L10 - Big Data, Artificial Intelligence & Brain Health + Neurorights

What are the 5 V’s of Big Data? + What do they do? + Give some examples for each

Big data is considered large and continually generated digital datasets & these are the characteristics of complex data

Volume: number of data points / amount of data generated

• Costco is a large chain company with multiple stores across the province and generates millions of transactions per day => large dataset of purchases and customer behaviour

Velocity: pace of data generation

• MRI has low velocity VS steps counted on IPhone has high velocity

Variety: different types of data (structured/unstructured)

• social media data includes videos, photos, comments

Veracity: data quality and accuracy

• bots and fake accounts on social media

• heart rate measured by Apple watch

Value: potential to create benefits and insights

• survey on customer behaviour used to improve future products

Define artificial intelligence VS machine learning + how does AGI differ from AI

AI = general umbrella term that encompasses machines and computers that have ability to mimic human intelligence (such as analyze data, categorize, recognize)

Machine learning = subset of AI / application of AI that allows systems to learn from experiences and data without human programming and apply what they learn to new data

AGI (artificial general intelligence) = hypothetical agent able to solve problems and make decisions on its own

Define supervised VS unsupervised learning (in machine learning) + give an example for each

Supervised learning = models that are trained on labelled, known data TO predict outcomes => need large volume of data

• medical records of patients who do or do not develop dementia

Unsupervised learning = models that learn from unlabelled data TO identify patterns => need huge processing power

• all reddit posts containing the word “dementia”

What are some applications of AI in neuroscience? (5)

Risk prediction => predict disease diagnosis using brain scans

• NSCI303 : trained machine learning to identify MRI data of healthy brain VS ones that have Alzheimer’s Disease

• LABELLED

Clinical decision making => identify disease region using intracranial EEG and surgically remove it

• trained machine learning model uses features of intracranial EEG output to identify origin of seizure

• UNLABELLED

Neurotech => control limb movement with neural activity

• trained machine learning model to map neural activity and limb movement

Brain modelling => understanding how brains represent space

• PSYC370 : trained machine learning model to “navigate space” to act like entorhinal cortex “grid cells”

Diagnosis & prognostication => sort data and determine what disorder someone has based on behaviours + what would happen next

• trained machine learning model to triage neurological illnesses on head CTs and radiology annotations

• LABELLED

What are some ethical issues of AI in neuroscience?

• Beneficence & Autonomy

  • culpability : responsibility based on intention, knowledge or control => if neurosurgery robot makes an error, is it the surgeon's or the robot’s fault

• moral accountability : duty to explain one’s reasons and actions => AI has found that someone will have a said disorder in the future but doctors (humans) may not be able to explain the diagnosis

• future access to present data : personal preference and emotional states could be at risk => anonymous data that is no longer anonymous because future AI models could potentially reveal someone’s identity by looking at their past memories

• Justice

  • cultural under-representation in AI model research

  • downplaying data of women & ethnic minorities

• What are LLMs?

• What type of data is used to train LLMs?

• Are LLMs designed to match their training data or to be true?

• LLMs are Large Language Models, generating new text

• Type of data used to train LLMs is large datasets from publicly available information online which can include encyclopedia, wikipedia, personal blogs, books, scientific papers

• LLMs outputs are designed to match the training data BUT not necessarily to be true

What are some ethical and/or environmental issues arising from LLMs?

• energy consumption

• carbon emissions

• water consumption for cooling

• data privacy (use of information with no consent)

• lack of transparency (information from “black box” because it’s hard for humans to understand how AI arrived at certain conclusions)

• malicious usage (generate deceptive content)

What is NeuroRights?

New sets of human rights set to protect an individual’s brain data from misuses of neurotechnologies and artificial intelligence

What are the 5 NeuroRights?

• Right to personal identity

• Right to free will (autonomy)

• Right to mental privacy

• Right to equal access to mental augmentation

• Right to protection from algorithmic bias

• How does Neurorights Initiative define personal identity?

• What principle guides the Neurorights Initiative in establishing guidelines to regulate neurotech?

• Individuals should have ultimate control over their own decision making without manipulation from external neurotechnologies

  • ☹ “ultimate control” will be compromised when policies are put into place

  • 🙂 shared end goal of achieving free will

• Based on the principle of justice & guarantee equality of access to all citizens