In this groundbreaking discussion paper, authors Connor Axiotes and Eddie Bolland, make the case for thinking ahead when it comes to artificial superintelligence, recommending we spur innovation, research, and investment into AI companies, whilst also proactively adapting policies to the effects of such a monumental shift in technology.
The paper recommends:
1. Invest in British Computing Resources
a. Allow our existing British public computing power and new exascale capacity to be used by our world leading universities for AI safety work - because universities are being crowded out by private labs with much more access to ‘cloud compute.’
b. The introduction of a new ‘British Compute Reserve.’
2. Setting the UK up for Success
a. Planning reform - build on the green belt, and implement street votes to make the UK a country high skilled AI safety researchers want to live in.
b. Lower corporation tax to an internationally competitive level so that more AI companies want to set up here in the UK.
3. Create a Public Comprehensive AI Monitoring System
a. Begin to monitor the largest AI models. A multilateral and unobtrusive monitoring of lab training runs would systematically track their capabilities and the extent of their alignment, to make sure innovative AI systems are safe and have few harmful emergent properties.
b. Making third-party external audits mandatory for largest/riskiest lab training runs.
4. UK to Lead the World in International Agreements on the Safe Deployment of Advanced AI Systems
a. The UK should take the lead on creating an International Agency for AI (IAAI).
b. A P5 statement on air-gapping nuclear weapons facilities from AI to reduce the chance of accidental nuclear strikes.
c. Lay out the structure and objectives of Bletchley Park’s 2023 AI Safety Summit.
5. Expand Educational Grants and High-Skilled Visa Scheme
a. Increase youth engagement in STEM through tax-credits to private companies to address long term skills shortages.
b. Eliminate obstacles to obtaining the High Potential Individual visa.
c. Align High-skilled Visa schemes with the priorities of prospective applicants to maintain the UK’s position as a global leader in attracting AI talent.
d. Expand university courses alongside changing patterns of demand for priority areas to prevent future skills shortages.
e. Integrate the Adam Smith Institute’s model for visa auction markets.
6. Regulatory Markets for AI
a. The UK should utilise ‘Regulatory Markets’ - private regulatory experts to bring their experience in helping with safety-based, innovation-inducing AI legislation. This would help to solve the knowledge gap between the government and the relevant regulatory body.
7. Government Investment in AI Safety
a. The Great British AI Prizes: cash prizes for open research questions in AI safety, such as ‘how do we stop larger models from hallucinating?’
b. If sovereign capabilities such as a public LLM are sought after, then AI alignment researchers and academics should be able to access them for safety work.
8. Facilitate the Safe Use of APIs for Innovative SMEs and Researchers
a. Enable SMEs and researchers to develop products and carry out safe research through APIs accessed on the research resource.
b. Implement risk based requirements for API access to reduce the risk of misuse and encourage private participation.
9. Effective Procurement to Increase Efficiency and Innovation
a. Introduce Challenge Based Procurement to improve the efficiency and reduce the barriers for smaller 5 firms.
b. The Office for AI should identify opportunities for procurement to support proof of concept work too risky for nationwide deployment.
c. Procurement for AI assurance within the public sector to support private sector firms and ensure safe deployment.
10. Saving Lives with AI-Powered Medicine while Reducing Engineered Pandemic Risk
a. The NHS should invest in Generalist Medical AI capabilities through the NHS AI Lab.
b. Introduction of Three Lines of Defence Structure to ensure the UK is proactively prepared for biosecurity risks.
c. Invest in pathogen monitoring systems and introduction of bio-engineering licences.
11. Implement a Review on the Possible Labour Effects of Future AGI
a. Commission a White Paper on what the introduction of a universal basic income (UBI) or a negative income tax (NIT) would look like in a worst-case scenario;
b. Introduce NIT and UBI trials to prepare for the possibility of AI caused unemployment.