IB Computer Science Topic A4.4.2: Ethical Aspects of the Increasing Integration of Computer Technologies Into Daily Life Notes

Ethical Guidelines Age Faster Than the Technology They Govern

1. A guideline is written against the capabilities that exist when it is written, and capabilities change faster than guidelines do.
2. Ethical norms are not static, so a practice that was acceptable when the rule was drafted can be unacceptable a decade later.
3. Harms often appear only at scale, long after a small pilot was approved as harmless.
4. This article judges four emerging technologies against four questions: the effect on society, on individual rights, on privacy and on equity.

Why Reassessment Matters: Trust, Early Warning and the Innovation Trade-Off

1. Evolving values
   1. Public expectations about surveillance, data collection and automated decisions have shifted, and guidelines have to move with them.
   2. Guidelines that fall behind those expectations produce public distrust of the technology and of the organisation using it.
2. Unintended consequences
   1. A capability released for one purpose gets used for others, and those later uses were never assessed.
   2. Scheduled reassessment is how a risk gets caught while it is still cheap to fix.
3. The innovation trade-off
   1. Rules that are too tight slow useful work and push development towards places with weaker oversight.
   2. Rules that are too loose leave the cost of the harm with the public rather than with the developer.
   3. Risk-tiered regulation is one answer, because the requirement scales with the harm the system could do.

Quantum Computing: Today's Encrypted Data Is Tomorrow's Plain Text

1. Privacy and individual rights
   1. A large quantum computer would break the public key cryptography that protects banking, medical records and messaging.
   2. This creates a harvest now, decrypt later problem, because encrypted traffic captured today can be stored until hardware exists that can read it.
   3. Post-quantum cryptography replaces the vulnerable algorithms with ones no known quantum algorithm breaks, but every system has to be migrated first.
2. Society
   1. The same machines could shorten drug discovery, materials design and logistics planning by simulating systems classical computers cannot handle.
   2. The risk and the benefit come from one capability, which is why banning the technology is not a serious proposal.
3. Equity
   1. The cost of building and cooling a quantum computer concentrates access in a few wealthy states and corporations.
   2. Whoever gets a working machine first can read what everyone else still believes is private, so a capability gap becomes a power gap.

Augmented Reality: Recording the World Means Recording Everyone In It

1. Privacy
   1. An AR device needs cameras and microphones running continuously to know what to overlay, so it captures whatever is in front of the wearer.
   2. Bystanders never agreed to being recorded, and usually cannot tell that they are.
   3. Pairing the camera with face recognition would let a wearer identify a stranger in the street and call up information about them.
2. Individual rights and safety
   1. Moving through a public space without being identified is treated as part of the right to privacy, and an always-on overlay erodes it.
   2. A display in the wearer's line of sight competes for their attention, so distraction causes accidents in the real environment.
3. Society and equity
   1. Whoever supplies the overlay decides what it shows, so commercial interests shape what a user notices about a place.
   2. Areas that are not mapped in detail receive a poorer overlay, which repeats existing inequality in digital services.

Virtual Reality: Immersion Collects Data a Browser Never Could

1. Privacy
   1. A headset tracks eye movement, gaze duration, head motion, posture and hand tremor so that it can render the scene correctly.
   2. That stream is biometric and can indicate attention, emotional response and some health conditions, none of which a click log reveals.
   3. Movement patterns are distinctive enough to identify a user, so removing the account name does not anonymise the recording.
2. Individual rights and well-being
   1. Harassment in social VR is experienced as physical, because the system is built to produce a sense of presence.
   2. There is no text transcript of a VR interaction, so reporting and moderation are much harder than on a text platform.
   3. Long sessions can cause disorientation on returning to the real setting, and the long-term effects on children have not been established.
3. Society and equity
   1. VR gives access to training, surgical rehearsal and remote collaboration that would otherwise be too expensive or too dangerous.
   2. Headset cost, motion sickness and some disabilities exclude part of the population from any service delivered this way.

Pervasive AI: Decisions Made About You, In Places You Cannot See

1. Individual rights
   1. Automated decisions in hiring, credit, welfare and policing affect people who often have no way of knowing a model was involved.
   2. Consent stops being meaningful when the only alternative is to go without a service everyone else uses.
   3. A right of appeal needs a human who can review the decision and overturn it.
2. Privacy
   1. Ambient devices such as smart speakers and building sensors collect continuously rather than when a user chooses to interact.
   2. Combining ordinary signals reveals facts nobody disclosed, the same inference problem that recommender systems create.
3. Society
   1. Widespread automation changes and removes jobs, which A4.4.1 covers as the societal impact of machine learning.
   2. Depending on a small number of shared AI services means many organisations fail at the same moment when one of them fails.
   3. A biased model deployed everywhere repeats one bad decision at national scale instead of in one office.
4. Equity
   1. People without a device, a connection or the confidence to use one are excluded from services that assume all three.
   2. Automated decisions fall hardest on those least able to challenge them, so an existing gap widens rather than a new one opening.

Reducing Harm From Emerging Technologies

1. Assess before deployment, so a risk is found at pilot scale rather than at national scale.
2. Regulate by risk tier, so the requirements match the harm a system could cause.
3. Require transparency, so a person can find out that a model was used and on what basis.
4. Design for privacy by default, so collection is limited and sensitive features are off until a user turns them on.
5. Allow independent audit, because an internal review has an interest in its own answer.
6. Fund digital literacy, so the public can judge a specific claim about a technology instead of accepting or rejecting all of it.
7. Reassess on a schedule, because every measure above was written against the capabilities that existed at the time.

Applying the Four Lenses to a Technology You Have Not Studied

1. An exam scenario can name a technology that is not in the syllabus, and the four lenses still produce a full answer.
2. The ethical issues raised by machine learning specifically are set out in A4.4.1.

Where Answers Lose Marks on This Statement

1. Describing how a technology works instead of what it implies, which answers a question that was not asked.
2. Delivering a verdict that a technology is good or bad, when Discuss requires both sides to be weighed before any conclusion is reached.
3. Leaving out equity, which is the lens most answers skip and the easiest one to make specific.