1 of 4

CAN WE ‘TRUST’ AI?

8th March 2023

Mark Irura

​

Contact: irura.gachara@gmail.com

2 of 4

TRUST…

2

… firm belief in the reliability, truth, or ability of someone or something.�

​

There has always been debate if artificial general intelligence is possible at all;

chatGPT stands as a touchstone for what AGI could resemble.

The quick, human-like responses of generative AI may be part of the cause of mistrust

The capabilities in diverse fields have evoked hope and caution.

3 of 4

SHOULD SUCH A TOOL DO THIS?

3

OpenAI’s moderation strategy seems to have less to do with building rigid controls on top of the AI system, and more to do with actually training the AI to be more ethical.

Or are these the wrong questions?

4 of 4

BUT THIS IS NOT ENOUGH

  • Can AI artefacts be built to be ethical?
  • What does that mean?
  • What is needed?

​

  1. Value alignment
  2. Identify relevant human values
  3. Are there universal human values?
  4. Who gets a say? Why these?

  • How to behave?
  • Ethical theories: How to behave according to these values?
  • How to prioritize those values?
  • How to pass them on?

4

Source: Dignum, V (2018)