Should AI development be paused or restricted?

Should AI development be paused? I'll weigh the arguments for and against restrictions.

Whether AI development should be paused or restricted is an active, unresolved debate with no consensus. The major divide is between a temporary, verifiable pause on the most powerful training runs versus targeted restriction and "governed acceleration."

Arguments For a Pause or Restrictions

The most prominent call for a pause came in March 2023, when the nonprofit Future of Life Institute published "Pause Giant AI Experiments: An Open Letter". It called on "all AI labs to immediately pause for at least 6 months the training of AI systems more powerful than GPT-4", citing risks such as AI-generated propaganda, extreme automation of jobs, human obsolescence, and a society-wide loss of control.

The letter said: "recent months have seen AI labs locked in an out-of-control race to develop and deploy ever more powerful digital minds that no-one - not even their creators - can understand, predict, or reliably control". It received more than 30,000 signatures, including Yoshua Bengio, Stuart Russell, Elon Musk, Steve Wozniak and Yuval Noah Harari.

The specific asks were:

More recently, Anthropic CEO Dario Amodei penned a nearly 4,000-word essay laying out the case for why AI development should be decelerated so that adequate guardrails can be deployed. Similarly, proponents argue approaches like establishing an international body akin to the IAEA for AI, overseeing compute resources and enforcing moratoriums on artificial superintelligence training until alignment is proven, and adopting frameworks like the EU AI Act for risk classification, documentation, and certification.

General concerns cited for restriction include:

Supporting stance is often framed as: “Pausing AI development ensures ethical safeguards are embedded before irreversible impacts occur.”

Arguments Against a Blanket Pause

Opposing stance is often framed as: “A pause on AI advancement risks stifling innovation and losing competitive advantage.”Specific arguments include:

1. Unworkable and unenforceable: Critics describe it as a prisoner's dilemma: "There's no way to collectively get all of the different entities that are working on these language models to collectively pause." Not only is it hard to get all labs to pause, but most labs would probably continue to train once they realize others were not stopping. It "looks good on paper, but it's completely unrealistic,".

Humanity is insufficiently consensus-driven, and stakeholders have too many easy ways to bypass bans.

2. Geopolitical competition: Even if one views a ban, pause, or slowdown as desirable, "ask yourself what China will do? Beijing called Anthropic CEO Dario Amodei’s 'We must pace the frontier' warning 'fearmongering,' suggesting that the alarm about AI was a US plot to hinder China’s own innovation." If China isn’t on board, capable AI models will be available regardless of what the US does. As one commentator put it: "If you pause AI in the US for 6 months, China will own everything." Pausing AI development is not realistic because even if the United States could pause domestically, it could not stop development in China or the spread of open-source models.

3. It could reduce safety, not increase it: Most importantly, pausing development would also pause some of the work needed to make AI safer. Governments also need advanced AI expertise to evaluate systems, detect misuse, and defend critical infrastructure. Slowing AI research could slow the safety research needed to make AI systems more secure.

A related argument is asymmetry between offense and defense: "Pausing AI development risks accentuating asymmetry between able defenders and rogue actors, whether they be nation states, criminals, terrorists, or simply misaligned models. AI defense can reduce or offset the risk of AI offense. If we delay access to the most capable models, defenders’ hands would be tied."

Authors argue: "We should not ban 'superintelligence' by pausing or slowing AI development, which could put defenders at a disadvantage and hold back AI’s potential to reduce risk." AI may also help reduce risks from large-scale threats from asteroids, volcanos, solar storms, floods, and wildfires by improving prediction and detection, and the economic growth dividend could help society adapt.

Bill Gates is against pausing AI development, believing that it would disadvantage ethical developers and not effectively stop misuse by malicious actors.

4. Costs to beneficial uses: A pause might slow down the development of AI applications in crucial areas such as education, healthcare, and food. AI has potential to increase productivity, reduce costs, accelerate medical research, improve education, and help solve problems humans struggle to tackle alone, and slowing development could delay life-saving discoveries.

The Middle Path: Restricted but Not Paused

Rejecting a broad pause does not mean accepting uncontrolled development. The better approach is described as governed acceleration: continue advancing AI while imposing stronger controls as systems acquire greater capability and authority.

Proposed alternatives to a blanket pause include:

Proponents of this view argue the debate should move away from choosing between stopping AI development and allowing unrestricted deployment, and instead continue development but exercise great care in deployment, requiring more safety measures and more human supervision as systems become more capable and autonomous.

In short, while few support completely unrestricted development, critics of a pause argue that a blunt, global pause is both infeasible and potentially counterproductive, and that restricted, risk-based governance is more effective than a moratorium.