The Challenges of AI

At the conclusion of his introductory account of AI Toby Walsh (‘The Shortest History of AI’) poses a number of contemporary challenges facing us. These make immediate sense to me, carry the potential of a real threat to our social wellbeing, and warrant reproducing here. So, thanks to Toby for what follows. He lists the following:

1. The bias challenge – AI might create or perpetuate biases that society find unacceptable.
2. The privacy challenge – AI might identify individuals and provide personal information about them beyond what the public expects.
3. The misrepresentation challenge – AI might generate content that misrepresents someone’s behaviour, opinions or character, as well as about important issues like the climate or political parties.
4. The access to data challenge – the most powerful AI systems require access to very large data sets, which are held by few organisations.
5. The access to computation challenge – the most potent AI requires access to significant amounts of computing power, available only to a few organisations.
6. The black box challenge – AI models often cannot explain why they produce a particular result.
7. The open-source challenge – making AI code openly available may promote transparency and innovation but may enable bad actors to commit harm. On the other hand, making code proprietary may prevent this but may concentrate market power.
8. The intellectual property and copyright challenge – large AI models are often trained on copyrighted content without the consent of owners or their compensation. It remains uncertain then who owns the outputs.
9. The liability challenge – if AI models are used by third parties and cause harm, it must be established who bears any liability.
10. The employment challenge – AI will disrupt the jobs that people do and that are available to be done. This disruption must be managed carefully.
11. The international coordination challenge – AI is being developed around the world, so any governance framework to regulate it require international cooperation;
12. The existential challenge – AI might pose a threat to the human species itself.
13. The environmental challenge – large AI models producing CO2 and consuming water.
14. The democratic challenge – AI being used to influence voters and upset elections around the world.   

These strike me as comprehensive and vital in the present, let alone the future.