top of page



Preparing for AI

Your weekly CogX newsletter on AI Safety and Ethics

Is the EU making a mistake?

Leading executives urged the EU to rethink the AI Act. Corporates worry the legislation will inhibit AI development and make it harder for European companies to compete overseas. Is the AI act too stringent? What’s the right balance between safety and growth?

Meanwhile new research found adding texture to images can trick AI detection software - detection rates fell from 99% to 3.3%.

Detection software is critical to combat misinformation and more robust solutions are urgently needed.

Read more about this and more in the CogX Must Reads - including the Pope’s AI ethics handbook, Juergen Schmidhuber’s AI optimism, and new research from Anthropic.

Top Stories
AI image detectors can be tricked

Our frontline defence against AI generated misinformation is AI detection software.

So it’s especially worrying that a NYT investigation found that adding grain - i.e. texture - to an AI generated photo can trick the system. We need more robust solutions.

Pope joins AI ethics debate

The Vatican has released an AI ethics handbook to help address critical emerging issues on AI.

The handbook emphasises embedding moral ideals and ethical standards into technology from inception and for AI companies to serve the common good of people.

Politics and Regulation
Corporates pushback against hardline EU rules

More than 150 leading executives including from Airbus, Renault and Heineken, signed an open letter urging the EU to rethink their stringent approach to AI regulation.

Corporates worry the EU will become uncompetitive and regulations will shackle AI development and growth. They want the laws to focus on broad risks rather than “rigid compliance.”

Japan takes softer regulatory approach

Japan is leaning towards a US-style regulatory approach rather than the EU’s.

It sees AI as an opportunity to boost economic growth, wants to be a leader in advanced chips and worries that burdensome regulation will make it less competitive.

Latest Research
Will AI actually lead to more jobs?

New research from the ECB, Oxford and others analysed job market data in Europe from 2011-19.

They found average employment shares in Europe actually increased in occupations more exposed to AI. White collar jobs saw a 5% boost in employment. The research shows AI can increase demand for skilled workers too.

Do LLMs represent diverse perspectives?

Research from Anthropic found default LLM responses to reflect opinions from US and European populations.

When prompted to consider another country’s perspective, responses shifted but often reflected harmful cultural stereotypes. More work is needed on evaluation frameworks to detect biases.

AI Thinkers
Juergen Schmidhuber’s AI optimism

CogX Festival 2023 speaker and Father of Modern AI, Juergen Schmidhuber, warned that those speaking out on AI dangers are simply seeking publicity.

We should be hugely optimistic about AI - 95% of all AI research and development is about making people happier and advancing human life.

Focus on today’s risks not theoretical risks

A group of European professors argue that we should stop speculating about theoretical existential threats and focus on tangible risks today.

Existential risk angst has distracted us from misinformation, exploitation of workers annotating datasets, the carbon footprint of data centres and algorithmic biases.

Use Cases
Latest hack increase fear of deepfake scams

Progress Corp had its file transfer system compromised this month leading to sensitive information from PwC, Shell and others being obtained.

Normally the hackers would seek a ransom. But experts say hackers are using sensitive information combined with deepfake software to cash in with lucrative ID theft scams. The future of hacking has changed.

The hidden side to building AI models

Behind every AI model are huge numbers of people intensively labelling data to train it.

The tedious work of annotating raw information often falls to young Africans, particularly in Kenya, who can be paid as little as $1 an hour. Allegations of exploitation are rife.

CogX Must Reads of the week

In case you missed it

Check out the Munk debate on AI risks with leading AI researchers

We'd love to hear your thoughts on this week’s Issue and what you’d like to see more of.

bottom of page