human mistakes such as distracted driving, fatigue, or drunk driving. It can also provide opportunities for those that are unable to drive/walk around to get around by themselves, creating a sense of independence. But when we have something like AI in charge of human lives, there is the worry that what if the AI glitches or chooses to save the people in the car over the pedestrians crossing the street. If these things were to happen, who would be held responsible? Experts in the Ghost in the Machine would view the self-driving cars with deep skepticism. There is no way to fully determine if a self-driving car is ready to be used without deploying them into the real world, turning pedestrians’ safety in test subjects. Experts would argue that corporations would use the algorithmic “black box” to avoid legal accountability and shift blame away from themselves if a crash were to happen.
What is a Luddite? Describe the historical meaning, and how it is used currently.
During the industrial revolution, Luddites was used to describe textile artisans and weavers who destroyed industrial machinery. This was a public act to show how they disagreed with how factories were using this technology to lower wages, worsen working conditions, and produce lower quality goods. Today, Luddite is used to describe anyone that is “anti-technology” whether it be by ignorance or fear of modern technology.
Do you think the Luddite’s would have supported or adopted the metaphor of the term’s “centaur” and “reverse centaur” when it comes to human use of technology? Which would they have preferred?
The metaphor of “centaur” describes a model where the human is still in control (the top half) and the AI is an enhancement tool (the bottom half). In a “reverse centaur”, the human is the tool (the bottom half) and the AI is in control (the top half). I think that Luddite would disagree and fear the “Reverse centaur” model. I would say that they would be more in support of the centaur model considering their backlash was from the control of technology.
The What AI is doing to young brains article mentions concerns with young people using “social AI.” Why are psychological scientists concerned about children and teens using AI in this way?
With the youth using “social AI” it is replacing real-world human relationships which is necessary for developing social skills such as conflict resolution and emotional regulation. Since AI is super prone to valdiating and are constantly available on the internet, they can cause a rise to feelings of loneliness. This can lead a cycle of confiding in AI models and feeling loneliness due to young people offloading emotional support.
Automata
historical mechanical machines built to perform repeating tasks; the earliest ancestors of modern automated AI systems
Computers
humans, typically women, who were employed to manually perform long mathematical calculations
Robots
physical machines capable of carrying out complex series of actions automatically, often guided by computer programing
Algorithm
a step-by-step set of rules, instructions or mathematical procedures used to solve a problem or perform a task
Neural network
An AI model structured in connected computational layers to recognize patterns, loosely mimicking how neurons fire in biological brains
Artificial neural network
subfield of AI that has networks of interconnected nodes arranged in layers, inspired by the human brain
Machine learning
subfield of AI focused on training systems to discover patterns in data and improve automatically
Training sets
initial datasets fed into AI models to teach it
Training stage
the initial heavy computing phase where an AI model analyzes training data and learns patterns
Inference stage
operational phase where a finished AI model applies what it’s learned to process new input
Hallucination
false sense that lead to false beliefs
Confabulation
unknowingly fill in a gap in memory
Bullshit
Language intended to appear persuasive without regard to its actual truth or logical consistency
Moore’s Law
Computational power doubles every two years
Goodhart’s Law
When a measure becomes a target, it ceases to be a good measure
Reverse Centaur
AI (top half) is in charge of humans (bottom half)
Stochastic parrot
Haphazardly stitching together sequences of linguistic forms to mimic
Metacognition
“thinking about thinking: which is something AI cannot do
Productive struggle in learning
uncomfortable effort humans exert to learn difficult concepts
Cognitive labor
intellectual work requiring active human thought, analysis, and reasoning
Cognitive offloading
offloading thinking onto someone else or thing
Deskilling
weakening of human skills due to overreliance on AI
Upskilling
learning higher-level skills to effectively collaborate with and manage new AI tools in the workplace
Luddite
19th century workers who broke machines due to hate for lower wages and poor work conditions that technology was creating.
What does it mean that the lines far to the left (the blue “handwriting recognition” line) are slower to rise than the lines farther to the right (like the green “reading comprehension” line)?
It shows how it took AI a long time to be as good as a human at handwriting recognition. This can be because during that time traditional machine learning and early computer vision made it harder for AI to learn and improve. This is especially apparent when we look at reading comprehension which did not become a benchmark task till later but AI rapidly became as good as a human in it. This shows that modern computer and machine learning has had a dramatic impact on AI learning as it has made AI learn skills even quicker than before.
What does it mean when a colored line passes the gray line at zero?
The gray line at zero represents human baseline performance. When a colored line passes that line, it represents that the AI system has surpassed average human performance in that benchmark test.