Should AI development be paused or restricted?

Should we hit pause on AI? I'll map the strongest arguments on both sides of the restriction debate.

There is no consensus on whether AI development should be paused or restricted, but the debate intensified in 2023-2024. The strongest call for a pause came from the Future of Life Institute.

The Case For a Pause or Restrictions

In March 2023 the nonprofit Future of Life Institute issued an open letter calling on all AI labs to "immediately pause for at least six months the training of AI systems more powerful than GPT-4". The letter stated the pause should be "public and verifiable, and include all key actors" and that "If such a pause cannot be enacted quickly, governments should step in and institute a moratorium".

The letter was initially signed by over a thousand people, including many notable technology leaders, ultimately growing to more than 1,300 people, including several notable technology experts and later more than 31,000 people. Notable signatories include Tesla and Twitter owner Elon Musk, Apple co-founder Steve Wozniak, Skype co-founder Jaan Tallinn, and DeepMind researchers Yoshua Bengio and Stuart Russell.

The core argument was that AI labs are currently locked in an "out-of-control race" to develop and deploy machine-learning systems "that no one – not even their creators – can understand, predict, or reliably control".

Proponents argued:

1. Safety and Alignment: "Powerful AI systems should be developed only once we are confident that their effects will be positive and their risks will be manageable". This confidence "must be well justified and increase with the magnitude of a system's potential effects".

Supporters wanted time to make systems "more accurate, safe, interpretable, transparent, robust, aligned, trustworthy, and loyal".

2. Need for Governance to Catch Up: The pause was intended to allow AI research and development to refocus while governance catches up. During the pause, the letter called for AI developers to work with policymakers to "dramatically accelerate" the development of robust AI governance systems, including creating new regulatory authorities, tracking and watermarking highly-capable AI systems and establishing liability for AI-caused harm.

Advocates argue that, assuming the pause was achieved, it "would create a breathing space where the policy and regulation required to safeguard AI development can be developed". This view notes that technology has always run ahead of regulations and that without checks, consequences could have profound adverse effects on society, putting safety and privacy at risk.

Other cited risks include privacy violations, human rights and safety concerns from chatbots like ChatGPT, deepfakes and copyright infringement from training data, generating "confidently incorrect or biased responses", and potential misuse by cyber criminals or authoritarian states. More existential concerns raised by signatories include that contemporary AI systems are now becoming human-competitive at general tasks and thereby "can pose profound risks to society and humanity".

The Case Against a Blanket Pause

Critics largely agree regulation is needed, but argue a 6-month moratorium is the wrong tool.

1. Delaying Benefits: "It’s already clear that AI will bring benefits to drug development, medical diagnosis, climate modeling and weather forecasting, education, and many other areas". A nationwide cessation "would have the inevitable result of delaying access to the technology’s benefits".

2. Legally Dubious and Difficult to Enforce: Analyses note there is no U.S. federal or state government entity that has clear legal authority to issue a moratorium on the training of large AI systems. The Federal Trade Commission's mission is protecting the public from deceptive practices, and an FTC moratorium would impede companies from competing.

Enforcement would also be difficult because "The U.S. government is clearly not going to start engaging in prohibition-era-style raids on companies suspected of performing unauthorized AI training" and instead would likely rely on self-reporting with no easy way to generate the list of entities subject to certification.

Unlike nuclear weapons where materials like uranium and centrifuges are hard to come by, "With AI systems the key ingredients are data and computing power, both of which are readily accessible and have an essentially limitless list of non-moratorium-violating uses".

3. Line-Drawing and "What Next?" Problems: Critics ask what would be the size threshold for AI systems subject to the moratorium, what metric would be deemed sufficient, and who would do the measuring. A six-month moratorium would also quickly lead to lack of consensus on what to do next as expiration nears.

4. Geopolitical Implications and Ineffectiveness: A moratorium in the U.S. "would have no force internationally. Governments and companies in other countries would continue to invest in building large AI systems", putting the U.S. at a disadvantage.

Many experts argue "A pause is virtually impossible at this point... The technology is moving too rapidly and becoming increasingly accessible so individuals can build and run their own LLMs. So how would such a pause be enforced?".

Some argue that six months is "equivalent to throwing a brick into the Grand Canyon. Our only option is to build safety features into the rocket after its launch".

Restriction vs. Regulation

Most experts on both sides agree that some form of restriction or regulation is needed, differing only on form:

The Brookings Institution summarizes the latter view: "It is far better for the U.S. to remain at the forefront of AI—advancing the state of the art, and using that knowledge to better identify and mitigate risks—than for the U.S. government to attempt to impose a legally dubious, unenforceable, and easily circumvented nationwide halt".

In short, the question is less whether AI should be governed and more how: whether through a temporary, verifiable moratorium on frontier model training to allow governance to catch up, or through ongoing, enforceable regulation that manages risks without stopping development entirely.