What you need to know

• Industry pundits see a big weakness in a new US Presidential Executive Order on AI: It responds to current threats, not new risks posed by Generative AI, including one AI pioneer, Louis Rosenberg, calls the most powerful form of mass persuasion ever.
• Overall, the Executive Order is broadly ambitious, and likely to garner bipartisan as well as industry support.
• Rosenberg, who has advised US and Australian law makers and regulators, has positive things to say about aspects of the Order, including the focus on consumer safety and corporate transparency.
• His views are shared by local experts including Norton Rose Fulbright technology lawyer and digital transformation co-lead, Nick Abrahams, who calls out the focus on addressing deep fakes, and new safeguards on areas such as health and criminal justice.

There is a real danger the new US Presidential Executive Order on AI, released last week, will focus too much on familiar dangers while ignoring potentially vast new harms, according to one of the world’s top augmented and virtual reality pioneers and entrepreneurs.

Louise Rosenberg, CEO of Unanimous AI, holds more than 300 patents in augmented reality, virtual reality and AI and is a leading light in the responsible AI movement. He believes US President, Joe Biden's new Executive Order on AI fails to account for new dangers such as Generative AI enabling an entirely new form of manipulation - interactive conversational influences.

Even so, Rosenberg has several positive things to say about the order. He and other specialists Mi3 spoke with said responses to the Executive Order on AI are likely to be bipartisan, at least initially.

The Executive Order last week outlined a number of areas where the US intends to apply extra scrutiny and regulation. These include:

• Using existing legislative tools such as the Defense Production Act to require digital giants to share their AI safety test results and other critical information with the US Government.
• Authorising the National Institute of Standards and Technology to set standards governing the release of AI.
• Applying those standards to critical infrastructure sectors and suppliers, with oversight from the Department of Homeland Security.
• Watermarking any government content created by AI.
• New protections around privacy and workers' rights, although some of these will require additional legislation.

Rosenberg has had a number of conversations with Senate staffers this year and gave a presentation to the Federal Trade Commission (FTC) on the topic of AI risks. He has also had multiple conversations with AI policymakers in the Australian Government, including the eSafety Commissioner.

While many parts of the Executive Order are aspirational, Rosenberg also saw it ushering in new legal requirements.

“The one part of the Executive Order that is an actual requirement is the need for companies developing large-scale AI systems such as Large Language Models to disclose their efforts if they are building systems larger than a threshold size, to perform red team testing to validate safety and to disclose the results of that testing," he told Mi3.

“This is an important step for transparency. To appreciate the significance of corporate transparency, we can look at historical situations in which tobacco companies, oil companies, and even social media companies have internally known their products or technologies cause harm and kept that knowledge secret. Hopefully, this requirement for AI companies avoids those scenarios.”

Rosenberg is less convinced, however, the Order will avoid the missteps of the past.

“What's missing? Technology regulators often fail to protect the public by focusing on the familiar dangers rather than the new dangers. This happened with social media where regulators looked at influence campaigns on social media [i.e. advertising] and considered the issue to be similar to traditional print, television and radio advertising. What they failed to appreciate is social media enables targeted influence, which was a totally new danger and contributed to the polarisation of society.”

Rosenberg also saw the Executive Order making a similar mistake about artificial intelligence, focusing on the ability of Generative AI systems to efficiently create traditional misinformation and disinformation.

“Yes, it's a danger that AI tools have made it far easier to create fake photos, fake audio clips, fake videos, and fake documents than ever before. But this is not a new danger. The new danger is Generative AI is enabling an entirely new form of influence - interactive conversational influence - in which users will engage chatbots, voice bots, and soon photorealistic video bots, that will be able to convey misinformation and disinformation through interactive conversations that target users individually, adapting to their personal interests, values, political leanings, education level, personalities, and speaking style,” he said.

"This will be the most powerful form of mass persuasion ever deployed. Sometimes referred to as the AI Manipulation Problem, it has not been addressed by the Executive Order.”

Perhaps surprisingly, Rosenberg believed Europe and even Australia are further ahead when it comes to technology policy than the US, especially when it comes to issues such as privacy.

“I generally look at the efforts in Europe and also Australia as being a little more forward-thinking than the US, especially related to privacy and a willingness to push back on corporations that are bad actors. This has certainly been the case with traditional social media platforms.”

Still, when it comes to AI, Rosenberg described Biden’s Executive Order as a bold step with the potential to put the US in a strong leadership position if it follows through.

“This would be good for the US, as it's a national strength to protect the public from AI. Those who say AI regulation puts countries at a disadvantage don't understand that ensuring safe AI for citizens is a national advantage, period,” he added.

Shared AI concerns

Nick Abrahams, a Sydney-based technology lawyer, global co-leader of digital transformation for Norton Rose Fulbright and one of the co-founders of LawPath, spied common areas of concern between the US and European approaches.

“Both the US order and Europe’s AI Act emphasise the safety and ethical implications of AI. Privacy protections remain a core concern across different regulations," he said. “The potential for AI-induced unemployment is a shared concern worldwide.”

There are a number of positive aspects to the US executive order, according to Abrahams, starting with the fact it seems to have fairly comprehensive coverage. The order addresses a wide range of areas including safety, privacy and civil rights as well as the issue of deep fakes.

“The directive to watermark AI-generated content can help mitigate the risk from deep fakes – which is a pressing issue," he said.

Abrahams also noted the Executive Order seeks to combat algorithmic bias and discrimination in crucial areas like healthcare, housing, and the criminal justice system.

As to how the new framework outlined in the Executive Order will be received, both Rosenberg and Abrahams believe the tech sector will generally be welcoming – if only for a larger measure of certainty.

Per Abrahams: “Technology companies will want to ensure excessive regulation doesn’t limit innovation but the proposed regulations seem measured so it seems unlikely there will be a significant backlash.”

Rosenberg meanwhile said tech companies were supportive of the proposed regulations, “Largely because they realise the public is concerned and if the public fears their products it will hinder success.”

“On the other hand, smaller companies are worried that the requirements imposed by the Executive Order will favor large and/or heavily funded companies that can afford red teams and extensive testing. There needs to be provisions to ensure smaller companies are not put at a disadvantage.”