AI Inquiry Atlas

An unofficial, independent guide to the public records of the Joint Select Committee on Artificial Intelligence. Not connected with the Parliament of Australia or the committee. Its makers also made a submission (no. 481; see About). Summaries and scores are written by AI and can be wrong: check the original before relying on them.

Submission 473

Mr Scott Masters

Submitter group: Individuals.

Position labels are not shown for individual submitters.

An AI model identified the submitter group and chose the extracts. Counts do not measure importance or merit.

Open in the interactive guide

Extracts from the submission, in the submitter’s own words.

Extracts

Quotes are the submitter’s own words. Text taken from documents, including scanned pages read by text-recognition software, can differ from the PDF. Check the original. Listed in the same order as on the submission’s page in the interactive guide.

  1. The lesson that was hammered into me, again and again, doing that work was that the model does not do what you want by default. I would give the model a metric to optimise (profit for example) and it would maximise that metric in an unhelpful way.

  2. Artificial Superintelligence (ASI) is the point where AI can outperform a top expert human on any task. As I have watched the improving abilities of each new model that is released, this milestone feels plausibly two years away.

  3. Meanwhile, the problem I encountered a decade ago, that the model does not do what you want by default, also called misalignment, remains. We are able to gesture at what we consider to be good and bad behaviour. We are not able to specify an AI’s goals in detail.

  4. Respected experts such as Yoshua Bengio10, Geoffrey Hinton11 and Eliezer Yudkowsky12 have warned repeatedly that unregulated development of AI capabilities risks human extinction.

  5. Australia should focus on adopting AI more broadly, rather than competing to create frontier models itself.

Extracts from recommendations

Extracts chosen by an AI model, in the submitter’s own words. Listed in the same order as on its page in the interactive guide.

  1. I hope that the committee will treat this as a warning shot. I cannot stress enough that in the Hugging Face incident we got extremely lucky that the goals of the misaligned AI remained narrow and unambitious7.

  2. An international treaty is needed to pause training new AI models until we can be confident of alignment. Fortunately, Australia has tremendous soft power in this space. Australia’s under-16s social media ban has been copied all over the world and for better or worse the details of that policy are being used globally as a template to draw from. Australia should use its soft power to advocate for a world- spanning treaty to pause AI.

  3. Holding AI companies legally liable for misaligned behaviour gives them a monetary incentive to solve the problem. If an AI model autonomously hacks a third party the model’s creators should be held responsible for the damage. If an AI model can be coaxed into giving terrorists instructions on producing a lethal virus its creators should be regarded as accessories to that crime.

  4. Australia should focus on adopting AI more broadly, rather than competing to create frontier models itself.

  5. The members of this committee need to raise these issues with the parliament, the government, the media and the public.

Read the original

Open in the interactive guide