80,000 Hours Podcast
#166 – Tantum Collins on what he’s learned as an AI policy insider at the White House, DeepMind and elsewhere
Episode guests
Podcast summary created with Snipd AI
Quick takeaways
- The evolution of collective decision-making and government should be approached deliberately to avoid inadvertently strengthening autocracies on the global stage.
- China's scientific innovation and research output in AI challenge the misconception that it can only copy and not innovate.
- Collaboration between US and Chinese labs on AI research is significant, challenging the perception of a traditional arms race.
- Allocating increased public funding for AI research and development can address multiple concerns and garner support from different interest groups.
- Improving interpretability of AI models and enhancing cybersecurity measures are key for protecting against risks and ensuring trust and safety.
- Immigration reform that facilitates collaboration among international AI experts can benefit various sectors and foster innovation and collaboration.
Deep dives
China's Scientific Innovation and AI Research
China has a long history of scientific innovation and has been responsible for a significant amount of scientific research, including in the field of AI. The number of AI research papers coming out of China has increased significantly, surpassing other countries. This challenges the misconception that China can only copy and not innovate.
US-China AI Collaboration and Arms Race
While there is increased competition and attention given to AI between the US and China, it does not resemble a traditional arms race. Collaboration between US and Chinese labs on AI research is significant and has even increased in recent years. While there are concerns about the US losing its lead, factors such as access to cutting-edge compute and human capital give the US an advantage. The fear of China surpassing the US remains uncertain.
Chinese Regulations and AI Diffusion
China has implemented regulations regarding AI, but it is too early to determine their effectiveness. Some argue that these measures may hinder China's ability to catch up to the US in AI capabilities. However, the innovation and diffusion of AI vary in China, with a focus on innovation and challenges in effective diffusion. This challenges the narrative that China can only copy and diffuse AI technologies without innovation.
Need for Increased Public Funding for AI
One proposal that could address multiple concerns is the allocation of increased public funding for AI research and development. This could include funding for alignment issues, ethics research, improved benchmarking and standards, and the development of privacy-preserving machine learning techniques. Expanding the funding would allow for addressing various risks and challenges associated with AI, from consumer protection to cybersecurity, and could potentially garner support from different interest groups by creating a larger pie of resources to allocate.
Improving Interpretability and Security
Another idea that could gather broad support is the improvement of interpretability of AI models and enhancing cybersecurity measures. Both initiatives are essential for protecting against risks and ensuring consumer trust and safety. The application of interpretability can address concerns related to bias, prejudice, and ethical issues in AI systems. Meanwhile, enhancing cybersecurity can contribute to preventing misuse of AI and protecting sensitive data from breaches. These proposals align with the interests of groups focused on extinction risks, AI ethics, national security, and privacy concerns.
Immigration Reform and Collaboration
Immigration reform that facilitates collaboration and knowledge exchange among international AI experts is an area of shared interest. Ensuring the availability of visas and reducing restrictions for qualified individuals to join AI research efforts can benefit various sectors, including health, cybersecurity, ethics, and governance. Promoting diversity and global cooperation within the AI community can foster innovation and collaboration on complex and emerging challenges. Immigration reform can be an inclusive policy that addresses societal needs, facilitates collaboration, and contributes to AI development.
Improving task allocation and collaboration within organizations
One area of interest in improving organizational dynamics is using machine learning to enhance task allocation and collaboration. By leveraging tools like recommendation engines, organizations can better match individuals with projects, improving efficiency and productivity. For example, tools that suggest papers or conferences based on individual interests and expertise can enhance research productivity. Additionally, machine learning systems can aid in code review, debugging, and project matching, enhancing collaboration and information sharing. While there are challenges to address, such as ensuring precision and avoiding information bubbles, leveraging AI in task allocation can lead to more effective and productive organizations.
Using AI for decision-making support
AI systems can also be used to provide decision-making support for managerial and strategic purposes. Tools like the paper recommender and conference recommender can help individuals stay informed and up-to-date with relevant research. Internal systems can recommend lunch pairings, facilitating networking and knowledge sharing. Additionally, machine learning can assist in organizational planning, identifying rising and declining research domains, and highlighting potential talent pools based on bibliometric data and metadata. While opportunities for AI-driven decision-making support exist, considerations should be given to interpretability, biases, and the balance between performance and transparency.
Challenges and considerations in applying AI to organizational dynamics
While applying AI to improve organizational dynamics presents numerous opportunities, it also requires careful consideration. Pitfalls include the risk of dehumanization and loss of human interactions, the challenge of balancing precision and recall in matching individuals with tasks, and the potential for bias or opaque decision-making. Additionally, aligning incentives within organizations is crucial for the successful implementation of AI systems. Ethical considerations around trust, fairness, and the impact of AI on employee experiences must also be addressed. By balancing technical capabilities, ethical considerations, and stakeholder interests, organizations can harness AI to enhance productivity, collaboration, and decision-making processes.
Panpsychism and the Spectrum of Consciousness
The speaker discusses panpsychism, which is the idea that consciousness exists along a spectrum, with varying degrees of consciousness in all things. They argue that while humans may have a higher level of consciousness compared to other beings, consciousness is continuous and can exist in a rock or an ant colony. The speaker is swayed by the notion that consciousness is distributed across a spectrum, leading them to consider the possibility of consciousness in non-carbon-based systems like AI. They mention that the concept of panpsychism has been debated, but they haven't found a better theory to explain consciousness.
Transitioning Between Tech and Policy
The speaker advises individuals interested in transitioning between technical roles in the tech industry and policy roles in government. They suggest connecting with people who have similar backgrounds and are already working in policy. They also emphasize the importance of understanding the different working styles and cultures between industry and government before making the transition. Additionally, they mention the value of having technical expertise in AI coupled with policy knowledge, as there is a shortage of individuals with both backgrounds in the policy space.
"If you and I and 100 other people were on the first ship that was going to go settle Mars, and were going to build a human civilisation, and we have to decide what that government looks like, and we have all of the technology available today, how do we think about choosing a subset of that design space?
That space is huge and it includes absolutely awful things, and mixed-bag things, and maybe some things that almost everyone would agree are really wonderful, or at least an improvement on the way that things work today. But that raises all kinds of tricky questions.
My concern is that if we don't approach the evolution of collective decision making and government in a deliberate way, we may end up inadvertently backing ourselves into a corner, where we have ended up on some slippery slope -- and all of a sudden we have, let's say, autocracies on the global stage are strengthened relative to democracies." — Tantum Collins
In today’s episode, host Rob Wiblin gets the rare chance to interview someone with insider AI policy experience at the White House and DeepMind who’s willing to speak openly — Tantum Collins.
Links to learn more, highlights, and full transcript.
They cover:
- How AI could strengthen government capacity, and how that's a double-edged sword
- How new technologies force us to confront tradeoffs in political philosophy that we were previously able to pretend weren't there
- To what extent policymakers take different threats from AI seriously
- Whether the US and China are in an AI arms race or not
- Whether it's OK to transform the world without much of the world agreeing to it
- The tyranny of small differences in AI policy
- Disagreements between different schools of thought in AI policy, and proposals that could unite them
- How the US AI Bill of Rights could be improved
- Whether AI will transform the labour market, and whether it will become a partisan political issue
- The tensions between the cultures of San Francisco and DC, and how to bridge the divide between them
- What listeners might be able to do to help with this whole mess
- Panpsychism
- Plenty more
Chapters:
- Cold open (00:00:00)
- Rob's intro (00:01:00)
- The interview begins (00:04:01)
- The risk of autocratic lock-in due to AI (00:10:02)
- The state of play in AI policymaking (00:13:40)
- China and AI (00:32:12)
- The most promising regulatory approaches (00:57:51)
- Transforming the world without the world agreeing (01:04:44)
- AI Bill of Rights (01:17:32)
- Who’s ultimately responsible for the consequences of AI? (01:20:39)
- Policy ideas that could appeal to many different groups (01:29:08)
- Tension between those focused on x-risk and those focused on AI ethics (01:38:56)
- Communicating with policymakers (01:54:22)
- Is AI going to transform the labour market in the next few years? (01:58:51)
- Is AI policy going to become a partisan political issue? (02:08:10)
- The value of political philosophy (02:10:53)
- Tantum’s work at DeepMind (02:21:20)
- CSET (02:32:48)
- Career advice (02:35:21)
- Panpsychism (02:55:24)
Producer and editor: Keiran Harris
Audio Engineering Lead: Ben Cordell
Technical editing: Simon Monsour and Milo McGuire
Transcriptions: Katy Moore
Remember Everything You Learn from Podcasts
