New frontiers Flashcards by Ava Cervas

Q

AI

A

How well did you know this?

1

Not at all

2

3

4

5

Perfectly

Q

brain-related data can be characterized by…5 V’S

A

Volume: number of data points
Variety: data may cross different types (structured/unstructured)
Velocity: pace of data generation; how fast can results show up?
Veracity: data quality and accuracy; EX: an apple watch vs laboratory sleep equipment
Value: potential to create benefits and insights

How well did you know this?

1

Not at all

2

3

4

5

Perfectly

Q

Artificial Intelligence

A

Artificial systems that appear to think like humans

How well did you know this?

1

Not at all

2

3

4

5

Perfectly

Q

Machine Learning

A

Systems that can learn from experience or data without direct human programming

How well did you know this?

1

Not at all

2

3

4

5

Perfectly

Q

Machine learning involves…

A

Training models on patterns in one set of data (training data) so that they can apply what they have learned to new data - classify, predict, decide, etc.
Can produce “black box” results – difficult to explain and interpret

How well did you know this?

1

Not at all

2

3

4

5

Perfectly

Q

2 types of machine learning

A

Supervised learning: Models are trained on known, labelled data; We know/are telling the AI what we’re training it to do, there is a right answer, we want to improve its efficiency in getting that right answer
Unsupervised learning: Models learn from unlabelled data; Not telling the data what its looking for

How well did you know this?

1

Not at all

2

3

4

5

Perfectly

Q

AI applications in Neuroscience (5)

A

Risk prediction
Clinical decision making
Neurotech
Brain modelling
Diagnosis & pronostication

How well did you know this?

1

Not at all

2

3

4

5

Perfectly

Q

Risk Prediction

AI applications in Neuroscience

A

Goal: predict Alzheimer’s Disease diagnosis using brain scans
Method: Train ML model, using labelled MRI data (heathy vs. Alzheimer’s Disease), to predict AD using neural activity. Identify most predictive brain regions

How well did you know this?

1

Not at all

2

3

4

5

Perfectly

Q

Clinical decision making:

AI applications in Neuroscience

A

Goal: surgically remove epileptogenic brain region to treat seizures using intracranial EEG (electrodes implanted on surface of brain)
Proposed ML model uses unlabelled features of the raw iEEG output to identify seizure origin
Trying to see if machine can split these electrodes into categories

How well did you know this?

1

Not at all

2

3

4

5

Perfectly

Q

Neurotech

AI applications in Neuroscience

A

Goal: Control limb prosthesis with neural activity
Train ML model on the mapping between neural activity and limb movement

How well did you know this?

1

Not at all

2

3

4

5

Perfectly

Q

Brain modelling

AI applications in Neuroscience

A

Goal: understand how rat brains represent space
Trained ML model to “navigate space” with training data that simulate real rodent behaviour + neural activity
Model developed representations resembling real rat entorhinal cortex “grid cells”

How well did you know this?

1

Not at all

2

3

4

5

Perfectly

Q

Diagnosis & prognostication

AI applications in Neuroscience

A

Problem: Need to triage acute neurological illnesses quickly (e.g., stroke, hemorrhage, hydrocephalus – “time is brain”)
Model type: supervised ML model trained on head CTs and radiology annotations
Result: Accelerated time to diagnosis in simulated clinical environment

How well did you know this?

1

Not at all

2

3

4

5

Perfectly

Q

Ethical issues around AI in Neuroscience: Accountability - Culpability

A

Culpability: Responsibility based on intention, knowledge, or control
AI introduces potential harms no one person could predict or prevent

How well did you know this?

1

Not at all

2

3

4

5

Perfectly

Q

Ethical issues around AI in Neuroscience: Accountability - Accountability

A

Moral Accountability: Duty to explain one’s reasons and actions to others; being answerable
AI processes may be unexplainable to their users
e.g., doctor can not explain AI-assisted diagnosis

How well did you know this?

1

Not at all

2

3

4

5

Perfectly

Q

Social Robots

A

How well did you know this?

1

Not at all

2

3

4

5

Perfectly

Q

Areas that social robots can be found (6)

A

Healthcare
Defense/security
Aerospace
Automotive
Electronics
Domestic

How well did you know this?

1

Not at all

2

3

4

5

Perfectly

Q

Defining social robot: 4 components

A

Sensors
Actuators
Autonomous
Social

How well did you know this?

1

Not at all

2

3

4

5

Perfectly

Q

Sensors

Defining social robot

A

identifies and responds to its
environment

How well did you know this?

1

Not at all

2

3

4

5

Perfectly

Q

Actuators

Defining social robot

A

physically moves; has a body

How well did you know this?

1

Not at all

2

3

4

5

Perfectly

Q

Social

Defining social robot

A

capable of social interaction; follows social rules

How well did you know this?

1

Not at all

2

3

4

5

Perfectly

Q

The Expectations Gap in robotics

A

There is a mismatch between expected capabilities and actual capabilities of social robots.
“Robots were pop-culture icons before they even existed”

How well did you know this?

1

Not at all

2

3

4

5

Perfectly

Q

Current robots look like (3)

A

Pet - EX: dog robot
Station - EX: for sick children, a robot that substitutes their place in class
Aide

Q

Real social robot abilities are fairly limited - Wizard-of-oz method

A

Studies often rely on Wizard-of-Oz method (a man behind the curtain)
Dynamic, adaptive movement is difficult and current robots are pretty bad at it
Conversation is difficult and robot speech is often pre- scripted (…until now?)
Emotion modeling is difficult and robots are bad at “reading the room”

Q

Potential applications for brain health (4)

A

Emotional support: companionship, touch
Monitoring: pain monitoring, fall monitoring, “smart” integrations
Health: physical therapy, distraction, dispensing medication
Connection: telepresence, reporting

Manufacturers claim many benefits, but...

...evidence is often fairly weak

Potential positive/negative outcomes for older adults with dementia - robots

* Agitation * Anxiety * Depression * Engagement * Interaction * Stress * Loneliness * Quality of life

Reasonable evidence for benefits to children in healthcare environments

* Anxiety reduction * Health info (diabetes) * Anger and depression (cancer) * Exercise (cerebral palsy)

Some evidence for benefits to children with Autism Spectrum Disorder

* Improved social functioning, e.g., * Increasing body awareness * Encouraging collaborative skills * Prolonging children’s attention span * Mediating social interaction * Learning appropriate physical interaction * But not emotional or motor outcomes

Why not give up on social robots?

Maybe we should! But... * Companies keep making and marketing them * The healthcare system faces intense demands * Some people really like them and might use them over other treatment modalities (especially kids) * Patient engagement: A path forward? * Can improve robots by patient-oriented training/research

Brain-Computer Interfaces

Components of Brain-Computer Interfaces

1. Record 2. Process 3. Control 4. Feedback

Record - levels of brain activity recording (2) ## Footnote Components of Brain-Computer Interfaces

* Semi-invasive * Invasive

Record - levels of brain activity recording: **Semi-invasive** ## Footnote Components of Brain-Computer Interfaces

electrocorticography (ECoG) electrodes places outside the dura mater (epidural) or under the dura mater (subdural)

Record - levels of brain activity recording: **Invasive** ## Footnote Components of Brain-Computer Interfaces

cortical implants are the only well-established invasive recording sensor

Record - levels of brain activity recording: **Utah Array** ## Footnote Components of Brain-Computer Interfaces

first and most studied, FDA approved implantable sensor

Record - levels of brain activity recording: **Neuropixels** ## Footnote Components of Brain-Computer Interfaces

* Multi-electrode array with hundreds of sensors along a single, thin probe * Can record from hundreds of neurons simultaneously * In 2022, tested in 9 patients under anaesthesia receiving brain surgery to validate technology

Record - levels of brain activity recording: **Stentrode** ## Footnote Components of Brain-Computer Interfaces

wire and electrode implant threaded into brain's blood vessels

Record - levels of brain activity recording: **Neural Lace** ## Footnote Components of Brain-Computer Interfaces

* Many flexible probes inserted via surgical robot (very fragile) * Signal is weaker, dispersed by bone, hair, skin * Typically require craniotomy * Surgically opening the skull

Process ## Footnote Components of Brain-Computer Interfaces

brain activity is processed using a computer to identify the user's desired action

Control ## Footnote Components of Brain-Computer Interfaces

a signal is sent to the application to carry out the desired command

Feedback ## Footnote Components of Brain-Computer Interfaces

* Feedback is provided to the user to indicate their action was successful * BCI: an AI interface with the brain that bypass ADD

Bidirectional BCI

1. Myoelectric prosthesis 2. Deep brain stimulation (DBS)

Myoelectric prosthesis ## Footnote Bidirectional BCI

* Electrodes implanted in motor and somatosensory cortices * Stimulating somatosensory cortex as though hand is being touched improves task performance

Deep brain stimulation (DBS) ## Footnote Bidirectional BCI

* Two intracranial electrodes implanted into basal ganglia, thalamus, or brain stem, with an implantable pulse generator (IPG) * Stimulates based on a pre-determined control policy

Deep brain stimulation (DBS) - Applications ## Footnote Bidirectional BCI

Applications: movement disorders such as Parkinson’s Disease

Deep brain stimulation (DBS) - Potential Applications ## Footnote Bidirectional BCI

Alzheimer’s diasease, OCD, Tourette syndrome, major depression disorders, addiction, anorexia...

Stimulating without recording

Visual implants (retina or occipital lobe)

Corporate accountability; end of use ## Footnote Ethics

* Second Sight (company) collapsed without any notice to patients * What happens to a device after a research study? * Who is responsible to pay for removal or maintenance?

User Safety ## Footnote Ethics

* Complications from implantation, potential scarring * Unknown interactions with plasticity of developing brain * Unknown effect of removal

Autonomy ## Footnote Ethics

May also produce incorrect or unwanted actions, undermining autonomy (e.g., wrong movements, reveal inner thoughts)

Cyborgization & personhood ## Footnote Ethics

* Users do not feel that they “melt with technology” or “become part of a hybrid,” maybe because these statements imply losing their essence or identity * They seem to prefer metaphors like “controlling a tool”

Judging agency ## Footnote Ethics

* Who is responsible for BCI outputs? * Sometimes users attribute errors to themselves * At other times, users attribute errors to the tool

What is “normal”?

* BCIs have the potential to bring users closer to “normal” health or behaviour, in alignment with a deficit model of disability * If some individuals choose to use a BCI, what impact does that have on other individuals who share their condition? * E.g., for the Deaf community, is a cochlear implant to be viewed as an enhancement or a treatment?