Social Impact

EU risk levels

Unacceptable, high and limited risk

EU unacceptable

Systems using subliminal techniques, purposeful

manipulation/deception

Systems exploiting vulnerabilities of individuals or groups

Systems identifying individuals through biometrics (e.g. faces)

- With exceptions for law enforcement

Systems for ‘social scoring’

Systems for inferring emotions, in workplaces and educational

settings.

EU high risk 

safety (medical)or social issues (bias, employment, justice)

AI Act provisions about Gen AI 

Publish summaries of the content 

Copy with EU copyright

Write technical documentation for ‘downstream users’

The Bletchley Park AI Safety Summit

Cooperation with over 28 countries NZ joined in 2024

This event focused on large Gen AI models

recognizes risk of bio weapons, toxins, alteration of genes and cybersecurity

-transparency, evolution metric, safety testing

Recognises the risks of ‘frontier’ AI models

Commits signatories ‘to work together in an inclusive manner to

ensure human-centric, trustworthy and responsible AI that is safe,

and supports the good of all through existing international fora and

other relevant initiatives, to promote cooperation to address the

broad range of risks posed by AI’

UK legisatation

Still a year away

security institute, hand over model to be tested, copyright 

US Policy

Joe biden had an order to disclose risks to national security results

develop standards for red teaming and evaluation

Donald trump removed that replaced it with how to remove ai barriers

China

No comprehensive policy’s, but big reports & committed

detailed legal requirements on AI content labelling and

watermarking are already in place

-thinking about cybersecurity, biosecurity and open source

Job impact

number of young employees declines

productivity

optimus humanoid in manufacturing, or self driving delivery vehicles

Managers - gig work

Issues in Ai in workplace gig work

monitoring

Recruiters use of Ai

In CV and Cover letter screening

Ai interviews

Ai taking notes

employee/candidate use of Ai

Using Ai for Cover letters and CV

help in practice interviews

and ai in workforce

1 Scenario of AI in NZ

The main effect of AI is to improve the productivity of workers.

No mass unemployment; instead, NZ workers are more efficient

Positive uses of Ai in recruitment

Positive uses of AI in recruitment

We can also work to build fairness into recruitment tools.

1. We can delete features that aren’t relevant from training sets.

Gender, ethnicity are often irrelevant. . .

If assessors classify applicants based on redacted application

materials, it’s harder for them to be biased.

2. We could include audit functions, that show percentages of hired

people from different demographic groups.

This way, biases will at least be visible. (Within the company and

beyond.)

Working alongside AI

Say you’re a worker making decisions. . .

- Perhaps you’re a doctor, looking at X-rays and detecting

fractures. . .

Say an AI system is working alongside you, to help you.

- Say it’s pretty reliable. . . perhaps 95%. . .

How do you stay in the loop??

- It’s hard to stay in control, if the system works well!

Also - who’s responsible, if the thing you are jointly doing goes wrong?

Ali Knott AIML131 Week 11 Lecture 2 17 / 24

Scenario 2

Replacing onshore

NZ workers are displaced by AI, into lower-value work.

The AI systems doing the displacing are NZ-owned.

(So NZ can recover some of their profits through taxation.

Replacing offshore’

NZ workers are displaced by AI, into lower-value work.

The AI systems doing the displacing are owned offshore.

-International tech tax

International Tech Tax

There is an international tech tax being organised by the OECD, for

large multinational companies. The US walked out in June.

‘Pillar 1’ sets things up so each country taxes LMCs according to

the revenue they make in that country.

‘Pillar 2’ establishes a ‘global minimum tax’ of 15%, so companies

can’t run to countries with low corporation tax.